esm

Protein sequence design uses models like ESM3 to complete masked sequences and generate novel proteins. You provide a partially masked sequence, and the model predicts and fills the missing residues to return a fully formed sequence.

Structure prediction and inverse folding map 3D coordinates from sequences or design sequences folding to target structures. ESM3 supports multimodal generation across sequence, structure, and function tracks for engineering workflows.

Protein embeddings are efficient representations for similarity search, clustering, and downstream ML feature extraction. ESM C models generate these dense vectors to enable embedding-driven similarity or clustering tasks.

Variant library construction for protein engineering involves generating and completing engineered protein sequences. ESM3 enables multimodal generation across sequence, structure, and function to support structure-conditioned redesign.

Forge API access provides scalable inference for running ESM3 and ESM C models. It is optional for generating proteins and embeddings but required when you need to scale generation configuration parameters and multimodal track processing.

Functional annotation and multimodal protein generation are supported by ESM3. The model interfaces process sequence, structure, and function tracks simultaneously to generate, complete, and engineer protein sequences.