text
string | predicted_class
string | confidence
float16 |
|---|---|---|
Participants called for a scholarly approach to ISTCs. Acknowledging the academic relevance of global surgery can lead to more significant improvements through increased participation in education, research, and service related to global surgery. Academics specifically called for interdisciplinary collaboration in global surgery efforts, to include fields such as health education, sociology, economics, and anthropology in order to make sure ISTCs are yielding the best results via the best approaches . | other | 99.9 |
Currently, there is insufficient research in the field of global surgery, especially regarding the effectiveness of capacity-building partnerships . Needs assessments should be developed by those in LMICs, so that ISTCs can be directed to address LMICs’ specific needs. Attention should be given to cultural humility of global surgery interventions, which is important in ensuring that the interventions align with the goals of LMICs. Assessments of education methodology and technology are important in addressing how best to teach surgery in resource-limited settings. Equally relevant is the need for longitudinal outcomes research that assesses whether ISTCs are producing successful outcomes, including increased number of surgical healthcare providers in LMICs, greater number of surgeries, and improved patient outcomes. There is also a need for a standardized system of metrics to evaluate these outcomes in order to inform policy makers . Increased research output in regards to global surgery is a feasible step towards improving efforts such as ISTCs and can simultaneously serve to increase the presence of global surgery in the academic arena. | other | 60.16 |
Program evaluation should be a systematic part of ISTCs, and recommendations from both local and visiting teams should be taken into account during this evaluation. Reports should outline what the needs and goals were, whether those needs were addressed, how the goals were met, what was successful, what was unsuccessful, and what the follow-up plans are. To facilitate collaboration, these reports should be publicly shared so that others can learn from the findings. These types of evaluations can help prevent duplication of unsuccessful models and can instead promote improvement and effectiveness. Additionally, establishing an equal and bidirectional partnership is important to achieving success, which is why evaluation from both sides of the partnership is essential . Program evaluation in a bidirectional manner is already being done by some who are involved in ISTCs, but it should continue and become a more regular part of global surgery work such that these efforts can further improve. See Fig. 3. | other | 99.9 |
Global communication technology can be integrated into ISTCs in order to provide year-round education over long distances. Technology allows for a continued presence of surgical healthcare providers from HICs in LMICs, which encourages collaboration and formalizes partnerships, while accelerating and continuing learning [34, 35]. | other | 99.9 |
Skype can be used for telementoring and patient consults, so that surgeons in LMICs can work with those in developed countries to discuss cases, share scans, and receive clinical advice. Google Glass, Wi-Fi enabled video cameras, and YouTube can be, and in many cases have already been, used to broadcast patient rounds and surgeries to students both in HICs and LMICs. Showing operations can help with learning new procedures, solidifying knowledge of previously learned procedures, and gaining exposure to more cases. Some surgeons are also using telesimulation to teach surgical skills to those in LMICs [36, 37]. Participants pointed out that technology is being implemented by many who are engaged in global surgery work and has been a feasible addition to ISTCs. Although patient privacy is a concern, measures can be taken to ensure that privacy laws are followed and patient consent is obtained. | other | 99.8 |
The Internet can be used to create online teaching modules in surgical specialty education to supplement the hands-on education of surgical healthcare providers in LMICs . It can also be used to disseminate knowledge via shared textbooks, journal articles, and other less accessible materials . Online curriculum modules and educational materials have been successful in supplementing and standardizing thoracic surgery in the U.S., and this model could potentially supplement education in LMICs . This increased access to knowledge is invaluable in improving surigcal care and facilitating the education of more surgical healthcare providers worldwide. See Fig. 4. | other | 99.9 |
Technology, however, should be used with caution, as mentioned by the study participants. It should not be used as a substitute for personal interaction and on-the-ground work in LMICs, but rather as a supplement to maintain a partnership and reinforce learning . Face-to-face interaction during ISTCs allows for surgical procedural education that is adapted to local realities, as well as for the development of close friendships and stronger partnerships. | other | 99.9 |
ISTCs often begin with individual efforts, and as mentioned by participants, this can be discouraging since there is no systematic method for involvement and implementation. Individual ISTCs are often done in many different ways, but a more structured approach is needed to globally coordinate efforts [18–21]. This coordination could happen at the national level, through discussions with the Ministry of Health, Ministry of Finance, and healthcare representatives from different regions in the country. This could also involve professional bodies who are monitoring surgical activities in the country or region and can convey information to national governmental bodies. A brief overview of a proposed systematic process is outlined below and in Fig. 2: expansion and consolidation phases. | other | 99.8 |
In order for ISTCs to have a greater impact globally, more ISTCs need to be formed, which is the goal of expansion phase. This includes the process of reaching out to new potential sites and initiating contact. The consolidation phase, which supports sustainability, includes maintaining partnerships through a continued presence. This is the phase most ISTCs are currently in. Sites should be revisited, so that there is assurance that new techniques, surgical education programs, and other interventions have been sustained, and so that relationships are strengthened. Vital to consolidation phase and the sustainability of ISTCs is the process of training the trainers – involving local stakeholders to create future plans for expansion of ISTCs to other institutions in the country as well as to more rural regions. This increases capacity for “south-south” partnerships between LMICs as well. See Fig. 2. | other | 99.9 |
Once an ISTC has met these goals for a particular site, expansion phase can be repeated. However, repeating expansion phase does not imply termination of relationships that were maintained in consolidation phase, but instead focuses on a comprehensive picture of how more relationships can be formed. These cycles of expansion and consolidation could allow ISTCs to both proliferate in number and be sustainable with long-term successful outcomes. This is essential so that efforts are not only reaching tertiary hospitals, but are also building surgical capacity in rural hospitals and clinics . | other | 99.9 |
As mentioned, there are significant advantages to maintaining partnerships long-term, especially given their bidirectional nature. While helping build surgical capacity in LMICs, surgeons from HICs learn how to navigate complex surgical cases in low-resource settings and how to effectively develop training programs. Other benefits from involvement in ISTCs for surgical healthcare providers from HICs include becoming well-rounded and more adaptable in surgery, improving educational skills, and developing teamwork and leadership skills that are beneficial in surgical programs at their home institutions. Overall, continuing a bidirectional exchange of knowledge in ISTCs is important to increasing collaboration and unity in such efforts by equally engaging both sides of the partnership and by maintaining relationships long-term. | other | 99.9 |
This pattern of expansion and consolidation may be more difficult to implement, as it first requires a stronger foundation and systematic process for ISTCs, which may take several years. As more is learned from the implementation and continuation of current ISTCs, this process can become more streamlined. | other | 99.94 |
Participants mentioned concern that organizing ISTCs could lead to bureaucracy– something that surgical healthcare providers often encounter at their own institutions. However, an administrative body to keep track of current efforts, to receive requests/needs, to receive offers to help, to track funding, to collect data/research, and to disseminate results could better unify, systematize, and improve the operation of ISTCs. | other | 99.94 |
A possible model for organization is a three-tiered system: local, national, and international. At each level, there could be a “Board of Global Surgery,” with the local boards feeding information to the national and international boards. This bottom-up approach could help with the dissemination of information regarding ISTCs globally, without restricting individual initiatives. These boards could include groups of individuals that could function to provide organizational support to ISTCs worldwide. On a national/regional scale, there are professional bodies in place that direct, monitor, and evaluate surgical services in many countries around the world, including LMICs. Professional bodies could potentially play a significant role in organizing and unifying global surgery efforts on a national level, and can then feed information to a larger international body that coordinates ISTCs worldwide. On an international scale, a board could monitor and organize these national efforts and serve as an international data collection/research hub for global surgery work . See Fig. 5. | other | 99.9 |
