Camp: Combinatorial Engineering of Proteins

Abstract

Protein recombination has long been a key method in protein engineering to diver- sify and optimize sequences. We enhance and evolve this approach by using a pro- tein language model, where we found that when attention in the language model is represented as a spline, abrupt transitions in the spline identify optimal crossover sites for recombination. As we show, these sites also correlate with transitions between various secondary structure elements in the corresponding protein struc- ture. We use these sites to guide recombination of sequence blocks from diverse sources using MCMC sampling. Language models also enable generation of novel recombinant blocks beyond traditional MSAs, increasing diversity, while a direct preference optimization algorithm is used to fine-tune these blocks for reduced immunogenicity. This method integrates modern deep learning architectures with traditional protein engineering techniques to improve success rate of the libraries designed for wetlab verification.