SymFold uses a dual-path AI to design proteins more accurately from their 3D shapes.

Proteins are made of chains of amino acids, which fold into specific three-dimensional shapes. In protein inverse folding, scientists want to find the exact amino acid sequence that will build a given target shape. Past tools do this in a single line: one model guesses a rough sequence, and a language model tweaks it later. If the first guess is poor, the final result suffers. The researchers created a new system called SymFold. It uses two balanced paths working together. One path uses evolutionary knowledge about sequences, while the other uses structural knowledge about protein shapes. These two paths iteratively guide sequence creation. Tests show SymFold beats previous tools, setting a new top benchmark for inverse folding.

Why it matters

This helps researchers in biotechnology and medicine design better enzymes and drugs by creating proteins that fold into desired shapes more reliably. It improves the tools used for protein engineering, which could speed up development in these fields.

Questions answered