Run protein folding and embeddings models

Generate protein embeddings and search protein sequence databases

Computing per-residue embeddings for downstream ML tasks (docking, binding prediction, active-site detection) when you need a pretrained protein encoder.

• · protein FASTA sequence
• · ESM model variant (ESM1b/ESM2/ESMFold)

• · 768-dim or 1280-dim embeddings per token
• · folded 3D structure (ESMFold)

• · PyTorch
• · fair-esm package
• · optional: CPU ok but GPU 10x faster

Python 3.8+; ESM model weights auto-download on first use (~500MB); CUDA optional

Embeddings frozen at model release (not fine-tunable in this context); sequence length limits (>1024 tokens need chunking); memory-hungry on CPU

• · State-of-the-art protein language model
• · ESMFold included for structure prediction
• · Token-level embeddings for fine-grained analysis
• · ESM2 256M up to 15B parameter variants