Protein Engineering with Lightweight Graph Denoising Neural Networks

  • One way of engineering proteins is by introducing mutations

  • This study uses a graph neural network to model the distribution of amino acid sequences that are more likely to pass natural selection. This is used as a guidance for scoring proteins towards properties that we want via mutagenesis.

  • They do wet-lab validation of their designs; more than 40% of designed single-site mutants outperform their counterparts, and they can design multiple mutations in a single round for improved properties.