Reviewer Validation
Run ProPepX on the official test dataset using the fine-tuned model weights. This validates reproducibility of the reported test performance.
FAQs & Guide
A quick guide for using ProPepX, preparing valid inputs, interpreting prediction outputs, and reproducing the test dataset validation reported in the ProPepX study.
User Guide
ProPepX is a deep learning-based web tool for predicting protein–peptide interaction patterns and potential binding residues from protein and peptide sequences.
Select Single Sequence, paste one protein sequence and one peptide sequence, choose the embedding, mode, and dataset, then click Run ProPepX Prediction.
Yes. Use the Multiple Sequence tab and provide protein and peptide sequences in FASTA format. Each sequence should contain a valid FASTA header and sequence body.
You can paste plain amino acid sequences or FASTA format. FASTA headers are ignored automatically. Only standard amino acid letters should be used.
The maximum protein length is 1418 amino acids. The maximum peptide length is 50 amino acids. Longer sequences should be trimmed before submission.
You may upload an AlphaFold-predicted protein–peptide complex structure in .cif or .mmcif format. This is optional and can be used for structure-based visualization or downstream interpretation.
Prediction Settings
Use ProtTrans for standard protein language model-based inference. Other embeddings should be selected only if they are supported by the deployed model weights.
Global mode evaluates the full protein and peptide sequence to identify likely binding-associated residues across the complete input sequence.
The output highlights predicted protein and peptide binding residues, residue-level scores, and visual summaries. Higher residue scores indicate stronger predicted involvement in binding.
Test Dataset Validation
Run Test Dataset Validation allows users to run the official benchmark/test dataset used in the ProPepX study. This helps verify that the deployed model reproduces results consistent with the base paper.
This feature supports transparency, reproducibility, and deployment verification. It confirms that the correct model weights, preprocessing steps, inference pipeline, and deployment environment are working properly.
The validation metrics should closely match those reported in the ProPepX paper or benchmark table. Small differences may occur due to hardware, software libraries, or floating-point precision, but the overall performance pattern should remain consistent.
Troubleshooting & Use
The first request may take longer because model files and dependencies are initialized. Long protein sequences and batch predictions may also require more processing time.
Common causes include empty sequences, invalid amino acid characters, unsupported file formats, sequences exceeding the allowed length, or mismatched batch FASTA entries.