LevSeq: Rapid\nGeneration of Sequence-Function Data\nfor Directed Evolution and Machine Learning

Sequence-function data provides valuable information\nabout the\nprotein functional landscape but is rarely obtained during directed\nevolution campaigns. Here, we present Long-read every variant Sequencing\n(LevSeq), a pipeline that combines a dual barcoding strategy with\nnanopore sequencing to rapidly generate sequence-function data for\nentire protein-coding genes. LevSeq integrates into existing protein\nengineering workflows and comes with open-source software for data\nanalysis and visualization. The pipeline facilitates data-driven protein\nengineering by consolidating sequence-function data to inform directed\nevolution and provide the requisite data for machine learning-guided\nprotein engineering (MLPE). LevSeq enables quality control of mutagenesis\nlibraries prior to screening, which reduces time and resource costs.\nSimulation studies demonstrate LevSeq’s ability to accurately\ndetect variants under various experimental conditions. Finally, we\nshow LevSeq’s utility in engineering protoglobins for new-to-nature\nchemistry. Widespread adoption of LevSeq and sharing of the data will\nenhance our understanding of protein sequence-function landscapes\nand empower data-driven directed evolution.