Accelerating physical simulations of proteins by leveraging external knowledge

MELD-Bracket Ranks Binding Affinities of Diverse Sets of Ligands

Affinity ranking of structurally diverse small-molecule ligands is a challenging problem with important applications in structure-based drug discovery. Absolute binding free energy methods can model diverse ligands, but the high computational cost of the current methods limits application to data sets with few ligands. We recently developed MELD-Bracket, a Molecular Dynamics method for efficient affinity ranking of ligands [ JCTC 2022, 18 (1), 374-379]. It utilizes a Bayesian framework to guide sampling to relevant regions of phase space, and it couples this with a bracket-like competition on a pool of ligands. Here we find that 6-competitor MELD-Bracket can rank dozens of diverse ligands that have low structural similarity and different net charges. We benchmark it on four protein systems─PTB1B, Tyk2, BACE, and JAK3─having varied modes of interactions. We also validated 8-competitor and 12-competitor protocols. The MELD-Bracket protocols presented here may have the appropriate balance of accuracy and computational efficiency to be suitable for ranking diverse ligands from typical drug discovery campaigns.

Accelerating Protein Folding Molecular Dynamics Using Inter-Residue Distances from Machine Learning Servers

Recently, predicting the native structures of proteins has become possible using computational molecular physics (CMP)─physics-based force fields sampled with proper statistics─but only for small proteins. Algorithms with better scaling are needed. We describe ML x MELD x MD, a molecular dynamics (MD) method that inputs residue contacts derived from machine learning (ML) servers into MELD, a Bayesian accelerator that preserves detailed-balance statistics. Contacts are derived from trRosetta-predicted distance histograms (distograms) and are integrated into MELD's atomistic MD as spatial restraints through parametrized potential functions. In the CASP14 blind prediction event, ML x MELD x MD predicted 13 native structures to better than 4.5 Å error, including for 10 proteins in the range of 115-250 amino acids long. Also, the scaling of simulation time vs protein length is much better than unguided MD: tsim ∼ e0.023N for ML x MELD x MD vs tsim ∼ e0.168N for MD alone. This shows how machine learning information can be leveraged to advance physics-based modeling of proteins.

The ugly, bad, and good stories of large-scale biomolecular simulations

Molecular modeling of large biomolecular assemblies exemplifies a disruptive area holding both promises and contentions. Propelled by peta and exascale computing, several simulation methodologies have now matured into user-friendly tools that are successfully employed for modeling viruses, membranous nano-constructs, and key pieces of the genetic machinery. We present three unifying biophysical themes that emanate from some of the most recent multi-million atom simulation endeavors. Despite connecting molecular changes with phenotypic outcomes, the quality measures of these simulations remain questionable. We discuss the existing and upcoming strategies for constructing representative ensembles of large systems, how new computing technologies will boost this area, and make a point that integrative modeling guided by experimental data is the future of biomolecular computations.

MELD-accelerated molecular dynamics help determine amyloid fibril structures

It is challenging to determine the structures of protein fibrils such as amyloids. In principle, Molecular Dynamics (MD) modeling can aid experiments, but normal MD has been impractical for these large multi-molecules. Here, we show that MELD accelerated MD (MELD x MD) can give amyloid structures from limited data. Five long-chain fibril structures are accurately predicted from NMR and Solid State NMR (SSNMR) data. Ten short-chain fibril structures are accurately predicted from more limited restraints information derived from the knowledge of strand directions. Although the present study only tests against structure predictions - which are the most detailed form of validation currently available - the main promise of this physical approach is ultimately in going beyond structures to also give mechanical properties, conformational ensembles, and relative stabilities.

