feat: added ESM-2 preprocessing pipeline

[jdpearce4]

Resolves #45

Changes

• Added the preprocessing pipeline from internal codebase.
• Added instructions to main README.md
• Created preprocessing README.md
• Replaced wget and gunzip dependencies with python modules

Testing

• tested human ESM-2 embeddings generation with defaults
• tested new species (sheep) generation with defaults

[Copilot]

Pull Request Overview

This PR adds a comprehensive ESM-2 preprocessing pipeline for generating protein embeddings from FASTA files. The implementation replaces shell-based dependencies (wget/gunzip) with pure Python modules and provides a complete workflow for downloading, processing, and converting protein sequences to gene-level embeddings.

Key changes:

• Complete preprocessing pipeline with ESM-2 model integration for protein embedding generation
• Python-based file downloading and decompression replacing shell dependencies
• Support for 20+ species with extensible configuration through JSON manifest

Reviewed Changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
preprocess/requirements.txt	Dependencies for ESM-2 preprocessing pipeline
preprocess/protein_embedding.py	Main script for generating protein embeddings using ESM-2 models
preprocess/get_stable_id_mapping.py	Utility functions for mapping gene/protein/transcript stable IDs
preprocess/fasta_manifest_pep.json	Configuration manifest with Ensembl download URLs for supported species
preprocess/README.md	Comprehensive documentation for the preprocessing pipeline
README.md	Updated main README with protein embedding availability table and preprocessing instructions

Comments suppressed due to low confidence (2)

preprocess/requirements.txt:4

• The biopython dependency is missing a version specification. Consider pinning to a specific version for reproducibility, e.g., 'biopython==1.81'.

biopython

preprocess/requirements.txt:5

• The h5py dependency is missing a version specification. Consider pinning to a specific version for reproducibility, e.g., 'h5py==3.9.0'.

h5py