There are, however, some challenges to creating this level of organization. As participants mentioned, some are resistant to centralized organization due to concerns over increased bureaucracy and loss of autonomy in global surgery work. Additionally, it is difficult to determine whether new organizational bodies are needed or whether organization should occur through already existing groups. Although developing this type of organizational structure might be difficult, it is necessary to begin considering fluid models that could ease the process of involvement in ISTCs. Those who are involved in global surgery efforts should work together and expand networks in a unified way . | other | 99.94 |
ISTCs are a long-term sustainable intervention involving partnerships that are geared towards building surgical capacity in LMICs. However, currently, the approach of HICs to these collaborations is fragmented. This study suggests that in order to organize ISTCs, those involved should consolidate efforts, collaborate between groups involved in ISTCs, establish a system of support, take on a scholarly approach, and integrate technology into ISTCs. This study also introduces the possibility of systematizing ISTCs through action and reflection, expansion and consolidation, and a single multi-level administrative organization. | other | 99.9 |
The overarching goal of ISTCs is to improve access to surgical care and to reduce the morbidity and mortality from surgical conditions in LMICs. Better coordination of efforts in specific ways can improve the number, scope, and sustainability of ISTCs in an attempt to achieve that goal. Future studies could analyze specific collaborations and assess their progress in regards to the implementation of the above recommendations and framework. They could also address the practicality of these interventions and the implications of their implementation. A similar study should be done to address the LMIC perspective of ISTCs and other sustainable global surgery efforts such that the results from both studies can lead to overall improvement and better coordination of ISTCs worldwide. | other | 99.6 |
Insects are the largest group of species on the planet1. Their success is critically dependent on their mobility. Although insects are relatively small in terms of both body size and neural system, their capability of handling complex environment outperforms most artificial robots. Simulating the locomotion of virtual insects are of great interest to researchers from various fields including experimental biology, computer animation and robotics. | other | 99.94 |
Experimental biologists used the simulation model to repeat experiments which otherwise would be impossible to conduct on the vivo subjects, in order to illuminate the underlying principles of their locomotory mechanisms. The early attempt is Walknet2, which simulated the movement of walking stick insects. The follow-up works extended the standard Walknet, which only simulated the gait in a procedurally-kinematic fashion, to advanced components including searching for foot placements3 and dynamics simulation4. Other studies simulated the gait pattern of cockroaches by incorporating rudimentary motor-neuron activation and agonist/antagonist Hill-type muscle pairs5–7 to analyze the gait stability of cockroach in standard and perturbed events. However, in this model the gait pattern is limited to the horizontal plane and excludes the effect of gravity. A neuron-mechanical model of a cockroach is further proposed to simulate the individual and population behaviours of real neurons8. However the parameters in this complex model is hand-tuned and do not allow fast-extending the skill repertoire of virtual insects. Our work uses the evolutionary algorithm to automatically set the model parameters and avoids the process of manual efforts. | study | 100.0 |
Researchers in the field of computer animation target at the problem of reproducing the mobile capability of real insects on their virtual representation. Virtual insects are widely used in graphical applications, including video games, movies and virtual/augmented reality etc. In contrast to the large collection of existing works in biped simulation, the field of insect simulation is less explored. Previously researchers set up a specialized system of synchronized cameras to capture and synthesis insect motion9. However this method requires manual post-processing and specialized hardware. Researchers also tackled this problem by procedurally but kinematically modelling the gait of multi-legged creatures10. By carefully selecting the foot placement, this method is able to adapt to different terrains and recover from perturbations. Due to the lack of a fully actuated model, the motion synthesized in this way is not physically-plausible and cannot take full advantage of the information from the interaction with the environment. Previous methods11,12 modelled the controller of virtual insects as a network of nonlinear oscillators and tuned the parameters of the network via the optimization strategy of Covariance Matrix Adaptation (CMA). The network was designed to mimic the Central Pattern Generator, which was found in real insects and in charge of their gait pattern. Researchers also modelled and animated virtual myriapoda by developing a hybrid animation system that combines kinematic and dynamic simulation13. While the body was driven by both rigid body simulation (for hard skeleton) and finite element method (for soft interlinks), the leg movements were synthesized in a kinematic fashion. | study | 88.3 |
Robotics researchers used the simulation framework to speed up the process of robot design and control. A representative hexapod robot, RHex14, was designed to perform the double-tripod gait similar to the gait pattern observed in hexapod insects. This gait pattern allowed inherent locomotion stability. The open-loop controller was further improved by the introduction of Central Pattern Generator (CPG). The biological evidences show that the CPG is responsible for producing the basic gait pattern of real animals15. Bio-inspired robots, such as walking salamander16, swimming fish17 and flying bird18, are controlled by mimicking the machinery of CPG. More details can be found in the review paper on applications of CPG for locomotion control in animals and robots19. However, existing models of CPG are generally constructed as a network of nonlinear oscillators, rather than spiking neurons. The advantage of using nonlinear oscillator is the inherent convergence to limit cycle or point attraction for most oscillator models, thus capable of endogenously generating the gait pattern without the high-level commands. However, the challenge from oscillator-based networks is how to connect the sensory information/high-level commands with the low-level control strategy. Previous method11 used a precomputed look-up table. This inevitably creates a large discretized look-up table. In comparison, real organisms, including insects, use spiking neurons for the general purposes of environmental sensory, information transmission, decision-making and muscle actuation. Such observation inspires this work to explore the solution of using spiking neuron, as the alternative building block to nonlinear oscillator, to construct the neural system. | review | 99.9 |
Building a simulation model to match its real equivalence is a non-trivial task. On one hand, a complex model may introduce too many parameters, which could be impossible to tune by hand. On the other hand, a simplified model may not capture the important details of the model and thus reduce the credibility and capability of the virtual model. In this work, we propose a neuro-musculo-skeletal model for virtual insects (using specifically the ant as an example) and an automatic solution to the problem of parameter optimization. More specifically:We propose a uniform encoding of sensory information, high-level commands and control strategy, with the spiking patterns of the neural system. Such an encoding paradigm provides the equivalent representation as real organism and shows the potential of using spiking neurons as the building block for accomplishing general tasks. The network also provides a continuous mapping from the sensory information/high-level commands to low-level control, in contrast to the discretized mapping in previous work11.We use the evolutionary algorithm to automatically find the optimal values for both the controller and muscle parameters. Given the high dimensions of the model state spaces, we resolve the challenge by dividing the optimization into a two-stage procedure. The first stage optimizes the parameters of controller and muscles for the standard gait of walking straight forward, while the second stage progressively extends the skill repertoire by considering curve trajectories and uneven terrain.We integrate the prior information (foot trajectories) of real ants during the optimization process via the evolution algorithm. The introduction of ground-truth data from real ants allows the quantitative comparison between the proposed model in this work and existing models11. The result shows that the trajectories generated from this model presents smaller deviations from the ground-truth data. | study | 100.0 |
We propose a uniform encoding of sensory information, high-level commands and control strategy, with the spiking patterns of the neural system. Such an encoding paradigm provides the equivalent representation as real organism and shows the potential of using spiking neurons as the building block for accomplishing general tasks. The network also provides a continuous mapping from the sensory information/high-level commands to low-level control, in contrast to the discretized mapping in previous work11. | other | 99.7 |
We use the evolutionary algorithm to automatically find the optimal values for both the controller and muscle parameters. Given the high dimensions of the model state spaces, we resolve the challenge by dividing the optimization into a two-stage procedure. The first stage optimizes the parameters of controller and muscles for the standard gait of walking straight forward, while the second stage progressively extends the skill repertoire by considering curve trajectories and uneven terrain. | other | 97.6 |