Computing Ligands Bound to Proteins Using MELD-Accelerated MD

Predicting the poses of small-molecule ligands in protein binding sites is often done by virtual screening algorithms such as DOCK. In principle, molecular dynamics (MD) using atomistic force fields could give better free-energy-based pose selection, but MD is computationally expensive. Here, we ask if modeling employing limited data (MELD)-accelerated MD (MELD × MD) can pick out the best DOCK poses taken as input. We study 30 different ligand-protein pairs. MELD × MD finds native poses, based on best free energies, in 23 out of the 30 cases, 20 of which were previously known DOCK failures. We conclude that MELD × MD can add value for predicting accurate poses of small molecules bound to proteins.

Liquid network connectivity regulates the stability and composition of biomolecular condensates with many components

One of the key mechanisms used by cells to control the spatiotemporal organization of their many components is the formation and dissolution of biomolecular condensates through liquid-liquid phase separation (LLPS). Using a minimal coarse-grained model that allows us to simulate thousands of interacting multivalent proteins, we investigate the physical parameters dictating the stability and composition of multicomponent biomolecular condensates. We demonstrate that the molecular connectivity of the condensed-liquid network-i.e., the number of weak attractive protein-protein interactions per unit of volume-determines the stability (e.g., in temperature, pH, salt concentration) of multicomponent condensates, where stability is positively correlated with connectivity. While the connectivity of scaffolds (biomolecules essential for LLPS) dominates the phase landscape, introduction of clients (species recruited via scaffold-client interactions) fine-tunes it by transforming the scaffold-scaffold bond network. Whereas low-valency clients that compete for scaffold-scaffold binding sites decrease connectivity and stability, those that bind to alternate scaffold sites not required for LLPS or that have higher-than-scaffold valencies form additional scaffold-client-scaffold bridges increasing stability. Proteins that establish more connections (via increased valencies, promiscuous binding, and topologies that enable multivalent interactions) support the stability of and are enriched within multicomponent condensates. Importantly, proteins that increase the connectivity of multicomponent condensates have higher critical points as pure systems or, if pure LLPS is unfeasible, as binary scaffold-client mixtures. Hence, critical points of accessible systems (i.e., with just a few components) might serve as a unified thermodynamic parameter to predict the composition of multicomponent condensates.

Enhancing fragment-based protein structure prediction by customising fragment cardinality according to local secondary structure

BACKGROUND

Whenever suitable template structures are not available, usage of fragment-based protein structure prediction becomes the only practical alternative as pure ab initio techniques require massive computational resources even for very small proteins. However, inaccuracy of their energy functions and their stochastic nature imposes generation of a large number of decoys to explore adequately the solution space, limiting their usage to small proteins. Taking advantage of the uneven complexity of the sequence-structure relationship of short fragments, we adjusted the fragment insertion process by customising the number of available fragment templates according to the expected complexity of the predicted local secondary structure. Whereas the number of fragments is kept to its default value for coil regions, important and dramatic reductions are proposed for beta sheet and alpha helical regions, respectively.

RESULTS

The evaluation of our fragment selection approach was conducted using an enhanced version of the popular Rosetta fragment-based protein structure prediction tool. It was modified so that the number of fragment candidates used in Rosetta could be adjusted based on the local secondary structure. Compared to Rosetta's standard predictions, our strategy delivered improved first models, + 24% and + 6% in terms of GDT, when using 2000 and 20,000 decoys, respectively, while reducing significantly the number of fragment candidates. Furthermore, our enhanced version of Rosetta is able to deliver with 2000 decoys a performance equivalent to that produced by standard Rosetta while using 20,000 decoys. We hypothesise that, as the fragment insertion process focuses on the most challenging regions, such as coils, fewer decoys are needed to explore satisfactorily conformation spaces.

CONCLUSIONS

Taking advantage of the high accuracy of sequence-based secondary structure predictions, we showed the value of that information to customise the number of candidates used during the fragment insertion process of fragment-based protein structure prediction. Experimentations conducted using standard Rosetta showed that, when using the recommended number of decoys, i.e. 20,000, our strategy produces better results. Alternatively, similar results can be achieved using only 2000 decoys. Consequently, we recommend the adoption of this strategy to either improve significantly model quality or reduce processing times by a factor 10.

