validate cv
Browse files- constants.py +11 -1
- data/example-predictions-cv.csv +0 -0
- submit.py +7 -2
- validation.py +105 -6
constants.py
CHANGED
|
@@ -36,7 +36,17 @@ REQUIRED_COLUMNS: list[str] = [
|
|
| 36 |
"vh_protein_sequence",
|
| 37 |
"vl_protein_sequence",
|
| 38 |
]
|
| 39 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 40 |
|
| 41 |
# Huggingface API
|
| 42 |
TOKEN = os.environ.get("HF_TOKEN")
|
|
|
|
| 36 |
"vh_protein_sequence",
|
| 37 |
"vl_protein_sequence",
|
| 38 |
]
|
| 39 |
+
# Cross validation
|
| 40 |
+
CV_COLUMN = "hierarchical_cluster_IgG_isotype_stratified_fold"
|
| 41 |
+
# Example files
|
| 42 |
+
EXAMPLE_FILE_DICT = {
|
| 43 |
+
"GDPa1": "data/example-predictions.csv",
|
| 44 |
+
"GDPa1_CV": "data/example-predictions-cv.csv",
|
| 45 |
+
}
|
| 46 |
+
ANTIBODY_NAMES_DICT = {
|
| 47 |
+
"GDPa1": pd.read_csv(EXAMPLE_FILE_DICT["GDPa1"])["antibody_name"].tolist(),
|
| 48 |
+
"GDPa1_CV": pd.read_csv(EXAMPLE_FILE_DICT["GDPa1_CV"])["antibody_name"].tolist(),
|
| 49 |
+
}
|
| 50 |
|
| 51 |
# Huggingface API
|
| 52 |
TOKEN = os.environ.get("HF_TOKEN")
|
data/example-predictions-cv.csv
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
submit.py
CHANGED
|
@@ -11,7 +11,12 @@ from constants import API, SUBMISSIONS_REPO
|
|
| 11 |
from validation import validate_csv_file
|
| 12 |
|
| 13 |
|
| 14 |
-
def make_submission(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 15 |
if user_state is None:
|
| 16 |
raise gr.Error("You must submit your username to submit a file.")
|
| 17 |
|
|
@@ -34,7 +39,7 @@ def make_submission(submitted_file: BinaryIO, user_state, anonymous_state):
|
|
| 34 |
with path_obj.open("rb") as f_in:
|
| 35 |
file_content = f_in.read().decode("utf-8")
|
| 36 |
|
| 37 |
-
validate_csv_file(file_content)
|
| 38 |
|
| 39 |
# write to dataset
|
| 40 |
filename = f"{submission_id}.json"
|
|
|
|
| 11 |
from validation import validate_csv_file
|
| 12 |
|
| 13 |
|
| 14 |
+
def make_submission(
|
| 15 |
+
submitted_file: BinaryIO,
|
| 16 |
+
user_state,
|
| 17 |
+
anonymous_state,
|
| 18 |
+
submission_type: str = "GDPa1",
|
| 19 |
+
):
|
| 20 |
if user_state is None:
|
| 21 |
raise gr.Error("You must submit your username to submit a file.")
|
| 22 |
|
|
|
|
| 39 |
with path_obj.open("rb") as f_in:
|
| 40 |
file_content = f_in.read().decode("utf-8")
|
| 41 |
|
| 42 |
+
validate_csv_file(file_content, submission_type)
|
| 43 |
|
| 44 |
# write to dataset
|
| 45 |
filename = f"{submission_id}.json"
|
validation.py
CHANGED
|
@@ -4,8 +4,10 @@ import gradio as gr
|
|
| 4 |
from constants import (
|
| 5 |
REQUIRED_COLUMNS,
|
| 6 |
MINIMAL_NUMBER_OF_ROWS,
|
| 7 |
-
ANTIBODY_NAMES,
|
| 8 |
ASSAY_LIST,
|
|
|
|
|
|
|
|
|
|
| 9 |
)
|
| 10 |
|
| 11 |
|
|
@@ -46,7 +48,90 @@ def validate_csv_can_be_read(file_content: str) -> pd.DataFrame:
|
|
| 46 |
raise gr.Error(f"β Unexpected error reading CSV file: {str(e)}")
|
| 47 |
|
| 48 |
|
| 49 |
-
def
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 50 |
"""
|
| 51 |
Validate the DataFrame content and structure.
