MuthuS97/PIPES-M

[] (https://colab.research.google.com/drive/1OoX9zDwdSD88UGXxlFctnq_UPcnkkWdp?usp=sharing)

PIPES-M, a deep learning-based binary classifier designed to predict protease inhibitor (PI) activity from primary protein sequences.

PIPES-M is a fine-tuned sequence classification model built on the ESM-2 protein language model (EsmForSequenceClassification):

• Base model: facebook/esm2_t30_150M_UR50D (150 million parameters, 30 layers)
• Pre-trained on UniRef50 via masked language modeling

Fine-tuning was performed on a high-quality curated dataset comprising:

• Positive examples: known protease inhibitors (<250 AA) from the MEROPS and Uniprot database
• Negative examples: non-inhibitors selected from UniProt using sequence similarity and Pfam domain analysis

Training used sequence-only input, requiring no structural data. The classification head leverages evolutionary and physicochemical features encoded by ESM-2.

Maximum sequence length is fixed at 250 residues; longer sequences are truncated from the N-terminus, appropriate for the typical size range of small secreted inhibitors.

---

license: creativeml-openrail-m