Protein Function Prediction (PFP)

This repository provides a framework for protein function prediction using a combination of sequence, structure, and text-based features, leveraging deep learning and protein language models.

Features

• Data Preparation: Scripts to process protein sequences, annotations, and generate training/test datasets.
• Model Training: Deep neural network models for multi-label protein function prediction, including support for various embedding types (e.g., ESM, T5, structure-based, and fused models).
• Ontology Tools: Utilities for handling Gene Ontology (GO) terms and relationships.
• Text Embedding: Generate protein-related text embeddings using transformer models (e.g., BiomedBERT).
• Structure Processing: Tools for building and processing protein structure graphs.

Directory Structure

• prepare_data.py: Prepare datasets from raw sequence and annotation files.
• train_InterLabelGO.py: Main training script for protein function prediction models.
• AP_align_fuse.py: Model for aligning and fusing sequence and text embeddings.
• prot2text.py: Generate text embeddings for protein descriptions.
• structure.py: Utilities for processing protein structures and building graph representations.
• Network/: Contains model architectures and training utilities.
• utils/: Ontology and miscellaneous utility scripts.
• cluster/: Scripts for HPC.