Applications of Molecular Dynamics Simulation in Structure Prediction of Peptides and Proteins

Compared with rapid accumulation of protein sequences from high-throughput DNA sequencing, obtaining experimental 3D structures of proteins is still much more difficult, making protein structure prediction (PSP) potentially very useful. Currently, a vast majority of PSP efforts are based on data mining of known sequences, structures and their relationships (informatics-based). However, if closely related template is not available, these methods are usually much less reliable than experiments. They may also be problematic in predicting the structures of naturally occurring or designed peptides. On the other hand, physics-based methods including molecular dynamics (MD) can utilize our understanding of detailed atomic interactions determining biomolecular structures. In this mini-review, we show that all-atom MD can predict structures of cyclic peptides and other peptide foldamers with accuracy similar to experiments. Then, some notable successes in reproducing experimental 3D structures of small proteins through MD simulations (some with replica-exchange) of the folding were summarized. We also describe advancements of MD-based refinement of structure models, and the integration of limited experimental or bioinformatics data into MD-based structure modeling.

Predicting Protein Dimer Structures Using MELD × MD

It is challenging to predict the docked conformations of two proteins. Current methods are susceptible to errors from treating proteins as rigid bodies and from an inability to compute relative Boltzmann populations of different docked conformations. Here, we show that by using the ClusPro server as a front end to generate possible protein-protein contacts, and using Modeling Employing Limited Data (MELD) accelerated molecular dynamics (MELD × MD) as a back end for atomistic simulations, we can find 16/20 native dimer structures of small proteins as those having the lowest free energy, starting from good-bound-backbone structures. We show that atomistic MD free energies can be used to identify native protein dimer structures.

How Effectively Can Adaptive Sampling Methods Capture Spontaneous Ligand Binding?

Molecular dynamics (MD) simulations that capture the spontaneous binding of drugs and other ligands to their target proteins can reveal a great deal of useful information, but most drug-like ligands bind on time scales longer than those accessible to individual MD simulations. Adaptive sampling methods―in which one performs multiple rounds of simulation, with the initial conditions of each round based on the results of previous rounds―offer a promising potential solution to this problem. No comprehensive analysis of the performance gains from adaptive sampling is available for ligand binding, however, particularly for protein–ligand systems typical of those encountered in drug discovery. Moreover, most previous work presupposes knowledge of the ligand’s bound pose. Here we outline existing methods for adaptive sampling of the ligand-binding process and introduce several improvements, with a focus on methods that do not require prior knowledge of the binding site or bound pose. We then evaluate these methods by comparing them to traditional, long MD simulations for realistic protein–ligand systems. We find that adaptive sampling simulations typically fail to reach the bound pose more efficiently than traditional MD. However, adaptive sampling identifies multiple potential binding sites more efficiently than traditional MD and also provides better characterization of binding pathways. We explain these results by showing that protein–ligand binding is an example of an exploration–exploitation dilemma. Existing adaptive sampling methods for ligand binding in the absence of a known bound pose vastly favor the broad exploration of protein–ligand space, sometimes failing to sufficiently exploit intermediate states as they are discovered. We suggest potential avenues for future research to address this shortcoming.

Computational Exploration of Conformational Transitions in Protein Drug Targets

Protein drug targets vary from highly structured to completely disordered; either way dynamics governs function. Hence, understanding the dynamical aspects of how protein targets function can enable improved interventions with drug molecules. Computational approaches offer highly detailed structural models of protein dynamics which are becoming more predictive as model quality and sampling power improve. However, the most advanced and popular models still have errors owing to imperfect parameter sets and often cannot access longer timescales of many crucial biological processes. Experimental approaches offer more certainty but can struggle to detect and measure lightly populated conformations of target proteins and subtle allostery. An emerging solution is to integrate available experimental data into advanced molecular simulations. In the future, molecular simulation in combination with experimental data may be able to offer detailed models of important drug targets such that improved functional mechanisms or selectivity can be accessed.