We integrate the prior information (foot trajectories) of real ants during the optimization process via the evolution algorithm. The introduction of ground-truth data from real ants allows the quantitative comparison between the proposed model in this work and existing models11. The result shows that the trajectories generated from this model presents smaller deviations from the ground-truth data. | study | 99.9 |
The simulation framework is composed of three parts: the spiking-neuron controller, the muscle actuation and the skeleton. The spiking-neuron controller receives the external inputs (sensory information and high-level commands) and sends out a series of spiking trains during each motion cycle and activates the muscle contraction. Each skeleton joint is connected with a pair of antagonistic muscles, generating the appropriate torques and resulting in the forward movements of the virtual insect. The parameters of the neuron controller and muscle actuation are updated with the technique of evolutionary algorithm, with the captured data of real insects. The overall algorithm of our method is presented in Fig. 1(a).Figure 1(a) Algorithm flowchart. (b) Block diagram of neural system. | study | 99.94 |
In real insects, spiking pattern of the neural system are used for the general purposes of information transmission and processing. The controller in the proposed model is modelled as a network of spiking neurons. The unit model is based on the model proposed in the work by Izhikevich20. The model is endowed with both biologically plausibility of Hodgkin–Huxley-type dynamics and computational efficiency of integrate-and-fire neurons. | other | 99.6 |
The equation for the spiking neuron model is:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v^{\prime} =0.04{v}^{2}+5v+140-u+I$$\end{document}v′=0.04v2+5v+140−u+I2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$u^{\prime} =a(bv-u)$$\end{document}u′=a(bv−u)with the auxiliary after-spike resetting:3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rm{if}}\,v\ge {\rm{30}}\,{\rm{mV}},then\{\begin{array}{ll}v\,\leftarrow & c\\ u\,\leftarrow & u+d\end{array}$$\end{document}ifv≥30mV,then{v←cu←u+dv represents the membrane potential of the neuron. u represents the activation of K+ ionic currents and inactivation of Na+ ionic currents, providing negative feedback to v. This model has four dimensionless parameters: a, b, c, d. The model is capable of generating multiple spiking patterns, including regular spiking, intrinsically spiking, chattering and so on. We here choose the most common type of spiking pattern (regular spiking) and its corresponding parameters are: a = 0.02, b = 0.2, c =−65 mV, d = 8. | other | 60.3 |
One novelty of this work is to provide a uniform encoding of both sensory information, high-level commands and control strategy. We choose specifically the slope angle as the example of the sensory information and the action of turning as the example of high-level commands. For the slope angle θs or the target orientation for next stride θt, the input currents I are defined as4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${I}_{sensor}=\frac{{\theta }_{s}}{{\theta }_{max}^{s}}{I}_{max}^{s}$$\end{document}Isensor=θsθmaxsImaxs5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${I}_{target}=\frac{{\theta }_{t}}{{\theta }_{max}^{t}}{I}_{max}^{t}$$\end{document}Itarget=θtθmaxtImaxt\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\theta }_{max}^{s},{\theta }_{max}^{t}$$\end{document}θmaxs,θmaxt are the maximum slope angle and target orientation. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${I}_{max}^{s},{I}_{max}^{t}$$\end{document}Imaxs,Imaxt are the corresponding maximum input currents. Here both variables are set to 20 mA. | study | 99.94 |
The input currents Isensor, Itarget are fed into an input layer of spiking neurons. The input layer is connected with a central layer of 16 spiking neurons, which are then fully-connected to the motor neuron. The output from the central layer is added with a bias component (denoted in blue in Fig. 1(b)), which generates the standard gait pattern of walking straight forward. This constructs a neural network with spiking neurons (Fig. 1(b)). The synaptic connection weights between the neuron layers determine how the external information adjusts the control signals from the bias neurons and eventually modifies the standard gait pattern. | study | 99.9 |
For the control strategy, each muscle is activated by a unit neuron model, which produces the overall spiking pattern of the neuron group activating the single muscle. Each joint is actuated by a pair of agonist/antagonist muscles, each of which is individually activated by one neuron controller. The separate control of individual muscles mimics the distributed hierarchy of the neural control in real insect. | other | 99.75 |
The activation process converts the spiking trains from the neural system into the activation signal on the muscle. This is modelled as a first-order differential equation:6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${a}_{t+1}=({v}_{t}-{a}_{t}){\rm{\Delta }}t+{a}_{t}$$\end{document}at+1=(vt−at)Δt+atwhere at is the activation signal at time t, vt is the excitation from the motor neuron (Equation 2). Δt is the stepsize. | study | 99.9 |
To simulate the contraction dynamics of a Hill-type muscle, we need to model two main features: the force—length and force—velocity relationships. The force—length (FL) part models the function of both active and passive forces in terms of the muscle length, and the force—velocity (FV) part describes how the contraction force varies with respect to the muscle’s contraction velocity.7\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${F}_{CE}={F}_{max}\,\ast \,{a}_{t}\,\ast \,{F}_{L}\,\ast \,{F}_{V}$$\end{document}FCE=Fmax⁎at⁎FL⁎FV8\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${F}_{L}=exp(-{|\frac{{L}_{ce}^{\beta }-1}{\omega }|}^{\rho })$$\end{document}FL=exp(−|Lceβ−1ω|ρ)9\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${F}_{V}=(\begin{array}{ll}({V}_{max}-{V}_{ce})/[{V}_{max}+({c}_{V0}+{c}_{V1}{L}_{ce}){V}_{ce}], & {\rm{if}}\,{V}_{ce}\le 0\\ \left[{b}_{V}-({a}_{V0}+{a}_{V1}{L}_{ce}+{a}_{V2}{L}_{ce}^{2}){V}_{ce}\right]/({b}_{V}+{V}_{ce}), & {\rm{if}}\,{V}_{ce}\, > \,0\end{array},$$\end{document}FV=((Vmax−Vce)/[Vmax+(cV0+cV1Lce)Vce],ifVce≤0[bV−(aV0+aV1Lce+aV2Lce2)Vce]/(bV+Vce),ifVce>0,where at is the activation signal from Equation 6. The variables Lce, Vce are the length and contraction velocity of the muscle. To simplify the model, we set Lce = 1 in the equation of Fv, which means that the change of muscle length is ignored when computing the force-velocity relationship. This assumption simplifies the model parameters as: cV = cV0 + cV1, aV = aV0 + aV1 + aV2. The parameters Ω = (Fmax, β, ρ, ω, Vmax, cV, bV, aV) are muscle-specific and determine the overall performance of a muscle. | study | 99.94 |
Existing works normally find the appropriate parameters using the collected data from vivo subjects23. However the experimental data are limited and may not be extended to different insects, or even different muscles on a single insect. Given the difference of the neural system between real and virtual insects, it is not of significance to follow the exact parameter values of real insects. Given the complicated model of nonlinear muscle, together with the parameters of neural system, we expect a high dimension of control space. It is challenging to reach an optimal solution for such optimization problem. In such circumstances, we propose the evolutionary algorithm to optimize the parameters of the controller and muscles simultaneously. | study | 99.94 |
During locomotion, an insect must solve essentially the same problem as a vertebrate although it has an external, rather than an internal skeleton. This is reflected in how the muscles are connected to the skeleton (Fig. 2). For the skeletal model, we follow the implementation from existing work11 and model the virtual insect as a hierarchy of connected rigid body.Figure 2Muscle attachment of the femur-tibia joint on a virtual ant. | study | 92.75 |
The main body is constructed as a set of three rigid bodies: head, thorax and abdomen, connected by two joints (head-thorax and thorax-abdomen). Each leg is actuated with three selected DOF: trunk-coxa, trochanter-femur and femur-tibia. Movement about the single joints and the resulting stepping patterns are generated by the activity of antagonistic muscle pairs. In the stick insect, the three major muscle pairs of a leg are the protractor and retractor coxae, the levator and depressor trochanteris, and the flexor and extensor tibiae. The protractor and retractor move the coxa, and thereby the leg, forward and backward. The levator and depressor move the femur up and down. The flexor flexes, and the extensor extends the tibia about the femur-tibia joint. | other | 95.25 |
Each body segment is modelled as cylinder with its length listed in Table 1. The radius of the cylinder is set to uniform (0.2 mm) across all leg segments. The data are extracted from the anatomy of real ants. During the locomotion cycle, legs in stance mode are dynamically actuated and legs in swing mode are kinematically animated. This is supported by biologically evidences that each leg only occupy a small proportion of body weight (5%) (Table 2), thus do not pose significant impact on overall body movements.Table 1Length of body segments. Unit: millimeter (mm).CoxaFemurTibiaTarsusFront0.252.492.332.37Middle0.252.822.743.13Back0.253.483.403.80Table 2Mass distribution of body segments. The dimensions of each part are the lengths along x, y, z axes.Dimension (mm)Mass (mg)Mass Proportion (%)Head2.67*2.20*0.505.2321.75Thorax3.07*1.01*0.503.8716.09Abdomen3.43*2.46*0.5011.3547.19Front Leg0.2*0.2*7.440.605.00Middle Leg0.2*0.2*8.940.504.12Back Leg0.2*0.2*10.930.705.82 | study | 100.0 |
