HelixGAN: A bidirectional Generative Adversarial Network with search in latent space for generation under constraints

Abstract:
Protein engineering has become an important field in biomedicine with application in therapeutics, diagnostics and synthetic biology. Due to the complexity of protein structure computational design remains a difficult problem. As helices are an abundant structural feature and play a vital role in determination of the protein structure, full computational design for helices would be an important first step. Here, we apply Wasserstein bi-directional Generative Adversarial Networks to generate the full atom helical structures, and we also introduce a novel Markov chain Monte Carlo searching mechanism with the encoder to design the desired helices matching the target “hotspot” residues structures. Our model completed 70.9% testcases by generating the desired helices within 3 A RMSD compared with target hotspots. We demonstrate that our approach is able to quickly generate structurally plausible solutions, bringing us closer to the final goal of fast computational protein design.

Suggested citation:
Xie, and Kim . (2021). HelixGAN: A bidirectional Generative Adversarial Network with search in latent space for generation under constraints, Machine Learning for Structural Biology (MLSB) Workshop at NeurIPS.