Latent generative landscapes as maps of functional diversity in protein sequence space

Abstract

Variational autoencoders are unsupervised learning models with generative capabilities, when applied to protein data, they classify sequences by phylogeny and generate de novo sequences which preserve statistical properties of protein composition. While previous studies focus on clustering and generative features, here, we evaluate the underlying latent manifold in which sequence information is embedded. To investigate properties of the latent manifold, we utilize direct coupling analysis and a Potts Hamiltonian model to construct a latent generative landscape. We showcase how this landscape captures phylogenetic groupings, functional and fitness properties of several systems including Globins, β-lactamases, ion channels, and transcription factors. We provide support on how the landscape helps us understand the effects of sequence variability observed in experimental data and provides insights on directed and natural protein evolution. We propose that combining generative properties and functional predictive power of variational autoencoders and coevolutionary analysis could be beneficial in applications for protein engineering and design.

Proteome-scale prediction of molecular mechanisms underlying dominant genetic diseases

“…In total, 21 interpretable features were used in the models, including properties derived from protein sequences, structures, networks, and gene mutational constraints (Badonyi and Marsh 2023). Additionally, 20 language model-based embeddings were also included, which are thought to represent protein function in their latent space [41]. As a measure of feature importance, we calculated the loss in AUROC relative to the full model ( Figure 3 ).…”

“…Additionally, 20 language model-based embeddings were also included, which are thought to represent protein function in their latent space [41]. As a measure of feature importance, we calculated the loss in AUROC relative to the full model (Figure 3).…”

Coevolutionary Information Captures Catalytic Functions and Reveals Divergent Roles of Terpene Synthase Interdomain Connections

“…Formally, H can be used to compute the probability of finding a particular sequence in the input protein family. H serves as a proxy for protein fitness − and has been described as a measure of the “typicality” of a protein sequence within its family, with more negative values signifying more family like or typical sequences . Coevolutionary information-based scoring has also been shown to predict specificity between histidine kinases and response regulators, compatibility between DNA recognition and allosteric response modules in LacI-type transcription inhibitors, folding kinetics, and mutational phenotypes in protein–RNA complexes …”

Machine Learning-Guided Protein Engineering

“…219 In addition, by exploring the latent manifold underlying the sequence information, we can uncover dependencies that may not be readily apparent in the raw latent space embeddings. 220 Despite its advantages, MSAs also have some drawbacks. First, it can be difficult to create an MSA that contains enough evolutionarily relevant sequences to establish strong patterns at key amino acid positions.…”

“…Building on these findings, the geometric structure of a latent space was recently used to guide the design of a haloalkane dehalogenase . In addition, by exploring the latent manifold underlying the sequence information, we can uncover dependencies that may not be readily apparent in the raw latent space embeddings …”