Real animal data on ground reaction forces and kinematics were obtained from a previous study on the leg functions of desert ants Cataglyphis fortis24,25. The motion of the walking ants in an enclosed chamber was captured by a Photron Fastcam SA3 (San Diego, CA, USA) camera at a frequency of 500 Hz from the lateral and the dorsal view (Fig. 3). Their ground forces were recorded with a 4 × 4 mm force platform of resolvable forces Fx = 5.4 μN, Fy = 2.9 μN and Fz = 10.8 μN and natural frequencies of fx = 380 Hz, fy = 279 Hz and fz = 201 Hz26. The signals were amplified by a data acquisition system (MGCplus, Hottinger Baldwin Messtechnik, Darmstadt, Germany) and recorded with a sampling frequency of 1200 Hz. After each recording, the body mass was measured with an analytical balance of ±0.2 mg reproducibility (readout = ±0.1 mg, ABS 80-4, Kern & Sohn, Germany). A total of 274 trials with ground force measurements and video sequences were recorded for level locomotion. Trials in which the ants stopped, changed direction or touched the force platform with their gasters were excluded from the dataset. Subsequently, the first 15 records for each set of legs (front legs, middle legs, hind legs) were selected for data analysis resulting in 45 different strides. Besides the petiole, the head-thorax joint and the claws, the trajectories of the femur-tibia joint were additionally tracked for this study with Digitizing Tools 2016071127 and MATLAB R20015b (The MathWorks, Natick, MA, USA). To further model the joint torques, body parts such as the head, the thorax–petiole-coxae, the gaster and the femur-tibia-tarsi (legs) were weighed and photographed for 15 randomly selected ants.Figure 3Video screenshots of real ant locomotion. The red curve denotes the trajectory of body center, the yellow triangle represents the supporting triangle formulated by touchdown positions of stance feet while the blue arrows denotes the force vectors at the stance foot. (a) dorsal view. (b) lateral view. | study | 100.0 |
Video screenshots of real ant locomotion. The red curve denotes the trajectory of body center, the yellow triangle represents the supporting triangle formulated by touchdown positions of stance feet while the blue arrows denotes the force vectors at the stance foot. (a) dorsal view. (b) lateral view. | other | 99.94 |
Previous paragraphs explain the hierarchy of our simulation framework. How to find the appropriate values for both the controller and muscle parameters is key to achieve stable performance of the locomotion. We here use the algorithm of Covariance Matrix Adaptation (CMA) to optimize the parameters. CMA is an evolutionary algorithm that is designed to solve nonlinear non-convex optimization problem. Existing works have applied this method to design the locomotion controller for virtual characters11,12,28. | study | 99.9 |
The optimization is initialized with the mean vector m and the standard deviation vector σcma. For each new generation, λ offspring individuals are sampled with a normal distribution:10\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\bf{x}}}_{{\bf{i}}}={\bf{m}}+\sigma N\mathrm{(0},{\bf{C}}),{\bf{for}}\,{\bf{i}}=1,\cdots ,\lambda $$\end{document}xi=m+σN(0,C),fori=1,⋯,λThe mean vector m represents the favourite solution from last generation. The covariance matrix C determines the shape of distribution ellipsoid. This covariance matrix is updated so that to increase the likelihood of successful individuals. | other | 78.5 |
All individuals are evaluated and ranked by a fitness function f(xi). μ individuals with the lowest fitness score are selected and named as the group of elites. The new mean of the next generation is updated from a weighted sum of these μ elites. The covariance matrix is updated similarly as a combination of covariance from previous generation. By iteratively repeating this process, the samples are expected to converge to the optimal solution. | other | 99.6 |
In the model proposed in this work, the character has 3 DOFs (3 pairs of muscles) for each leg, and the dimensions of the neuron and muscle parameters are 3 and 8 respectively for each neuron-muscle pair. The symmetry between left and right legs first reduces the dimensions of motor and muscle parameters to half. Together with the synaptic weights in the central neural system, the total dimension of parameters is 806. This far exceeds the dimensions of parameter space in existing works11,12,28 and the capability of the algorithm as declared by the author29. To optimize in the parameter space with such complexity, we resolve this challenge with a two-stage strategy with the algorithm of Covariance Matrix Adaptation (CMA). First, we optimize the input currents Is to the motor neurons and muscle parameters for the standard case of walking straight forward. Next, we optimize the offset currents from the central neurons, which are added to the standard case Is and generate the versatile motor skills including walking on curve trajectories and uneven terrains. | study | 100.0 |
The first is to move forward at a predetermined velocity ν* of body center (a vector pointing straight forward):11\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{v}=\sum _{k}{(\nu -{\nu }^{\ast })}^{2}$$\end{document}fv=∑k(ν−ν⁎)2where ν is the actual moving velocity of body center. The subscript k sums up all the simulation steps. The velocity is three dimensions (x, y, z) thus this objective function also penalizes the case when the body falls down or deviates from the forward direction. | other | 99.4 |
The second is to match the foot trajectory pf with the captured trajectory \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\bf{p}}}_{f}^{\ast }$$\end{document}pf⁎ from the real insect:12\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{t}=\sum _{k}{({{\bf{p}}}_{f}-{{\bf{p}}}_{f}^{\ast })}^{2}$$\end{document}ft=∑k(pf−pf⁎)2By minimizing this objective component, we expect to reproduce the similar gait pattern as the real insects. Ants typically demonstrate the double-tripod gait11,30. The intra-leg coordination between joints on the same leg is achieved by matching the trajectory of the particular leg, while the inter-leg coordination between different legs is enforced by the spatial-temporal information embedded in the trajectories of all legs on real ants. The use of motion trajectories allows the compact representation of the desired gait pattern, as part of the optimization goal. | study | 100.0 |
The third is to minimize the energy consumption of the neural system, that is to use as little current input I as possible:13\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{s}=\sum _{k}{I}^{2}$$\end{document}fs=∑kI2This objective is biologically valid since the individuals use less energy for neural control gains advantages during the process of evolution. | other | 99.6 |
The final objective function is defined as:14\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$f={\omega }_{s}\,{f}_{s}+{\omega }_{v}\,{f}_{v}+{\omega }_{t}\,{f}_{t}$$\end{document}f=ωsfs+ωvfv+ωtftωs = 1.0, ωv = 10.0, ωt = 0.5 are weights to balance the significance of three objective components. Our results are not sensitive to the exact values of these weights, so other values within the same order of magnitude may be used as well. | study | 82.2 |
To accomplish this goal, we generate Ns samples from 2D uniform distribution of two variables: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\theta }_{s}\in [-{\theta }_{max}^{s},{\theta }_{max}^{s}]$$\end{document}θs∈[−θmaxs,θmaxs], \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\theta }_{t}\in [-{\theta }_{max}^{t},{\theta }_{max}^{t}]$$\end{document}θt∈[−θmaxt,θmaxt] and enforce an objective function focusing on the trajectories of body center and foot:15\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$f=\sum _{{N}_{s}}({\omega }_{s}\,{f}_{s}+{\omega }_{v}\,{f}_{v})$$\end{document}f=∑Ns(ωsfs+ωvfv) | study | 99.6 |
The code is written in C++ and runs on a standard PC with four CPU cores at 3.4 GHz. The physics simulation is done with the Bullet31 physics engine with a simulation step of 5 ms. Our single-thread implementation runs in real time during the online simulation, while the optimization process runs with the technique of multi-thread, costing around 10 hours for Stage I and 22 hours for Stage II on the aforementioned hardware. In comparison to the discretized look-up table11, our work provides a continuous mapping with comparable timecost. Our method also avoids manually selecting the value of the substep width which critically determines the quality and time performance of the look-up table method11,32. During the optimization, the simulation for each individual is run for 100 seconds (or 100 locomotion cycles) for Stage I, or 10 seconds (or 10 locomotion cycles) for Stage II, or is terminated in advance if the body falls down (the height of body center falls below a threshold). Figure 4 gives a screenshot of the locomotion sequence of a virtual ant in a variety of scenarios.Figure 4Screenshot of locomotion sequence of a virtual ant. (a) Walking straight forward. (b) Walking along a curve. (c) Walking on uneven terrain. | study | 99.94 |
The total dimensions of the parameter space is 11*18 = 198 for Stage I. For each generation, λ = 120 individuals are sampled from the elites and μ = 40 elites are selected out of the whole population. The values for the parameters for all neuron and muscle units are initialized as: I = 10 mV, Fmax = 1, β = 1.5, ρ = 2.0, ω = 3.0, Vmax = 5.0, cV = −7.0, bV = 0.67, aV = −1.668. The triggering timing of the neuron models are set as Ts = 0,Te = 0.5 for left front, right middle, left back legs, and Ts = 0.5, Te = 1.0 for right front, left middle, right back legs. The success of the optimization is confirmed with the decreasing value of the objective function (Fig. 5). The objective function converges quickly in the first 2000 generations and slowly afterwards.Figure 5Objective function value of best individual from each generation for optimization Stage I (a) and Stage II (b). | study | 100.0 |
