SimpleDesign: A Joint Model for Protein Sequence and Structure Codesign
Apple Machine Learning Research
SimpleDesign introduces an end-to-end protein design model that jointly generates amino-acid sequences and 3D structures instead of using a two-stage latent-space pipeline. SimpleDesign is trained on over 2M sequence–structure pairs. As a result, it uses a single-stage objective with sequence cross-entropy and structure regression while employing Transformer-based multimodal backbones, achieving competitive performance on co-design and unconditional generation benchmarks.
Why it matters
Proteins are fundamental to biological processes, with their function determined by the complex interplay between the amino acid sequence and the three-dimensional structure. Developing generative models capable of understanding this intrinsically multi-modal relationship is crucial for fields like drug discovery and protein engineering. Existing models often rely on a multi-stage training process where autoencoders that tokenize data into latent representations are trained in a first stage. Secondly, a generative model is trained on the latent representation of the autoencoder(s), i.e…