|
| 52 |
|
|
@@ -54,18 +139,23 @@ def validate_dataframe(df: pd.DataFrame) -> None:
|
|
| 54 |
----------
|
| 55 |
df: pd.DataFrame
|
| 56 |
The DataFrame to validate.
|
|
|
|
|
|
|
| 57 |
|
| 58 |
Raises
|
| 59 |
------
|
| 60 |
gr.Error: If validation fails
|
| 61 |
"""
|
|
|
|
|
|
|
|
|
|
| 62 |
# Required columns should be present
|
| 63 |
missing_columns = set(REQUIRED_COLUMNS) - set(df.columns)
|
| 64 |
if missing_columns:
|
| 65 |
raise gr.Error(f"β Missing required columns: {', '.join(missing_columns)}")
|
| 66 |
|
| 67 |
# Should include at least 1 assay column
|
| 68 |
-
assay_columns =
|
| 69 |
if len(assay_columns) < 1:
|
| 70 |
raise gr.Error(
|
| 71 |
"β CSV should include at least one of the following assay columns: "
|
|
@@ -96,14 +186,21 @@ def validate_dataframe(df: pd.DataFrame) -> None:
|
|
| 96 |
)
|
| 97 |
|
| 98 |
# All antibody names should be recognizable
|
| 99 |
-
unrecognized_antibodies = set(df["antibody_name"]) - set(
|
|
|
|
|
|
|
| 100 |
if unrecognized_antibodies:
|
| 101 |
raise gr.Error(
|
| 102 |
f"β Found unrecognized antibody names: {', '.join(unrecognized_antibodies)}"
|
| 103 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 104 |
|
| 105 |
|
| 106 |
-
def validate_csv_file(file_content: str) -> None:
|
| 107 |
"""
|
| 108 |
Validate the uploaded CSV file.
|
| 109 |
|
|
@@ -111,10 +208,12 @@ def validate_csv_file(file_content: str) -> None:
|
|
| 111 |
----------
|
| 112 |
file_content: str
|
| 113 |
The content of the uploaded CSV file.
|
|
|
|
|
|
|
| 114 |
|
| 115 |
Raises
|
| 116 |
------
|
| 117 |
gr.Error: If validation fails
|
| 118 |
"""
|
| 119 |
df = validate_csv_can_be_read(file_content)
|
| 120 |
-
validate_dataframe(df)
|
|
|
|
| 4 |
from constants import (
|
| 5 |
REQUIRED_COLUMNS,
|
| 6 |
MINIMAL_NUMBER_OF_ROWS,
|
|
|
|
| 7 |
ASSAY_LIST,
|
| 8 |
+
CV_COLUMN,
|
| 9 |
+
EXAMPLE_FILE_DICT,
|
| 10 |
+
ANTIBODY_NAMES_DICT,
|
| 11 |
)
|
| 12 |
|
| 13 |
|
|
|
|
| 48 |
raise gr.Error(f"β Unexpected error reading CSV file: {str(e)}")
|
| 49 |
|
| 50 |
|
| 51 |
+
def validate_cv_submission(df: pd.DataFrame, submission_type: str = "GDPa1_CV") -> None:
|
| 52 |
+
"""Validate cross-validation submission"""
|
| 53 |
+
# Must have CV_COLUMN for CV submissions
|
| 54 |
+
if CV_COLUMN not in df.columns:
|
| 55 |
+
raise gr.Error(f"β CV submissions must include a '{CV_COLUMN}' column")
|
| 56 |
+
|
| 57 |
+
# Load canonical fold assignments
|
| 58 |
+
expected_cv_df = pd.read_csv(EXAMPLE_FILE_DICT[submission_type])[
|
| 59 |
+
["antibody_name", CV_COLUMN]
|
| 60 |
+
]
|
| 61 |
+
antibody_check = expected_cv_df.merge(
|
| 62 |
+
