Abstract
Machine learning has been increasingly used for protein engineering. However, because the general sequence contexts they capture are not specific to the protein being engineered, the accuracy of existing machine learning algorithms is rather limited. Here, we report ECNet (evolutionary context-integrated neural network), a deep-learning algorithm that exploits evolutionary contexts to predict functional fitness for protein engineering. This algorithm integrates local evolutionary context from homologous sequences that explicitly model residue-residue epistasis for the protein of interest with the global evolutionary context that encodes rich semantic and structural features from the enormous protein sequence universe. As such, it enables accurate mapping from sequence to function and provides generalization from low-order mutants to higher-order mutants. We show that ECNet predicts the sequence-function relationship more accurately as compared to existing machine learning algorithms by using ~50 deep mutational scanning and random mutagenesis datasets. Moreover, we used ECNet to guide the engineering of TEM-1 β-lactamase and identified variants with improved ampicillin resistance with high success rates.
Highlights
Machine learning has been increasingly used for protein engineering
It was found that epistasis interactions, quantified by deep mutational scanning (DMS) of proteins, can be used to infer protein contacts and structures[29,30]
A critical challenge in machine learning-guided protein engineering is the development of a machine learning model that accurately maps protein sequences to functions for unseen variants
Summary
Machine learning has been increasingly used for protein engineering. because the general sequence contexts they capture are not specific to the protein being engineered, the accuracy of existing machine learning algorithms is rather limited. It was found that using the learned representation as the feature input to fine-tune a supervised model improves fitness prediction on multiple protein mutagenesis datasets[18] As these models are trained on massive sequences such as those in UniProt[27] and Pfam[28], the learned representations only capture general context for a wide spectrum of proteins but may not be specific to the protein to be engineered. Lacking this specificity in the representation, the prediction model may not be effective in capturing the underlying mechanism (e.g., epistasis between residues) that determines the fitness of a protein and is not able to effectively prioritize bestperforming variants to assist the directed evolution. ECNet was successfully used to engineer TEM-1 β-lactamase variants with improved resistance to ampicillin
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
