Apple Unveils SimpleDesign: AI Model for End‑to‑End Protein Design

Apple researchers have introduced a new AI system called SimpleDesign, which creates amino‑acid sequences and three‑dimensional protein structures in a single step, eliminating the multi‑stage pipelines used by most existing tools.

How the model works

SimpleDesign was trained on more than two million paired protein sequences and structures, adding varying levels of noise to teach the network to predict structure from sequence, generate a sequence for a target structure, and co‑design both simultaneously. This end‑to‑end approach lets the model learn directly from raw data without first converting it into separate representations, streamlining training.

Test results

Computational benchmarks show SimpleDesign delivering competitive or superior performance compared with rival models in protein design and structure generation. However, the system has not yet been validated on real‑world proteins in laboratory settings to confirm safety and functionality.

SimpleDesign marks a significant advance in Apple’s life‑science research agenda, building on its earlier SimpleFold model that focused solely on structure prediction from sequence.