df[["antibody_name", CV_COLUMN]],
|
| 63 |
+
on="antibody_name",
|
| 64 |
+
how="left",
|
| 65 |
+
suffixes=("_expected", "_submitted"),
|
| 66 |
+
)
|
| 67 |
+
# All antibodies should be present if using CV
|
| 68 |
+
missing_antibodies_mask = antibody_check[f"{CV_COLUMN}_submitted"].isna()
|
| 69 |
+
n_missing_antibodies = missing_antibodies_mask.sum()
|
| 70 |
+
if n_missing_antibodies > 0:
|
| 71 |
+
missing_antibodies = (
|
| 72 |
+
antibody_check[missing_antibodies_mask]["antibody_name"].head(5).tolist()
|
| 73 |
+
)
|
| 74 |
+
raise gr.Error(
|
| 75 |
+
f"β Missing predictions for {n_missing_antibodies} antibodies. Examples: {', '.join(missing_antibodies)}"
|
| 76 |
+
)
|
| 77 |
+
# CV fold assignments should match
|
| 78 |
+
fold_mismatches = antibody_check[
|
| 79 |
+
antibody_check[f"{CV_COLUMN}_expected"]
|
| 80 |
+
!= antibody_check[f"{CV_COLUMN}_submitted"]
|
| 81 |
+
]
|
| 82 |
+
if len(fold_mismatches) > 0:
|
| 83 |
+
examples = []
|
| 84 |
+
for _, row in fold_mismatches.head(3).iterrows():
|
| 85 |
+
examples.append(
|
| 86 |
+
f"{row['antibody_name']} (expected fold {row[f'{CV_COLUMN}_expected']}, got {row[f'{CV_COLUMN}_submitted']})"
|
| 87 |
+
)
|
| 88 |
+
raise gr.Error(
|
| 89 |
+
f"β Fold assignments don't match canonical CV folds: {'; '.join(examples)}"
|
| 90 |
+
)
|
| 91 |
+
|
| 92 |
+
# Merge on both columns for assay validation
|
| 93 |
+
merged_cv_df = expected_cv_df.merge(df, on=["antibody_name", CV_COLUMN], how="left")
|
| 94 |
+
|
| 95 |
+
# Check for missing assay predictions
|
| 96 |
+
assay_columns = get_assay_columns(merged_cv_df)
|
| 97 |
+
for assay_column in assay_columns:
|
| 98 |
+
missing_antibodies = merged_cv_df[merged_cv_df[assay_column].isna()][
|
| 99 |
+
"antibody_name"
|
| 100 |
+
].unique()
|
| 101 |
+
if len(missing_antibodies) > 0:
|
| 102 |
+
raise gr.Error(
|
| 103 |
+
f"β Missing {assay_column} predictions for {len(missing_antibodies)} antibodies: {', '.join(missing_antibodies[:5])}"
|
| 104 |
+
)
|
| 105 |
+
|
| 106 |
+
# Step 5: Check that submission length matches expected
|
| 107 |
+
if len(merged_cv_df) != len(expected_cv_df):
|
| 108 |
+
raise gr.Error(
|
| 109 |
+
f"β Expected {len(expected_cv_df)} rows, got {len(merged_cv_df)}"
|
| 110 |
+
)
|
| 111 |
+
|
| 112 |
+
|
| 113 |
+
def validate_full_dataset_submission(df: pd.DataFrame) -> None:
|
| 114 |
+
"""Validate full dataset submission"""
|
| 115 |
+
if CV_COLUMN in df.columns:
|
| 116 |
+
raise gr.Error(
|
| 117 |
+
f"β Your submission contains a '{CV_COLUMN}' column. "
|
| 118 |
+
"Please select 'Cross-Validation Predictions' if you want to submit CV results."
