Learning the molecular grammar of protein condensates from sequence determinants and embeddings

Abstract

Intracellular phase separation of proteins into biomolecular condensates is increasingly recognized as a process with a key role in cellular compartmentalization and regulation. Different hypotheses about the parameters that determine the tendency of proteins to form condensates have been proposed, with some of them probed experimentally through the use of constructs generated by sequence alterations. To broaden the scope of these observations, we established an in silico strategy for understanding on a global level the associations between protein sequence and phase behavior and further constructed machine-learning models for predicting protein liquid–liquid phase separation (LLPS). Our analysis highlighted that LLPS-prone proteins are more disordered, less hydrophobic, and of lower Shannon entropy than sequences in the Protein Data Bank or the Swiss-Prot database and that they show a fine balance in their relative content of polar and hydrophobic residues. To further learn in a hypothesis-free manner the sequence features underpinning LLPS, we trained a neural network-based language model and found that a classifier constructed on such embeddings learned the underlying principles of phase behavior at a comparable accuracy to a classifier that used knowledge-based features. By combining knowledge-based features with unsupervised embeddings, we generated an integrated model that distinguished LLPS-prone sequences both from structured proteins and from unstructured proteins with a lower LLPS propensity and further identified such sequences from the human proteome at a high accuracy. These results provide a platform rooted in molecular principles for understanding protein phase behavior. The predictor, termed DeePhase, is accessible from https://deephase.ch.cam.ac.uk/.

Beta turn propensity and a model polymer scaling exponent identify intrinsically disordered phase-separating proteins

“…Other sequence-based predictors of protein LLPS, for example, PSCORE ( 10 ) and catGRANULE ( 11 ), similarly identify only a small subset of the human proteome as exhibiting high LLPS potential. Our work also builds on the recent finding that phase-separating IDRs are less hydrophobic by traditional scales than non-phase-separating IDRs, yet more compact ( 105 , 106 ). The compaction appears to occur through other mechanisms than hydrophobicity, including cation–π and charged interactions ( 10 , 80 , 106 ), as well as a high propensity for β-turns ( Fig.…”

“…As the ParSe model does not directly evaluate features proposed by other mechanisms, namely the patterning of either cation–π, π–π, or charged amino acids or nucleic acid binding; the correlation between these metrics points to an evolutionary constraint to include multiple sequence features in regions that promote LLPS ( 10 , 11 , 95 ). More generally, the correlation between predictors that are based on disparate molecular mechanisms will be useful for determining which molecular features are typically combined in LLPS proteins and which LLPS proteins instead rely on unique molecular grammars ( 80 , 95 , 96 , 98 , 100 , 105 , 106 ).…”

Beta turn propensity and a model polymer scaling exponent identify disordered proteins that phase separate

“…As the ParSe model does not directly evaluate for the sequence features proposed by other mechanisms, namely the patterning of either cation-π, π-π or charged amino acids; or nucleic acid binding; the correlation between these metrics points to an evolutionary constraint to include multiple sequence features in regions that promote LLPS (10, 11, 95). More generally, the correlation between predictors that are based on disparate molecular mechanisms will be useful for determining which molecular features are typically combined in LLPS proteins, and which LLPS proteins instead rely on unique molecular grammars (80, 95, 96, 98, 100, 105, 106).…”

The chromatin regulator HMGA1a undergoes phase separation in the nucleus

“…We further examined how some of the key physical parameters that have been proposed to govern phase separation vary across the sequence. 52,53 Proteins that undergo phase separation in vitro and in vivo often contain intrinsically disordered regions that are marked by low sequence complexity. 35,36,50,[54][55][56][57][58][59] These features enable proteins to establish multivalent homotypic and/or heterotypic interactions with their binding partners necessary to drive phase separation.…”