The total dimensions of the parameter space is 2*16 + 16*36 = 608 for Stage II. The values for the weight parameters are constrained as [0, 1]. The optimal values for the muscle parameters in Stage I (standard case of walking forward) are used at this stage. In practice, we found that it is difficult for the solution to converge given the high dimensions of the parameter space. To ensure the successful convergence, we initially sample the candidates from a narrower value range of [−\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha {\theta }_{max}^{s}$$\end{document}αθmaxs, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha {\theta }_{max}^{s}$$\end{document}αθmaxs] and [−\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha {\theta }_{max}^{t}$$\end{document}αθmaxt, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha {\theta }_{max}^{t}$$\end{document}αθmaxt], where α ∈ . α starts from value of 0 and progressively increase to 1, with a step of 0.2, so that the initial value stays sufficiently close to the optimal value. Figure 5(b) shows the optimization progress for Stage II. The result shows that when the exploration range is small (for cases when α = 0.2 or 0.4), the optimization converges fast and smooth, while for other cases the optimization requires more iterations to converge. | study | 100.0 |
Figure 6 plots the spiking pattern and muscle activation signals of the retractor neuron-muscle unit from the body-coxa joint on the left-sided and right-sided middle legs, when the character turns left. With the activation dynamics (Equation 6), the discrete spiking signal is converted into a continuous activation stimulus. The result shows that when the character turns left, the number of spike trains on the left-sided leg is greater than the one on the right-sided leg. This leads to a larger activation signal for the muscle actuation. The result also shows that the coordination between different legs is achieved by preferably adjusting the number of spike trains, instead of the amplitude of the spike.Figure 6Spiking pattern and muscle activation of the neuron-muscle unit from the body-coxa joint on the left-sided and right-sided middle leg, when the character turns left. (a) Spiking pattern on the left leg. (b) Activation signal on the left leg. (c) Spiking pattern on the right leg. (d) Activation signal on the right leg. | study | 100.0 |
Spiking pattern and muscle activation of the neuron-muscle unit from the body-coxa joint on the left-sided and right-sided middle leg, when the character turns left. (a) Spiking pattern on the left leg. (b) Activation signal on the left leg. (c) Spiking pattern on the right leg. (d) Activation signal on the right leg. | other | 99.9 |
One of the objective functions is to minimize the difference between the trajectory of the real and virtual insects. We here compare the joint angles at three DOFs on the middle leg for both the real and simulated subjects (Fig. 7). We found that the overall shape of the trajectory is consistent between the real and simulated cases. However in the result of real cases, the joint angles decrease to a minor extent before increasing incrementally. This is not observed in the simulated case, in particular for the body-coxa joint. This mismatch may be caused by the posture adjustment of real ants, or possibly by the hardware limitation (image resolution and framerate) of data collection. In the simulation case, the body-coxa joints move at a strict forward pattern in a single motion cycle.Figure 7Joint angles on the middle leg captured from real insects during a locomotion cycle. Dashed lines are trajectories from real ants, solid line (in bold) is from virtual ant with Hill-type muscle and dashed line (in bold) is from virtual ant with PD servo. | study | 100.0 |
Figure 8 plots the joint torques on the middle leg during the stance stage for both the simulated and real cases. The torque from the simulation cases are recorded as the torques applied on the joint constraints during the simulation step. The torque from the real cases are computed with the method of Virtual Model Control11. This method is based on Jacobian matrix, converting the ground force on the end-effector into the joint torques given the configuration of body hierarchy. The peak of the simulation torques arises in the first half of the stance stage for the body-coxa and femur-tibia joints, and in the middle of the stance stage for the coxa-femur leg. However, the muscle activation in the simulation case reaches its peak around the end of the stance stage, instead of the first half. This is possibly due to the fact that the joint rotation reaches its limit at the end of the stance stage, resulting in a small value of FL and FV (Equation 9) and thus a moderate value of joint torque. The torque profiles of real cases generally follow the shape of a parabolic curve and have the peak value around the middle of the stance stage. This is consistent with the ground force on the middle leg (Fig. 9(b)), which presents a similar parabolic curve shape.Figure 8Joint torques on the middle leg during the stance stage in the mode of walking straight forward.Figure 9Ground forces in the Z direction during the stance stage in the mode of walking straight forward. | study | 100.0 |
Figure 9 plots the ground forces on the real ant in the Z (vertical up) direction during the stance stage. For the front leg, the force magnitude present small variations during the whole stance stage. For the middle leg, the force magnitude increases initially, remains stable and decreases when the stance stage comes to the end. This creates a parabolic curve of the force profile. For the back leg, it presents a distinct peak during the first half of the stance stage, which is in consistent with the torque results in the simulation case. | study | 100.0 |
Replicating the function of neural system in real organism is non-trivial, given the fact that the machinery underlying the observed behaviour is not fully understood32. Existing works have proposed different paradigms to model the behaviour neural system. For example, previous work uses a network of nonlinear oscillator to construct the Central Pattern Generator11. We believe the use of different models has its own pros and cons. The oscillator with inherent properties of limit cycle and attraction point could generate the basic pattern with no intervention from higher control system. The high-level neural system is encoded as a look-up table11, which is pre-computed and discretized. In comparison, the current work, which is based on spiking neuron models, provides a unified representation of neural system. The spiking pattern of the unit neurons can be used to seamlessly connect the procedures of sensory processing, high-level commands and low-level control strategy. Such a unified representation lays the foundation for processing sensory information via large-scale networks. | study | 99.94 |
The use of nonlinear Hill-type muscle improves the naturalness of the synthetic motion on the examples of biped28, and research also show that it improves the 2D motion stability for insect simulation33. However the significance of non-linear muscle has not been fully evaluated in the example of a 3D insect model. We here compare the performance of two actuator choices: Proportional-derivative (PD) servo11 and nonlinear Hill-type muscle, in terms of the naturalness of the synthetic motion (Fig. 7). The PD servo is defined as:16\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tau ={k}_{g}({q}_{t+1}-{q}_{t})+{k}_{d}{\dot{q}}_{t}$$\end{document}τ=kg(qt+1−qt)+kdq˙t\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${q}_{t+1},{q}_{t},{\dot{q}}_{t}$$\end{document}qt+1,qt,q˙t are the rotational angles and velocity of the joint. kg, kd are the gain and damping coefficients. | study | 100.0 |
To quantitatively evaluate the difference between the ground-truth data and the simulated trajectories, we use the Euclidean distance between two vectors:17\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$d=||{{\bf{T}}}^{g}-{{\bf{T}}}^{s}||$$\end{document}d=||Tg−Ts||where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\bf{T}}}^{g}=[{T}_{1}^{g},{T}_{2}^{g},\cdots {T}_{n}^{g}],{{\bf{T}}}^{s}=[{T}_{1}^{s},{T}_{2}^{s},\cdots {T}_{n}^{s}]$$\end{document}Tg=[T1g,T2g,⋯Tng],Ts=[T1s,T2s,⋯Tns] are joint trajectories of rotation in one step cycle from the ground truth and simulated result respectively. Tg is computed as the average of all collected trajectories of real ants. | study | 100.0 |
The quantitative result of comparison (Table 3) shows that trajectories of our model presents less deviations from the ground-truth data than the model in11. This is mainly caused by the linear feature of the PD servo. We also notice that the peak values of the trajectories of the previous model11 are generally higher than the ones of the proposed model in this work. The reason is possibly caused by the fact of over-shooting34. The problem of over-shooting refers to the fact that the PD servos normally require a large value of kg, thus causing the over-protraction/elevation/extension of the foot.Table 3Quantitative comparison of the trajectory difference between the ground-truth data and the ones generated by our model and the previous model11.JointsOur ModelPrevious Model11body-coxa0.71881.4827coxa-femur0.34671.7306femur-tibia0.49624.1081 | study | 100.0 |
The neuro-musculo-skeletal model we developed yields very promising results. The CMA optimization algorithm successfully helps to find optimal parameter configuration for the model. The optimization process converges quickly in around 2000 iteration. And the simulated gait pattern is highly consistent with the experimental data collected from real ants walking in the free mode. | study | 99.94 |
An interesting and promising point is that the joint torques of the simulated and real cases also present high consistency. It will be a great help to biologist in studying of some special phenomenons of insects such as the incredible lifting ability of ants and impressive leaping ability of mantis. They can easily observe and analysis the stress conditions under various situations with the simulated model. And the study can stimulate the development of bionic design. | other | 99.9 |
