Protein Structure Analysis Project

Project Overview

This repository contains various implementations and experiments for protein structure analysis using different deep learning approaches.

Repository Structure

• GAT-experiments-23Apr: Final implementation of ESM-2 + GAT with/without KAN
• GAT-GCN-improved-20APr: Final implementation of ESM-2 + GCN with/without KAN
• protein_folding: Code for protein structure prediction using ESMFold
• esm-3-emebddings-generation: ESM-3 embeddings generation via API calls
• ESM-3_and_GCN: Experimental code combining ESM-3 with GCN

Important Notice ⚠️

Before running any code in this repository:

1. Data Paths:

   • Modify all data file paths according to your local setup
   • Data we used for training and testing: data/development_set/full_grouped_train_binding_sites_df.csv (Training data), data/development_set/full_grouped_test_binding_sites_df.csv (TestSet300), data/independent_set/grouped_test_46_new_binding_sites.csv (TestSetNew46)
   • Data file path for storing protein's predicted structures using ESMFold: data/esmFold_pdb_files
   • Check and update paths in configuration files to align with your local file path.
   • Ensure data files are in the correct locations

2. Execution Order:

   • Generate embeddings first using the ESM-3 code
   • Run model training scripts afterward
   • Follow each directory's specific README for detailed instructions

Prerequisites

• Python 3.9+
• Required packages (refer to requirements.txt)
• Access to ESM-3 API (for embeddings generation)

Contact

For questions or issues, please open a GitHub issue in this repository.