🔬Generating Molecules, Not Just Models

CASP 14 dominance

AlphaFold2 achieved unprecedented accuracy at the 2020 protein structure prediction competition, effectively solving the 50-year challenge of predicting single-chain protein structures when evolutionary data is available.

Structure vs. folding distinction

The breakthrough addressed predicting final static structures but not the dynamic folding process itself, leaving intermediate states and misfolding pathways poorly understood.

Proteins as cellular machinery

Proteins are sequences of 20 amino acid types that fold into functional machines, with their 3D shape determining biological function and disease mechanisms.

Small molecules and nucleic acids

Small molecules feature diverse atomic compositions distinct from proteins, while nucleic acids (DNA/RNA) use 4-base sequences similar to proteins but require different modeling approaches.

Dynamic and complex systems

Models struggle with intrinsically disordered proteins, multi-chain complexes, and conformational switching where proteins change shape based on cellular environment.

The orphan protein problem

Structure prediction fails for proteins lacking evolutionary homologs, as current methods rely heavily on multiple sequence alignments (MSA) to infer spatial constraints.

Co-evolutionary signals

Spatially close amino acids show correlated mutations across species through compensatory changes, providing distance constraints that enable accurate geometric reconstruction.

MSA dependency bottleneck

This reliance on evolutionary history means the 'solved' problem only applies to well-characterized protein families, leaving many therapeutic targets structurally opaque.