- PepSeqPred
-- PepSeqPred predicts epitope masks for protein sequences and supports full - developer workflows for preprocessing, training, evaluation, and HPC execution. -
-- PyPI - GitHub - Developer README -
-pip install pepseqpredfrom pepseqpred import load_pretrained_predictor
-
-predictor = load_pretrained_predictor(
- model_id="default",
- device="auto"
-)
-result = predictor.predict_sequence(
- "ACDEFGHIKLMNPQRSTVWY",
- header="example_protein"
-)
-print(result.binary_mask)from pepseqpred import load_predictor
-
-predictor = load_predictor(
- model_artifact="path/to/ensemble_manifest.json",
- device="auto"
-)
-result = predictor.predict_sequence(
- "ACDEFGHIKLMNPQRSTVWY"
-)
-print(result.binary_mask)- PepSeqPred is built for reproducible residue-level prediction under strong class imbalance. - Training and validation are documented as protocol-first; numeric scorecards are based on seeded evaluation - snapshots. -
-- Headline metrics below are means across seeded evaluation sets and paired deltas - (flagship2 - flagship1). PR metrics are primary because the positive class is rare. -
- -| Metric | -Flagship1 mean | -Flagship2 mean | -Delta (F2 - F1) | -95% bootstrap CI | -Sign-test p | -
|---|---|---|---|---|---|
| Best-fold PR-AUC (primary) | -- | - | - | - - to - - | -- |
| Best-fold PR lift | -- | - | - | - - to - - | -- |
| Best-fold ROC AUC | -- | - | - | - - to - - | -- |
| Overall PR-AUC | -- | - | - | - - to - - | -- |
- 
- 
- 
- - PepSeqPred is fully open-source. All pipeline code, scripts, and documentation are available on GitHub. -
-PepSeqPred
++ PepSeqPred predicts epitope masks for protein sequences and supports full developer workflows for + preprocessing, training, evaluation, and HPC execution. +
++ + PyPI + + + GitHub + + + Developer README + +
++ Model Comparison + PhaseB Validation + External Cocci +
+
+ pip install pepseqpred
+
+
+ {`from pepseqpred import load_pretrained_predictor
+
+predictor = load_pretrained_predictor(
+ model_id="default",
+ device="auto"
+)
+result = predictor.predict_sequence(
+ "ACDEFGHIKLMNPQRSTVWY",
+ header="example_protein"
+)
+print(result.binary_mask)`}
+
+
+ {`from pepseqpred import load_predictor
+
+predictor = load_predictor(
+ model_artifact="path/to/ensemble_manifest.json",
+ device="auto"
+)
+result = predictor.predict_sequence(
+ "ACDEFGHIKLMNPQRSTVWY"
+)
+print(result.binary_mask)`}
+
+ + PepSeqPred is built for reproducible residue-level prediction under strong class imbalance. Training and + validation are documented as protocol-first; numeric scorecards are based on seeded evaluation snapshots. +
++ PepSeqPred is fully open-source. Pipeline code, scripts, and curated website evidence sources are all in the + same repository for transparent review. +
+ +ffnn_ens_2.4_v100_28234651).
+ + Headline metrics below are means across seeded evaluation sets and paired deltas (flagship2 - flagship1). + PR metrics are primary because the positive class is rare. +
+ +| Metric | +Flagship1 mean | +Flagship2 mean | +Delta (F2 - F1) | +95% bootstrap CI | +Sign-test p | +
|---|---|---|---|---|---|
| + {metric.label} + {metric.primary ? " *" : ""} + | +{formatFixed(metric.flagship1Mean, metric.id.includes("lift") ? 3 : 6)} | +{formatFixed(metric.flagship2Mean, metric.id.includes("lift") ? 3 : 6)} | ++ {formatSigned(metric.delta, metric.id.includes("lift") ? 3 : 6)} + | ++ {formatSigned(metric.ciLow, metric.id.includes("lift") ? 3 : 6)} to{" "} + {formatSigned(metric.ciHigh, metric.id.includes("lift") ? 3 : 6)} + | +{formatSci(metric.signTestPValue, 2)} | +
+ Snapshot date: {modelComparisonData.snapshotDate}. Generated: {modelComparisonData.generatedAtUtc}. +
+{externalCocciData.description}
+ +| Metric | +n pairs | +Mean delta | +Median delta | +95% bootstrap CI | +Sign-test p | +
|---|---|---|---|---|---|
| {metricLabels[row.metric] ?? row.metric} | +{row.nPairs} | +{formatSigned(row.meanDiffBMinusA, 6)} | +{formatSigned(row.medianDiffBMinusA, 6)} | ++ {formatSigned(row.ciLow, 6)} to {formatSigned(row.ciHigh, 6)} + | +{formatSci(row.signTestPValue, 2)} | +
+ This page intentionally uses the most recent PhaseB validation run only:
+ {phaseBValidationData.runId} (last modified {phaseBValidationData.lastModifiedDate}).
+ Multi-dataset PhaseC validation is intentionally excluded until stability is established.
+
Train mode
+{phaseBValidationData.trainMode}
+Split type
+{phaseBValidationData.splitType}
+Runs / Folds
++ {phaseBValidationData.nRuns} / {phaseBValidationData.nFolds} +
+PR baseline mean
+{formatPercent(phaseBValidationData.prBaselineMean, 3)}
+| Metric | +Mean | +Std | +Min | +Max | +
|---|---|---|---|---|
| PR-AUC | +{formatFixed(summaries.prAuc.mean, 6)} | +{formatFixed(summaries.prAuc.std, 6)} | +{formatFixed(summaries.prAuc.min, 6)} | +{formatFixed(summaries.prAuc.max, 6)} | +
| F1 | +{formatFixed(summaries.f1.mean, 6)} | +{formatFixed(summaries.f1.std, 6)} | +{formatFixed(summaries.f1.min, 6)} | +{formatFixed(summaries.f1.max, 6)} | +
| MCC | +{formatFixed(summaries.mcc.mean, 6)} | +{formatFixed(summaries.mcc.std, 6)} | +{formatFixed(summaries.mcc.min, 6)} | +{formatFixed(summaries.mcc.max, 6)} | +
| ROC AUC | +{formatFixed(summaries.auc.mean, 6)} | +{formatFixed(summaries.auc.std, 6)} | +{formatFixed(summaries.auc.min, 6)} | +{formatFixed(summaries.auc.max, 6)} | +
| AUC10 | +{formatFixed(summaries.auc10.mean, 6)} | +{formatFixed(summaries.auc10.std, 6)} | +{formatFixed(summaries.auc10.min, 6)} | +{formatFixed(summaries.auc10.max, 6)} | +
| Balanced accuracy | +{formatFixed(summaries.balancedAcc.mean, 6)} | +{formatFixed(summaries.balancedAcc.std, 6)} | +{formatFixed(summaries.balancedAcc.min, 6)} | +{formatFixed(summaries.balancedAcc.max, 6)} | +
| Fold | +Best epoch | +PR-AUC | +F1 | +MCC | +AUC | +AUC10 | +Threshold | +
|---|---|---|---|---|---|---|---|
| {row.fold} | +{row.bestEpoch} | +{formatFixed(row.prAuc, 6)} | +{formatFixed(row.f1, 6)} | +{formatFixed(row.mcc, 6)} | +{formatFixed(row.auc, 6)} | +{formatFixed(row.auc10, 6)} | +{formatFixed(row.threshold, 6)} | +
![PepSeqPred logo]({assetPath("/assets/brand/PepSeqPred_logo_white.png")}){kind=logo}
+