Explore the vast range of titles available.

SignificanceLearning biological properties from sequence data is a logical step toward generative and predictive artificial intelligence for biology. Here, we propose scaling a deep contextual language model with unsupervised learning to sequences spanning evolutionary diversity. We find that without prior knowledge, information emerges in the learned representations on fundamental properties of proteins such as secondary structure, contacts, and biological activity. We show the learned representations are useful across benchmarks for remote homology detection, prediction of secondary structure, long-range residue–residue contacts, and mutational effect. Unsupervised representation learning enables state-of-the-art supervised prediction of mutational effect and secondary structure and improves state-of-the-art features for long-range contact prediction. In the field of artificial intelligence, a combination of scale in data and model capacity enabled by unsupervised learning has led to major advances in representation learning and statistical generation. In the life sciences, the anticipated growth of sequencing promises unprecedented data on natural sequence diversity. Protein language modeling at the scale of evolution is a logical step toward predictive and generative artificial intelligence for biology. To this end, we use unsupervised learning to train a deep contextual language model on 86 billion amino acids across 250 million protein sequences spanning evolutionary diversity. The resulting model contains information about biological properties in its representations. The representations are learned from sequence data alone. The learned representation space has a multiscale organization reflecting structure from the level of biochemical properties of amino acids to remote homology of proteins. Information about secondary and tertiary structure is encoded in the representations and can be identified by linear projections. Representation learning produces features that generalize across a range of applications, enabling state-of-the-art supervised prediction of mutational effect and secondary structure and improving state-of-the-art features for long-range contact prediction.

Macromolecular phase separation is thought to be one of the processes that drives the formation of membraneless biomolecular condensates in cells. The dynamics of phase separation are thought to follow the tenets of classical nucleation theory, and, therefore, subsaturated solutions should be devoid of clusters with more than a few molecules. We tested this prediction using in vitro biophysical studies to characterize subsaturated solutions of phase-separating RNA-binding proteins with intrinsically disordered prion-like domains and RNA-binding domains. Surprisingly, and in direct contradiction to expectations from classical nucleation theory, we find that subsaturated solutions are characterized by the presence of heterogeneous distributions of clusters. The distributions of cluster sizes, which are dominated by small species, shift continuously toward larger sizes as protein concentrations increase and approach the saturation concentration. As a result, many of the clusters encompass tens to hundreds of molecules, while less than 1% of the solutions are mesoscale species that are several hundred nanometers in diameter. We find that cluster formation in subsaturated solutions and phase separation in supersaturated solutions are strongly coupled via sequence-encoded interactions. We also find that cluster formation and phase separation can be decoupled using solutes as well as specific sets of mutations. Our findings, which are concordant with predictions for associative polymers, implicate an interplay between networks of sequence-specific and solubility-determining interactions that, respectively, govern cluster formation in subsaturated solutions and the saturation concentrations above which phase separation occurs.

Every heartbeat relies on cyclical interactions between myosin thick and actin thin filaments orchestrated by rising and falling Ca2+ levels. Thin filaments are comprised of two actin strands, each harboring equally separated troponin complexes, which bind Ca2+ to move tropomyosin cables away from the myosin binding sites and, thus, activate systolic contraction. Recently, structures of thin filaments obtained at low (pCa ∼9) or high (pCa ∼3) Ca2+ levels revealed the transition between the Ca2+-free and Ca2+-bound states. However, in working cardiac muscle, Ca2+ levels fluctuate at intermediate values between pCa ∼6 and pCa ∼7. The structure of the thin filament at physiological Ca2+ levels is unknown. We used cryoelectron microscopy and statistical analysis to reveal the structure of the cardiac thin filament at systolic pCa = 5.8. We show that the two strands of the thin filament consist of a mixture of regulatory units, which are composed of Ca2+-free, Ca2+-bound, or mixed (e.g., Ca2+ free on one side and Ca2+ bound on the other side) troponin complexes. We traced troponin complex conformations along and across individual thin filaments to directly determine the structural composition of the cardiac native thin filament at systolic Ca2+ levels. We demonstrate that the two thin filament strands are activated stochastically with short-range cooperativity evident only on one of the two strands. Our findings suggest a mechanism by which cardiac muscle is regulated by narrow range Ca2+ fluctuations.

Cavitation is the sudden, unstable expansion of a void or bubble within a liquid or solid subjected to a negative hydrostatic stress. Cavitation rheology is a field emerging from the development of a suite of materials characterization, damage quantification, and therapeutic techniques that exploit the physical principles of cavitation. Cavitation rheology is inherently complex and broad in scope with wide-ranging applications in the biology, chemistry, materials, and mechanics communities. This perspective aims to drive collaboration among these communities and guide discussion by defining a common core of highpriority goals while highlighting emerging opportunities in the field of cavitation rheology. A brief overview of the mechanics and dynamics of cavitation in soft matter is presented. This overview is followed by a discussion of the overarching goals of cavitation rheology and an overview of common experimental techniques. The larger unmet needs and challenges of cavitation in soft matter are then presented alongside specific opportunities for researchers from different disciplines to contribute to the field.

The prediction of interresidue contacts and distances from coevolutionary data using deep learning has considerably advanced protein structure prediction. Here, we build on these advances by developing a deep residual network for predicting interresidue orientations, in addition to distances, and a Rosetta-constrained energy-minimization protocol for rapidly and accurately generating structure models guided by these restraints. In benchmark tests on 13th Community-Wide Experiment on the Critical Assessment of Techniques for Protein Structure Prediction (CASP13)- and Continuous Automated Model Evaluation (CAMEO)-derived sets, the method outperforms all previously described structure-prediction methods. Although trained entirely on native proteins, the network consistently assigns higher probability to de novo-designed proteins, identifying the key fold-determining residues and providing an independent quantitative measure of the “ideality” of a protein structure. The method promises to be useful for a broad range of protein structure prediction and design problems.

Membrane bending is a ubiquitous cellular process that is required for membrane traffic, cell motility, organelle biogenesis, and cell division. Proteins that bind to membranes using specific structural features, such as wedge-like amphipathic helices and crescentshaped scaffolds, are thought to be the primary drivers of membrane bending. However, many membrane-binding proteins have substantial regions of intrinsic disorder which lack a stable three-dimensional structure. Interestingly, many of these disordered domains have recently been found to form networks stabilized by weak, multivalent contacts, leading to assembly of protein liquid phases on membrane surfaces. Here we ask how membrane-associated protein liquids impact membrane curvature. We find that protein phase separation on the surfaces of synthetic and cell-derived membrane vesicles creates a substantial compressive stress in the plane of the membrane. This stress drives the membrane to bend inward, creating protein-lined membrane tubules. A simple mechanical model of this process accurately predicts the experimentally measured relationship between the rigidity of the membrane and the diameter of the membrane tubules. Discovery of this mechanism, which may be relevant to a broad range of cellular protrusions, illustrates that membrane remodeling is not exclusive to structured scaffolds but can also be driven by the rapidly emerging class of liquid-like protein networks that assemble at membranes.

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/.