PLAID is a multimodal generative model that simultaneously produces protein sequences and structures by learning the latent space of protein folding models. It addresses limitations in previous models by generating all-atom structures and incorporating organism-specific constraints, useful for real-world applications like drug design. Using sequence databases for training, which are much larger than structural databases, PLAID showcases the ability to generate diverse and functional proteins. Additionally, the CHEAP method compresses the latent space for more effective modeling.
Table of contents
From structure prediction to real-world drug designGenerating “useful” proteinsTraining using sequence-only training dataHow does it work?Compressing the latent space of protein folding modelsWhat’s next?Further linksSome bonus protein generation fun!AcknowledgementsSort: