A novel Sentence Transformer for DNA tasks

This is a sentence-transformers model trained. It maps sentences & paragraphs to a 512-dimensional dense vector space and can be used for semantic textual similarity, semantic search, paraphrase mining, text classification, clustering, and more.

Model Details

Model Description

Model Type: Sentence Transformer
Maximum Sequence Length: 512 tokens
Output Dimensionality: 512 tokens
Similarity Function: Cosine Similarity

Model Sources

Documentation: Sentence Transformers Documentation
Repository: Sentence Transformers on GitHub
Hugging Face: Sentence Transformers on Hugging Face

Full Model Architecture

SentenceTransformer(
  (0): SentenceTransformer(
    (0): Transformer({'max_seq_length': 512, 'do_lower_case': False}) with Transformer model: MPNetModel 
    (1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False, 'include_prompt': True})
  )
  (1): CNN(
    (convs): ModuleList(
      (0): Conv1d(768, 256, kernel_size=(1,), stride=(1,))
      (1): Conv1d(768, 256, kernel_size=(3,), stride=(1,), padding=(1,))
      (2): Conv1d(768, 256, kernel_size=(5,), stride=(1,), padding=(2,))
    )
  )
  (2): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False, 'include_prompt': True})
  (3): Dense({'in_features': 768, 'out_features': 512, 'bias': True, 'activation_function': 'torch.nn.modules.activation.ReLU'})
  (4): Normalize()
)

Usage

Direct Usage (Sentence Transformers)

First install the Sentence Transformers library:

pip install -U sentence-transformers

Then you can load this model and run inference.

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("dsfsi/dna-paraphrase-mpnet-base-v2")
# Run on kmer DNA sequences
#
sentences = [
    'ATCCCC ATGGAA ATTTGG',
    'TTGGCC AAAGGG GGGTTT',
    'CCCTTT CTCTGTTT AAATTTTGG',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 512]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities.shape)
# [3, 3]

Downstream Performance

Average Train Accuracy

Dataset	Infersent1	all-MiniLM-L6-v2	all-mpnet-base-v2	paraphrase-mpnet-base-v2	Proposed model
H3	61±0.014	62±0.027	63±0.034	65±0.021	63±0.017
H4	63±0.030	60±0.014	61±0.011	64±0.025	66±0.012
enhancers	53±0.015	77±0.0203	79±0.022	83±0.020	74±0.022
promoter_all	54±0.011	75±0.007	77±0.008	83±0.006	76±0.005
enhancers_types	49±0.004	64±0.027	66±0.021	70±6.435	59±0.013
promoter_no_tata	54±0.007	76±0.011	77±0.013	82±0.011	76±0.011
promoter_tata	51±0.023	74±0.037	74±0.039	82±0.016	71±0.033
splice_sites_donors	54±0.022	55±0.019	56±0.012	58±0.021	58±0.011

Test Accuracy

Dataset	Infersent1	all-MiniLM-L6-v2	all-mpnet-base-v2	paraphrase-mpnet-base-v2	Proposed model
H3	61	64	65	67	67
H4	63	61	60	64	65
enhancers	53	78	76	87	67
promoter_all	54	76	78	83	78
enhancers_types	46	67	68	70	62
promoter_no_tata	54	76	78	83	78
promoter_tata	52	76	77	84	75
splice_sites_donors	58	59	60	63	62

Framework Versions

Python: 3.8.10
Sentence Transformers: 3.0.1
Transformers: 4.43.4
PyTorch: 2.4.0+cu121
Accelerate:
Datasets:
Tokenizers: 0.19.1

Citation

TBA

BibTeX

TBA

Contributing

Your contributions are welcome! Feel free to improve the model.

Model Card Authors

Mpho Mokoatle

Model Card Contact

Email: [email protected]