Inferring Protein Sequence-Function Relationships with Large-Scale Positive-Unlabeled Learning

Abstract

Machine learning can infer how protein sequence maps to function without requiring a detailed understanding of the underlying physical or biological mechanisms. It's challenging to apply existing supervised learning frameworks to large-scale experimental data generated by deep mutational scanning (DMS) and related methods. DMS data often contain high dimensional and correlated sequence variables, experimental sampling error and bias, and the presence of missing data. Importantly, most DMS data do not contain examples of negative sequences, making it challenging to directly estimate how sequence a↵ects function. Here, we develop a positive-unlabeled (PU) learning framework to infer sequence-function relationships from largescale DMS data. Our PU learning method displays excellent predictive performance across ten large-scale sequence-function data sets, representing proteins of di↵erent folds, functions, and library types. The estimated parameters pinpoint key residues that dictate protein structure and function. Finally, we apply our statistical sequence-function model to design highly stabilized enzymes.

Design of synthetic human gut microbiome assembly and butyrate production

“…In metabolic and protein engineering, computational models informed by high-throughput experimental measurements have been used to design optimized protein sequences with high protein stability 28 or identify promoter combinations driving enzymes in a metabolic pathway to maximize valuable molecule production 29,30 . In the microbiome field, empirical and top-down approaches are frequently used to design microbial communities as opposed to data-driven approaches 14 .…”

Combining evolutionary and assay-labelled data for protein fitness prediction

“…When the space becomes too large, a layer of optimization must be added because a comprehensive screen is no longer possible [16][17][18][19] . In both cases, a key component of the ML-guided protein engineering approach is a reliance on an accurate MLbased fitness model-one that predicts protein property from protein sequence [20][21][22][23][24][25] .…”

Microfluidic deep mutational scanning of the human executioner caspases reveals differences in structure and regulation

“…The reads from the Illumina FASTQ files were mapped to the caspase reference gene using Bowtie2 22 , and translated to amino acid sequences. The fitness effect of each observed amino acid substitution was estimated using a positive-unlabeled learning framework that compares sequences from the presorted population with the sorted population 23,24 .…”