The model could also be beneficial in studying of higher level behavior of insects. With the model, we can reproduce the motion of an insect with only the recorded foot trajectories. Thus, we can use techniques of machine learning to extrapolate or predict the trajectories of insects under different circumstances and then reconstruct the 3d motion with our model. Our model is intended to be widely applicable in its ability to model insect locomotions. Here we only conduct a primary validation on ants. It will be interesting to apply this model to other insect species in future, potentially including other locomotion constraints. | study | 99.2 |
In conclusion, we have proposed a neuro-musculo-skeletal model of virtual insects. By incorporating some nature-tested mechanisms of real insects, the model is capable of reproducing the gait pattern observed from real insects on virtual ones. There are a couple of directions for future works. The current implementation uses a single neuron to simulate the neuron group activating the same muscle. Improving the model complexity to match the real insect is worth exploring. To extend this framework to these advanced motion skills is one of the future efforts. How to encode more sophisticated behaviours, such as collective transport35, as patterns of spiking trains is even more challenging. | study | 99.94 |
Thyroid cancer is the endocrine malignancy1 that, although stable until the 1990s, has progressively and greatly increased thereafter2, 3. The vast majority of the thyroid cancers originate from the follicular cell epithelium1, which includes papillary thyroid carcinoma (PTC) that accounts for approximately 80% of all thyroid malignancies4. | review | 98.6 |
Thyroid oncogenesis is still under investigation, however a high frequency (70%) of activating mutations in components of the mitogen-activated protein kinase (MAPK) pathway was reported, such as BRAFV600E 5, 6 and HRAS/NRAS/KRAS point mutations7, 8. Also, fusions involving the RET9 and NTRK1 tyrosine kinases10 were described to promote thyroid cancers. More recently, the set of known PTC driver alterations was extended to include EIF1AX, PPM1D, and CHEK27. | review | 99.44 |
Additionally of being the most frequent mutation in many types of cancers including PTC7, 11, BRAFV600E confers poorer prognosis compared to other oncogenes. There is a growing number of evidence demonstrating that BRAFV600E correlates with metastasis, cancer recurrence12 and higher mortality in PTC13. BRAFV600E-expressing cells have a diversity of malignant characteristics, including increased DNA synthesis, dedifferentiation, and chromosomal instability14. Also, BRAFV600E stimulates more actively MEK-dependent invasion than the expression of RET/PTC oncoprotein through the expression of matrix metalloproteinases (e.g. MMP-3, MMP-9 and MMP-13), which, in part, can explain the more aggressive BRAFV600E-induced phenotype15. | study | 99.9 |
Similarly to melanoma16, BRAF mutation occurs at early stages of PTC development11, 17. Besides all BRAF oncogenic activities, its single exacerbated stimulation of the MAPK pathway is not sufficient to sustain malignant transformation, resulting in induced senescence18 that confers a barrier to tumor progression19. To bypass BRAF-induced senescence, cells may suffer a second event that allows malignant transformation, as possibly the epigenetic silencing of tumor suppressors DAPK, TIMP320, SLC5A820, 21 and hMLH122 and other BRAF-induced mechanisms that remain to be discovered11. In thyroid cancers, Thyroid-stimulating Hormone (TSH) is more involved in overcoming senescence; while BRAF overexpression suppresses thyroid hormone biosynthesis and leads to elevated TSH levels in vivo 14; it was shown that TSH signaling inhibits BRAFV600E-induced senescence through DUSP623. | study | 100.0 |
Recently, BRAFV600E-associated mRNA signature was determined in a mouse model and human samples24, which identified new genes not previously reported as related to BRAF mutation in thyroid cancer (e.g. MMD, ITPR3, AACS, LAD1, PVRL3, ALDH3B1, and RASA1) that will provide further support for future research on BRAF-induced PTC24. However, this analysis did not evaluate the expression of long noncoding RNAs (lncRNAs), which are progressively shown to be of fundamental importance in other types of cancer25, 26. Such analysis is necessary for the identification of BRAFV600E-correlated long noncoding RNAs. | study | 100.0 |
LncRNAs are RNAs longer than 200 nucleotides that have no coding potential27 and are involved in several processes, such as gene expression regulation through chromatin modulation28, 29, epigenetic control30, association with translational apparatus31, improving other mRNA stability32, serving as a scaffold for protein33, acting as decoys for miRNAs34, altering protein turnover35, among others. | review | 99.44 |
To date, as per the authors’ knowledge, only a few published studies identified differentially expressed (DE) lncRNAs between normal (N) and tumoral thyroid (T) in a limited set of paired samples36, 37. Although these findings laid the foundation for further investigation of lncRNAs related to PTC36, they need to be confirmed in a more numerous group of patients. | review | 99.6 |
In this study we confirmed previously reported lncRNAs and determined new DE lncRNAs in PTC in a larger set of samples and also identified BRAFV600E-correlated lncRNAs, providing possible candidates that can constitute secondary mechanisms of BRAF- induced malignancy in PTC. | study | 100.0 |
Comparative analysis between 59 pairs of matched normal and papillary thyroid cancer samples identified 455 differentially expressed lncRNAs (log2 fold change > 1 or < −1; adj. p-value < 1 × 10−7; Fig. 1A), being 71 lncRNAs upregulated and 384 lncRNAs downregulated in PTC (Supplemental Table S1). The same samples presented a total of 2016 mRNAs (log2 fold change > 1 or <−1; adj. p-value < 0.05; Supplemental Table S2) and 186 microRNAs (log2 fold change > 1 or < −1; adj. p-value < 0.05; Supplemental Table S3) differentially expressed.Figure 1Differentially expressed lncRNAs between N × T and WT × BRAFV600E were identified. (A) Volcano plot of DE lncRNAs between N × T (log2 fold change > 1 or < −1; adj. p-value < 1 × 10−7). (B) Volcano plot of DE lncRNAs between WT × BRAFV600E (log2 fold change > 1 or < −1; adj. p-value < 1 × 10−4). (C) Venn Diagram of common DE lncRNAs between N × T and WT × BRAFV600E. (D) Heatmap* of DE lncRNAs between N × T (log2 fold change > 3 or < −3; adj. p-value < 1 × 10−7). (E) Heatmap* of DE lncRNAs between WT × BRAFV600E (log2 fold change > 2.5 or < −2.5; adj. p-value < 1 × 10−4). *For hierarchical clustering, one minus Spearman rank correlation was performed. | study | 100.0 |
Differentially expressed lncRNAs between N × T and WT × BRAFV600E were identified. (A) Volcano plot of DE lncRNAs between N × T (log2 fold change > 1 or < −1; adj. p-value < 1 × 10−7). (B) Volcano plot of DE lncRNAs between WT × BRAFV600E (log2 fold change > 1 or < −1; adj. p-value < 1 × 10−4). (C) Venn Diagram of common DE lncRNAs between N × T and WT × BRAFV600E. (D) Heatmap* of DE lncRNAs between N × T (log2 fold change > 3 or < −3; adj. p-value < 1 × 10−7). (E) Heatmap* of DE lncRNAs between WT × BRAFV600E (log2 fold change > 2.5 or < −2.5; adj. p-value < 1 × 10−4). *For hierarchical clustering, one minus Spearman rank correlation was performed. | study | 100.0 |
Differential expression analyses were also performed to identify BRAFV600E-correlated lncRNAs. The comparison between BRAF wild type (WT) patients (n = 242) and BRAFV600E patients (n = 226), determined 437 differentially expressed lncRNAs (log2 fold change > 1 or < −1; adj. p-value 1X10−4; Fig. 1B), being 117 upregulated and 320 downregulated (Supplemental Table S4). The same comparison found a total of 924 mRNAs (log2 fold change > 1 or < −1; adj. p-value < 0.05; Supplemental Table S5) and 94 microRNAs (log2 fold change > 1 or < −1; adj. p-value < 0.05; Supplemental Table S6) differentially expressed. A total of 103 lncRNAs was differentially expressed in both analyses [(Normal × Tumor and WT × BRAFV600E), (Fig. 1C, Table 1 and Supplemental Table S7)].Table 1Top 5 upregulated and 5 downregulated common DE lncRNAs between Normal × Tumor and WT × BRAFV600E.Ensembllog2 FCAdj. p-valuelog2 FCAdj. p-valueN × TN × TWT × BRAFV600E WT × BRAFV600E ENSG00000214797.3 7.016.39E-095.194.92E-12 ENSG00000273132.1 5.494.17E-101.693.70E-12 ENSG00000230918.1 5.491.17E-101.363.89E-05 ENSG00000260328.1 5.394.27E-084.042.26E-16 ENSG00000256268.1 5.122.27E-141.004.52E-13 ENSG00000261185.1 −6.421.13E-10−2.813.55E-06 ENSG00000254489.1 −6.377.98E-15−3.278.33E-11 ENSG00000260412.1 −6.337.11E-09−3.772.59E-08 ENSG00000235070.3 −4.972.28E-12−3.011.10E-15 ENSG00000247416.2 −4.556.89E-13−2.401.09E-16 | study | 100.0 |
Experimental validation using qRT-PCR was performed to demonstrate the reliability of the bioinformatics analyses applied. From the top 25 positively DE lncRNAs and from the top 20 negatively DE lncRNAs, it were selected for validation those lncRNAs with low adjusted p-values to minimize expression variability, especially in the comparison BRAFWT × BRAFV600E tumor, among others characteristics (for detailed information see Methods). From a total of 5 DE lncRNAs selected for validation from the TCGA analysis (Fig. 2 and Supplemental Fig. S1, upper part), 4 DE lncRNAs were validated using thyroid cell lines (Fig. 2 and Supplemental Fig. S1, lower part), which strengths the reliability of this bioinformatics analysis, although the experimentally tested set of lncRNAs constitutes a relatively small sampling. Considering all comparisons, we obtained a very expressive validation efficiency [from a total of 8 different comparisons (Normal × Tumor and BRAFWT × BRAFV600E), 6 were validated]. Downregulation of ENSG00000235070.3 and ENSG00000255020.1 in PTC was confirmed in the tumor cell lines TPC1 (BRAFWT) and BCPAP (BRAFV600E) compared to the normal immortalized cell line NTHY (Fig. 2). Also, downregulation of their expression was in accordance to the bioinformatics analysis, since lower expression for both of them was observed in BCPAP (BRAFV600E) compared to TPC1 (BRAF wild type) (Fig. 2). Noteworthy, is that due to the very low abundance of ENSG00000255020.1 in the BCPAP cell line, qRT-PCR resulted in two unspecific melting peaks, which did not influenced the results. Upregulation of ENSG00000273132.1 in PTC was confirmed, however its overexpression in BRAFV600E tumors was not observed in the cell line BCPAP compared to TPC1 (Fig. 2), maybe due to the small log2 fold change value (1.69) of this comparison. Overexpression of ENSG00000230498.1 in BRAFV600E PTC compared to BRAF wild type tumors was also confirmed (Supplemental Fig. S1); nevertheless ENSG00000247311.2 was undetectable in both TPC1 and BCPAP cells (Supplemental Fig. S1).Figure 2Experimental validation of DE lncRNAs. Upper part of panel displays the expression levels of the indicated lncRNAs in the TCGA analyses. The nonparametric Mann–Whitney test was applied due to the non-Gaussian expression distribution and p-value was assigned. Lower part of panel displays the experimental validation of these lncRNAs measured by qRT-PCR and calculated with 2−ΔΔCt method using RPL19 (Ribosomal Protein L19) as endogenous control. Experiments with three biological replicates were performed using two technical replicates for each sample. These results are representative of at least two independent experiments. Values are plotted as expression mean ± Standard Error of Mean (SEM). Unpaired two-tailed t-Test assigned the p-value. | study | 100.0 |
