Abstract
Statistical analysis of evolutionary-related protein sequences provides information about their structure, function, and history. We show that Restricted Boltzmann Machines (RBM), designed to learn complex high-dimensional data and their statistical features, can efficiently model protein families from sequence information. We here apply RBM to 20 protein families, and present detailed results for two short protein domains (Kunitz and WW), one long chaperone protein (Hsp70), and synthetic lattice proteins for benchmarking. The features inferred by the RBM are biologically interpretable: they are related to structure (residue-residue tertiary contacts, extended secondary motifs (α-helixes and β-sheets) and intrinsically disordered regions), to function (activity and ligand specificity), or to phylogenetic identity. In addition, we use RBM to design new protein sequences with putative properties by composing and 'turning up' or 'turning down' the different modes at will. Our work therefore shows that RBM are versatile and practical tools that can be used to unveil and exploit the genotype-phenotype relationship for protein families.
Highlights
In recent years, the sequencing of many organisms’ genomes has led to the collection of a huge number of protein sequences, which are catalogued in databases such as UniProt or PFAM Finn et al, 2014)
Our study shows that Restricted Boltzmann Machines (RBM) are able to capture: (1) structure-related features, be they local, extended such as secondary structure motifs (a-helix and b-sheet)) or characteristic of intrinsically disordered regions; (2) functional features, that is groups of amino acids controling specificity or activity; and (3) phylogenetic features, related to sub-families sharing evolutionary determinants
Protein sequences v 1⁄4 ðv1; v2; :::; vN Þ are displayed on the Visible layer, and representations h 1⁄4 ðh1; h2; :::; hMÞ on the Hidden layer
Summary
The sequencing of many organisms’ genomes has led to the collection of a huge number of protein sequences, which are catalogued in databases such as UniProt or PFAM Finn et al, 2014). Sequences that share a common ancestral origin, defining a family (Figure 1A), are likely to code for proteins with similar functions and structures, providing a unique window into the relationship between genotype (sequence content) and phenotype (biological features). In this context, various approaches have been introduced to infer protein properties from sequence data statistics, in particular amino-acid conservation and coevolution (correlation) (Teppa et al, 2012; de Juan et al, 2013). Other methods rely on phylogeny Rojas et al, 2012, entropy (variability in amino-acid content) Reva et al, 2007, or a hybrid combination of both Mihalek et al, 2004; Ashkenazy et al, 2016
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