|
| 119 |
+
)
|
| 120 |
+
|
| 121 |
+
# All names should be unique (duplicates check from original validation)
|
| 122 |
+
n_duplicates = df["antibody_name"].duplicated().sum()
|
| 123 |
+
if n_duplicates > 0:
|
| 124 |
+
raise gr.Error(
|
| 125 |
+
f"β Standard submissions should have only one prediction per antibody. Found {n_duplicates} duplicates."
|
| 126 |
+
)
|
| 127 |
+
|
| 128 |
+
|
| 129 |
+
def get_assay_columns(df: pd.DataFrame) -> list[str]:
|
| 130 |
+
"""Get all assay columns from the DataFrame"""
|
| 131 |
+
return [col for col in df.columns if col in ASSAY_LIST]
|
| 132 |
+
|
| 133 |
+
|
| 134 |
+
def validate_dataframe(df: pd.DataFrame, submission_type: str = "GDPa1") -> None:
|
| 135 |
"""
|
| 136 |
Validate the DataFrame content and structure.
|
| 137 |
|
|
|
|
| 139 |
----------
|
| 140 |
df: pd.DataFrame
|
| 141 |
The DataFrame to validate.
|
| 142 |
+
submission_type: str
|
| 143 |
+
Type of submission: "GDPa1" or "GDPa1_CV"
|
| 144 |
|
| 145 |
Raises
|
| 146 |
------
|
| 147 |
gr.Error: If validation fails
|
| 148 |
"""
|
| 149 |
+
if submission_type not in EXAMPLE_FILE_DICT.keys():
|
| 150 |
+
raise ValueError(f"Invalid submission type: {submission_type}")
|
| 151 |
+
|
| 152 |
# Required columns should be present
|
| 153 |
missing_columns = set(REQUIRED_COLUMNS) - set(df.columns)
|
| 154 |
if missing_columns:
|
| 155 |
raise gr.Error(f"β Missing required columns: {', '.join(missing_columns)}")
|
| 156 |
|
| 157 |
# Should include at least 1 assay column
|
| 158 |
+
assay_columns = get_assay_columns(df)
|
| 159 |
if len(assay_columns) < 1:
|
| 160 |
raise gr.Error(
|
| 161 |
"β CSV should include at least one of the following assay columns: "
|
|
|
|
| 186 |
)
|
| 187 |
|
| 188 |
# All antibody names should be recognizable
|
| 189 |
+
unrecognized_antibodies = set(df["antibody_name"]) - set(
|
| 190 |
+
ANTIBODY_NAMES_DICT[submission_type]
|
| 191 |
+
)
|
| 192 |
if unrecognized_antibodies:
|
| 193 |
raise gr.Error(
|
| 194 |
f"β Found unrecognized antibody names: {', '.join(unrecognized_antibodies)}"
|
| 195 |
)
|
| 196 |
+
# Submission-type specific validation
|
| 197 |
+
if submission_type.endswith("_CV"):
|
| 198 |
+
validate_cv_submission(df, submission_type)
|
| 199 |
+
else: # full_dataset
|
| 200 |
+
validate_full_dataset_submission(df)
|
| 201 |
|
| 202 |
|
| 203 |
+
def validate_csv_file(file_content: str, submission_type: str = "GDPa1") -> None:
|
| 204 |
"""
|
| 205 |
Validate the uploaded CSV file.
|
| 206 |
|
|
|
|
| 208 |
----------
|
| 209 |
file_content: str
|
| 210 |
The content of the uploaded CSV file.
|
| 211 |
+
submission_type: str
|
| 212 |
+
Type of submission: "standard" or "cv"
|
| 213 |
|
| 214 |
Raises
|
| 215 |
------
|
| 216 |
gr.Error: If validation fails
|
| 217 |
"""
|
| 218 |
df = validate_csv_can_be_read(file_content)
|
| 219 |
+
validate_dataframe(df, submission_type)
|