Experimental validation of DE lncRNAs. Upper part of panel displays the expression levels of the indicated lncRNAs in the TCGA analyses. The nonparametric Mann–Whitney test was applied due to the non-Gaussian expression distribution and p-value was assigned. Lower part of panel displays the experimental validation of these lncRNAs measured by qRT-PCR and calculated with 2−ΔΔCt method using RPL19 (Ribosomal Protein L19) as endogenous control. Experiments with three biological replicates were performed using two technical replicates for each sample. These results are representative of at least two independent experiments. Values are plotted as expression mean ± Standard Error of Mean (SEM). Unpaired two-tailed t-Test assigned the p-value. | study | 100.0 |
For downstream analyses, we increased the stringency of differentially expressed lncRNAs between Normal × Tumor (log2 fold change > 3 or < −3; adj. p-value <1 × 10−7, n = 73; Fig. 1D) and between WT × BRAFV600E (log2 fold change >2.5 or < −2.5; adj. p-value < 1 × 10−4, n = 59; Fig. 1E) to analyze the lncRNAs that were most DE. Hierarchical clustering was used to organize patients or lncRNAs into groups according to the expression levels of DE lncRNAs. | study | 100.0 |
Results demonstrated that this set of lncRNAs is capable of clustering, majorly, normal and cancer patients in two distinct groups (Supplemental Fig. S2A). Clustering lncRNAs by Spearman correlation among all DE lncRNAs also identified two groups of lncRNAs highly positively correlated or negatively correlated (Supplemental Fig. S3A). | study | 100.0 |
Hierarchical clustering was also performed with a more stringent set of DE lncRNAs between WT and BRAFV600E, which allowed the clustering of two groups enriched with WT and BRAFV600E patients, respectively (Supplemental Fig. S2B). Clustering lncRNAs by Spearman correlation among all DE lncRNAs also identified two groups highly positively correlated or negatively correlated lncRNAs (Supplemental Fig. S3B). | study | 100.0 |
As almost the totality of the identified DE lncRNAs in both conditions (Normal × Tumor and WT × BRAFV600E) is uncharacterized, we used prediction methods to identify a possible interaction between lncRNAs and mRNAs/microRNAs. Predicted mRNAs and microRNAs (targets of DE lncRNAs) were compared to differentially expressed mRNAs and microRNAs (log2 fold change >1 or <−1; adj. p-value < 0.05) calculated from the same TCGA patients. Predicted mRNAs/microRNAs that were also identified as DE were considered as indirectly validated targets. | study | 100.0 |
A total of 1109 DE mRNAs (Table 2 and Supplemental Table S8) and 26 DE microRNAs (Supplemental Table S9) were found to be predicted targets of the DE lncRNAs between Normal and Tumor samples and were considered as indirectly validated targets. Gene ontology and KEGG pathways enrichment of these validated mRNAs demonstrated that most of the genes are involved in surface receptors responsible for signal transduction and cell adhesion, as well as, regulation of cell death, proliferation and apoptosis (Fig. 3A). Enriched pathways (Fig. 3B) were composed of cytokine-cytokine receptor interaction, pathways in cancer (Fig. 3C), focal adhesion, MAPK pathway and calcium signaling pathway. Validated microRNAs were also used to determine enriched pathways based on their predicted targets calculated elsewhere (Fig. 3D). Genes involved in cancer and MAPK pathways were the most enriched pathways. Interestingly, some genes predicted to be targets of the validated microRNAs were also DE expressed in our analysis (Supplemental Table S2), such as upregulation of the MAPK constituents, CACNG4, CACNA1E, DUSP4, TGFBR1, FGF1, FGF2 and MAP3K1. Enriched pathways were extended to genes involved in cancer, focal adhesion and calcium signaling (Fig. 3D).Table 2Top 5 upregulated and 5 downregulated DE lncRNAs between paired Normal × Tumor with examples of indirectly validated targets.Ensembllog2 FCAdj. p-valueUpregulated indirectly validated targetsDownregulated indirectly validated targetsN × TN × T ENSG00000223914.1 7.048.69E-11VGLL1, GDF6, FAM19A2, HRH1RPS6KA5, RNF150, ANK2, FOSB ENSG00000250748.2 7.014.56E-09FUT3, GPR115, CAPN8, COL7A1SVEP1, LMOD1, DPT, KCNA1 ENSG00000214797.3 7.016.39E-09TMEM130, HES2, KCP, DTX4CDHR3, PAK3, RASSF6, NWD1 ENSG00000273132.1 5.494.17E-10KLK6, ELFN2, C19orf59, SHISA6SRF, CPXM1, LAYN, FAM163A ENSG00000230918.1 5.491.17E-10GRM4, DPP4, LRP4, SHISA6EGR1, TFCP2L1, FOXJ1, ABCA9 ENSG00000253288.1 −7.431.03E-09SLC6A20, KLK10, FUT3, HRH1HAP1, SH2D6, FOXP2, ADH1B ENSG00000272479.1 −7.191.16E-09DMBX1, TMPRSS6, TMPRSS4, PPP1R1BCUX2, PAX1, CLCNKB, FOSB ENSG00000261185.1 −6.421.13E-10B3GNT3, ELFN2, LRP4, SHISA6NR4A1, C1QTNF7, RNF150, FAM180B ENSG00000254489.1 −6.377.98E-15SYTL5, HPCAL4, KCNQ3, CPNE4RBM24, PGA3, GFRA1, RNF150 ENSG00000260412.1 −6.337.11E-09CLDN16, PDE4C, LRG1, SHROOM4SLC26A4, CDHR3, PAK3, NWD1 Figure 3Indirectly validated targets of the DE lncRNAs between N × T are involved in cancer-related processes. (A) GO biological processes and (B) KEGG enriched pathways of the indirectly validated mRNA targets of the DE lncRNAs between N × T. (C) Proteins’ network of genes linked to the pathways in cancer, where black circles are validated targets and grey circles are connective proteins. (D) KEGG Pathways enrichment of the indirectly validated microRNAs targets of the DE lncRNAs between N × T. | study | 100.0 |
Indirectly validated targets of the DE lncRNAs between N × T are involved in cancer-related processes. (A) GO biological processes and (B) KEGG enriched pathways of the indirectly validated mRNA targets of the DE lncRNAs between N × T. (C) Proteins’ network of genes linked to the pathways in cancer, where black circles are validated targets and grey circles are connective proteins. (D) KEGG Pathways enrichment of the indirectly validated microRNAs targets of the DE lncRNAs between N × T. | study | 100.0 |
Between WT and BRAFV600E, 471 DE mRNAs (Table 3 and Supplemental Table S10) and 11 DE microRNAs (Supplemental Table S11) were indirectly validated. Gene ontology of these mRNAs demonstrated that most of the genes are also related to surface receptors involved in signal transduction and cell adhesion, but, additionally, with response to hormone stimulus and transmembrane transport (Fig. 4A). Enriched pathways (Fig. 4B) were constituted of calcium signaling pathway (Fig. 4C), cardiomyopathies and ECM-receptor interaction. KEGG enrichment pathway analysis of the validated DE microRNAs demonstrated participation of the MAPK and WNT pathways, as well as regulation of actin cytoskeleton and focal adhesion (Fig. 4D). Several pro-oncogenic genes were found to be upregulated in our analysis and were described as predicted targets of the validated DE microRNAs, as the example of MET and TGFBR1 genes (Supplemental Table S5).Table 3Top 5 upregulated and 5 downregulated DE lncRNAs between WT × BRAFV600E with examples of indirectly validated targets.Ensembllog2 FCAdj. p-valueUpregulated indirectly validated targetsDownregulated indirectly validated targetsWT × BRAF V600EWT × BRAF V600E ENSG00000255595.1 9.875.68E-05TCAP, ITGA2, LY6G6C, BEND6SLC5A5, ASTN1, PART1, IRX6 ENSG00000214797.3 5.194.92E-12HES2, DTX4, KCP, LDLRRNF157, HAP1, NWD1, SLC14A2 ENSG00000260328.1 4.042.26E-16TMPRSS4PRND, TMPRSS3, PREX2, CNTNAP2 ENSG00000230498.1 3.955.74E-14C1orf106, ADAMTS14, DMBX1, DUSP13FCGBP, GCGR, SOX3, PPP2R2C ENSG00000256916.1 3.623.51E-09ELFN2, SPTBN2, MUC16, ZNF469SSPO, ASXL3, CTNND2, CNTNAP2 ENSG00000267674.1 −8.553.89E-13VSIG1, LDLRADM2, ST3GAL6, NWD1, SLC5A8 ENSG00000237396.1 −6.353.05E-08SIGLEC6, SDK1, C1QL2, MUC21TFCP2L1, SCUBE1, SBSN, PAX1 ENSG00000227947.1 −6.225.72E-07ELFN2, EPHA10, SLC30A3, SYT1SFTPC, GATA5, SFRP1, MATN1 ENSG00000224568.1 −4.238.95E-18SLC6A14, C1orf106HIF3A, SYT13, SLC29A4, ARSF ENSG00000267214.1 −4.181.17E-06ELFN2, COL7A1, B4GALNT3FAM124A, SULT1A2, SLC29A4, CNTNAP2 Figure 4Indirectly validated targets of the DE lncRNAs between WT × BRAFV600E are involved in oncogenic pathways. (A) GO biological processes and (B) KEGG enriched pathways of the indirectly validated mRNA targets of the DE lncRNAs between WT × BRAFV600E. (C) Proteins’ network of genes linked to calcium signaling pathway, where black circles are validated targets and grey circles are connective proteins. (D) KEGG pathways enrichment of the indirectly validated microRNAs targets of the DE lncRNAs between WT × BRAFV600E. | study | 100.0 |
Indirectly validated targets of the DE lncRNAs between WT × BRAFV600E are involved in oncogenic pathways. (A) GO biological processes and (B) KEGG enriched pathways of the indirectly validated mRNA targets of the DE lncRNAs between WT × BRAFV600E. (C) Proteins’ network of genes linked to calcium signaling pathway, where black circles are validated targets and grey circles are connective proteins. (D) KEGG pathways enrichment of the indirectly validated microRNAs targets of the DE lncRNAs between WT × BRAFV600E. | study | 100.0 |
