In brief: A plain-language reference for newcomers to ESM3 and computational protein science.
Model fundamentals
Protein: a biological polymer made of amino acids. Amino-acid sequence: the ordered chain of residues. Residue: an amino-acid unit within a protein. Transformer: a neural-network architecture used by many language models. Token: a discrete unit processed by a model. Embedding: a learned numeric representation. Checkpoint: a saved set of model weights. Inference: using trained weights to produce an output. Fine-tuning: further training on a specific task or dataset.
Modeling modalities
Sequence track: a representation of amino-acid identity. Structure track: a representation related to protein geometry. Function track: model information associated with functional annotations. Masked generation: filling unspecified positions iteratively. Conditioning: specifying constraints before generation. MSA: multiple sequence alignment. PDB: a common data resource and coordinate file format for macromolecular structures. Fold: an overall structural organization of a protein.
Evaluation concepts
Baseline: a reference method for comparison. Data leakage: test information that contaminates the model-development process. Homology: similarity reflecting shared evolutionary origin. Distribution shift: differences between development and real-world test data. Confidence metric: a model-estimated signal of uncertainty. Experimental validation: laboratory evidence supporting a claim. Reproducibility: ability to repeat the method with documented inputs. Biosecurity: efforts to prevent biological capabilities from being misused.
Next steps
Use the terms with the ESM3 introduction and model comparison guides. Many concepts have specialized meanings across disciplines; consult primary papers for precise mathematical definitions.
Primary sources and further reading
Editorial review: October 2026. This guide is educational. Model availability and APIs may change; confirm exact versions in the linked documentation. Generated structures and sequences require independent scientific validation.