ESM — Protein Language Models

Overview

ESM (Evolutionary Scale Modeling) provides pretrained protein language models for generative protein design and representation learning. ESM3 is a multimodal generative model conditioned on sequence, structure, and function simultaneously. ESM C is an efficient embedding model optimized for extracting protein representations for downstream ML tasks.

When to Use

• Generating novel protein sequences conditioned on desired structure or function
• Extracting fixed-length embeddings from protein sequences for classification, clustering, or regression
• Predicting 3D structure from amino acid sequence
• Inverse folding: designing sequences that fold into a target structure
• Annotating proteins with functional keywords (GO terms, EC numbers)
• Comparing protein similarity via embedding distance instead of sequence alignment
• Chain-of-thought protein design: iterative refinement of sequence/structure/function
• For traditional physics-based structure prediction, use AlphaFold instead
• For sequence alignment and homology search, use BLAST/HMMER via BioPython instead

Prerequisites