Long noncoding RNAs are arising as key participants in cancer establishment and progression by several oncogenic mechanisms30, 32. On the other hand, it is of urge interest the determination of how these lncRNAs are activated and how they can be associated with specific events or genotypes, such as point mutations. BRAFV600E is the driver oncogenic mechanism with the greatest incidence in PTC7 and, therefore, any event correlated with this mutation will be necessary to understand BRAFV600E-induced aggressiveness. | study | 99.94 |
We have identified 455 DE lncRNAs between paired normal and PTC samples. A total of 76 (log2 fold change >1 or < −1; adj. p-value < 1 × 10−7) lncRNA were previously reported as DE in thyroid cancer compared to adjacent normal thyroid36 (Fig. 5A and Supplemental Table S12). This validation set, together with the lncRNAs confirmed by experimental approaches (Fig. 2 and Supplemental Fig. S1), confers consistency to our analysis. Additionally, a diversity of DE lncRNAs identified in our analysis were reported in individual studies as altered in PTC samples, such as ENSG00000259104.2 38, 39, ENSG00000236130 40, ENSG00000226363 37, ENSG00000271086 39, 41, ENSG00000223914 37 and ENSG00000187979 37.Figure 5Experimentally validated lncRNAs are important to tumor malignancy. (A) Examples of DE lncRNAs between Normal and PTC, which were confirmed in a validation set36. (B) Differentially expressed lncRNAs in PTC that alter tumor malignancy. | study | 99.94 |
ENSG00000259104.2 (PTCSC3), which is downregulated in the tumor samples (log2 fold change −1.40; adj. p-value 1.11E-12) was previously reported as having thyroid-specific expression and decreased expression in PTC38, 39. Interestingly, the risk allele [T] associated with SNP rs944289, located at PTCSC3’s promoter, affects the binding site of C/EBPα and C/EBPβ (PTCSC3 activators), reducing its expression. Restoration of PTCSC3 expression in PTC cells inhibited cell growth and affected the expression of genes involved in DNA replication/repair, cellular movement and cell death38. Also PTCSC3 ectopic expression reduces cell proliferation and increases cell cycle arrest and apoptosis39, while reducing cell motility and invasiveness through S100A4 downregulation42 (Fig. 5B). | study | 100.0 |
ENSG00000236130 (PTCSC2), was also reported as having decreased expression in PTC40, which was confirmed in our analysis (N × T log2 fold change −1.03; adj. p-value 3.12E-09). The risk allele [A] of rs965513 was significantly associated with low expression of unspliced PTCSC2 in unaffected thyroid tissue, however this correlation was not extended to PTC samples40. | study | 100.0 |
ENST00000426615 (ENSG00000226363) is another lncRNA that we identified as upregulated in PTC (NxT log2 fold change 3.87; adj. p-value 2.36E-05), which was experimentally demonstrated to be overexpressed in this cancer, inducing cell proliferation and motility37 (Fig. 5B). | study | 100.0 |
Our analysis also confirmed the differential expression (N × T: log2 fold change −2.42; adj. p-value 3.96E-11) of the previously reported lncRNA ENSG00000271086 (NAMA), which is downregulated in PTC compared to normal tissues39, 41 and in BRAFV600E tumors compared to wild type tumors39 (Fig. 5B). NAMA is induced by inhibition of the MAPK pathway, growth arrest and DNA damage41 and our analysis also demonstrated that NAMA is downregulated in BRAFV600E patients (WT × BRAFV600E log2 fold change −1.66; adj. p-value 2.02E-15). All these independently validated lncRNAs demonstrate the reliability of our study (Fig. 5). | study | 100.0 |
Similarity matrix based on Spearman correlation identified clusters of DE lncRNAs between Normal × Tumor (Supplemental Fig. S3A) and WT × BRAFV600E (Supplemental Fig. S3B) with similar expression patterns, which can provide evidence for further studies to determine common upstream regulators. | study | 100.0 |
Indirectly validated targets of the DE lncRNAs between Normal × Tumor are involved in a diversity of biological processes (Fig. 3A). For instance, it was noticed an overrepresentation of adhesion molecules, such as downregulation of CDH16, which was already reported as a potential marker for PTC43. Along with CDH16, many other cadherins were identified as validated targets of DE lncRNAs, such as CDH2, CDH3, CDH4, CDH6, CDH11 and CDH24 (Supplemental Table S8). Another highly enriched biological process was the regulation of programmed cell death (Fig. 3A), represented by the upregulation of the antiapoptotic SOX444 and TP6345 in PTC samples. Enriched pathways (Fig. 3B) as cytokine-cytokine receptor interaction, focal adhesion and MAPK pathways were already reported in the first study of DE lncRNA with paired Normal × PTC samples36, providing further support for future research. It was observed an enrichment of MAPK-related genes, represented in our results by upregulation, in the tumor samples, of TGB1, TGFB2 and TGFBR1 that were shown to activate the MAPK pathway46. Interestingly, pathway enrichment analysis of indirectly validated microRNAs (Fig. 3D) demonstrated a convergent tendency to genes involved in cancer, MAPK pathway and focal adhesion, which were also observed with the validated mRNAs. | study | 100.0 |
Indirectly validated targets of the DE lncRNAs between WT × BRAFV600E tumors were demonstrated to be involved with cell surface receptors responsible for signal transduction and with cell adhesion (Fig. 4A). Pathway enrichment analysis, identified genes involved in calcium signaling and ECM-receptor interaction, which were already reported as an early transcriptome change in BRAFV600E-associated mouse model24. Interestingly, we also observed several genes correlated with cardiomyopathies that are mostly related to calcium regulation in cardiac muscle cells (Fig. 4B). Calcium (Fig. 4C) and MAPK cascade (represented by BRAFV600E group) are tightly involved, where calcium modulates the protein interaction properties of ERKs, affecting the subcellular localization and influencing the distribution of their targets47. Calcium can also stimulates MEK through Ras activation48. Therefore, these results can support a future investigation to answer if BRAFV600E-stimulated MAPK activation can be reinforced by calcium modulation induced by the DE lncRNAs. Additionally, MAPK stimulation may be supported by the indirectly validated microRNAs, since pathway analysis demonstrated an enrichment of microRNAs’ targets in cancer and MAPK pathways (Fig. 4D). Interestingly, predicted targets of the DE microRNAs were also differentially expressed in our analysis, such as the upregulation of TGFBR1 and downregulation of PRKACB (Supplemental Table S5). | study | 100.0 |
Concluding, our extended cohort of paired Normal and PTC patients identified new DE lncRNAs and confirmed many other lncRNAs already reported. Additionally, to our knowledge, this is the first study to identify BRAFV600E-correlated lncRNAs in PTC, which will provide support for future studies aiming to identify BRAFV600E-linked events in attempt to optimize therapeutic treatment and diagnosis/prognosis of this aggressive PTC genotype. | study | 100.0 |
Thyroid Carcinoma (THCA) clinical information, mRNA and microRNA data expression data were downloaded from The Cancer Genome Atlas (TCGA) online platform (https://tcga-data.nci.nih.gov/tcga/), as January 2016. Mutations data were retrieved through cBioPortal49. LncRNA RPKM expression levels corresponding to TCGA patients were downloaded through TANRIC50, which obtained the genomic coordinates of 13,870 human lncRNAs from the GENCODE Resource (version 19)51 and further filtered out those lncRNA exons that overlapped with any known coding genes based on the gene annotations of GENCODE and RefGene, resulting in 12,727 lncRNAs50. | study | 100.0 |
BRAFV600E patients were selected to form the BRAFV600E group (n = 226), which excluded any other type of BRAF mutations. Wild Type group (n = 242) was formed by patients without any somatic mutation in BRAF gene, but patients with mutations in HRAS, NRAS, KRAS, EIF1AX, PPM1D, RET and NTRK1 were considered. It were selected only the patients with papillary thyroid cancer diagnosis. | study | 99.94 |
For differential expression analysis of mRNA and microRNA was used edgeR package52 through TCGAbiolinks53. To identify differentially expressed lncRNAs between groups, it was used the paired/unpaired Student t test to assess the statistical difference of mean expression values between the two groups50. LncRNA with median value equal zero were excluded and fold change was calculated using median expression values. In all differential analysis, p-values were adjusted for False Discovery Rate (FDR) < 0.05 as multiple hypothesis test correction method. | study | 100.0 |
Total RNA was phenol-chloroform extracted from cell lines Nthy-ori3-1 (NTHY–immortalized human thyroid follicular epithelial cell), TPC1 (papillary thyroid carcinoma- RET/PTC1 rearrangement) and BCPAP (papillary thyroid carcinoma–BRAFV600E) using TRIzol reagent (Invitrogen) according to the manufacter’s instructions. Four µg of total RNA was reverse transcribed using M-MLV Reverse Transcriptase (Invitrogen) in the presence of 100 ng of random hexamers primers according to the manufacter’s protocol. qPCR reaction was performed using 100 ng of cDNA, 1X Power SYBR Green PCR Master Mix (Applied Biosystems) and specific primers. Amplification and detection were performed using ViiA7 TM Real-Time PCR System (Applied Biosystems). Relative gene expression was calculated using the QGENE program and calculated with 2−ΔΔCt method using RPL19 (Ribosomal Protein L19) as endogenous control. | study | 99.94 |
Primers used (5′-3′): RPL19 (Fw-TCTCATGGAACACATCCACAA; Rv-TGGTCAGCCAGGAGCTTCTT), ENSG00000273132.1 (Fw-CTAGCTGCCAGCAGTGACAA; Rv-GCGAGAGCACAGATGACCAC), ENSG00000230498.1 (Fw-CCCTGGGTGATGAAGATGAG; Rv-TGGGATCCCTTTTTTGTCCG), ENSG00000235070.3 (Fw-TGACTCCAAGTTCACGCAGC; Rv-GTGGATGAGTTGTGTGCTGG), ENSG00000255020.1 (Fw-AGTGACGTGGGGAAGAAACG; Rv-CGACATATTTCAAGGGCGCC) and ENSG00000247311.2 (Fw-GCTGTGAGTGACTCTTCAGC and ACAGACACACCCAGGAACAA). | other | 99.9 |
Subsets and Splits