An artificial intelligence accelerated virtual screening platform for drug discovery.

Structure-based virtual screening is a key tool in early drug discovery, with growing interest in the screening of multi-billion chemical compound libraries. However, the success of virtual screening crucially depends on the accuracy of the binding pose and binding affinity predicted by computational docking. Here we develop a highly accurate structure-based virtual screen method, RosettaVS, for predicting docking poses and binding affinities. Our approach outperforms other state-of-the-art methods on a wide range of benchmarks, partially due to our ability to model receptor flexibility. We incorporate this into a new open-source artificial intelligence accelerated virtual screening platform for drug discovery. Using this platform, we screen multi-billion compound libraries against two unrelated targets, a ubiquitin ligase target KLHDC2 and the human voltage-gated sodium channel NaV1.7. For both targets, we discover hit compounds, including seven hits (14% hit rate) to KLHDC2 and four hits (44% hit rate) to NaV1.7, all with single digit micromolar binding affinities. Screening in both cases is completed in less than seven days. Finally, a high resolution X-ray crystallographic structure validates the predicted docking pose for the KLHDC2 ligand complex, demonstrating the effectiveness of our method in lead discovery.

GalaxyDock-DL: Protein-Ligand Docking by Global Optimization and Neural Network Energy.

With the recent introduction of deep learning techniques into the prediction of biomolecular structures, structure prediction performance has significantly improved, and the potential for biomedical applications has increased considerably. The prediction of protein-ligand complex structures, applicable to the atomistic understanding of biomolecular functions and the effective design of drug molecules, has also improved with the introduction of deep learning. In this paper, it is demonstrated that docking performance can be greatly enhanced by training an energy function that encapsulates physical effects using deep learning within the framework of the traditional protein-ligand docking method. The advantage of this method, called GalaxyDock-DL, lies in its minimal overfitting to the training data compared to several existing deep learning-based protein-ligand docking methods. Unlike some recent deep learning methods, it does not use information about known binding pocket center positions. Instead, the results of this docking method show a systematic dependence on the physical properties of the target protein-ligand complexes such as atomic thermal fluctuations and binding affinity. GalaxyDock-DL utilizes the global optimization technique of the conventional protein-ligand docking method, GalaxyDock, and a neural network energy function trained to stabilize the native state compared to non-native states, just as physical free energy does. This physical principle-based approach suggests directions not only for future structure prediction involving structurally flexible biomolecular complexes but also for predicting binding affinity, thereby providing guidance for the effective design of biofunctional ligands.

Protein Ensemble Generation Through Variational Autoencoder Latent Space Sampling.

Mapping the ensemble of protein conformations that contribute to function and can be targeted by small molecule drugs remains an outstanding challenge. Here, we explore the use of variational autoencoders for reducing the challenge of dimensionality in the protein structure ensemble generation problem. We convert high-dimensional protein structural data into a continuous, low-dimensional representation, carry out a search in this space guided by a structure quality metric, and then use RoseTTAFold guided by the sampled structural information to generate 3D structures. We use this approach to generate ensembles for the cancer relevant protein K-Ras, train the VAE on a subset of the available K-Ras crystal structures and MD simulation snapshots, and assess the extent of sampling close to crystal structures withheld from training. We find that our latent space sampling procedure rapidly generates ensembles with high structural quality and is able to sample within 1 Å of held-out crystal structures, with a consistency higher than that of MD simulation or AlphaFold2 prediction. The sampled structures sufficiently recapitulate the cryptic pockets in the held-out K-Ras structures to allow for small molecule docking.

Understanding the molecular mechanisms of odorant binding and activation of the human OR52 family.

Structural and mechanistic studies on human odorant receptors (ORs), key in olfactory signaling, are challenging because of their low surface expression in heterologous cells. The recent structure of OR51E2 bound to propionate provided molecular insight into odorant recognition, but the lack of an inactive OR structure limited understanding of the activation mechanism of ORs upon odorant binding. Here, we determined the cryo-electron microscopy structures of consensus OR52 (OR52cs), a representative of the OR52 family, in the ligand-free (apo) and octanoate-bound states. The apo structure of OR52cs reveals a large opening between transmembrane helices (TMs) 5 and 6. A comparison between the apo and active structures of OR52cs demonstrates the inward and outward movements of the extracellular and intracellular segments of TM6, respectively. These results, combined with molecular dynamics simulations and signaling assays, shed light on the molecular mechanisms of odorant binding and activation of the OR52 family.

Development of a novel histone deacetylase inhibitor unveils the role of HDAC11 in alleviating depression by inhibition of microglial activation.

Histone deacetylases (HDACs) are key epigenetic regulators and classified into four subtypes. Despite the various roles of each HDAC isoform, the lack of selective HDAC inhibitors has limited the elucidation of their roles in biological systems. HDAC11, the sole class-IV HDAC, is highly expressed in the brain, however, the role of HDAC11 in microglia is not fully understood. Based on the modification of MC1568, we developed a novel HDAC inhibitor, 5. Interestingly, 5 suppresses lipopolysaccharide-induced microglial activation by the initiation of autophagy and subsequent inhibition of nitric oxide production. Furthermore, we demonstrated that 5 significantly alleviates depression-like behavior by inhibiting microglial activation in mouse brain. Our discovery reveals that specific pharmacological regulation of HDAC11 induces autophagy and reactive nitrogen species balance in microglia for the first time, which makes HDAC11 a new therapeutic target for depressive disorder.

Benchmarking applicability of medium-resolution cryo-EM protein structures for structure-based drug design.

Cryo-electron microscopy (cryo-EM) is gaining large attention for high-resolution protein structure determination in solutions. However, a very high percentage of cryo-EM structures correspond to resolutions of 3-5 Å, making the structures difficult to be used in in silico drug design. In this study, we analyze how useful cryo-EM protein structures are for in silico drug design by evaluating ligand docking accuracy. From realistic cross-docking scenarios using medium resolution (3-5 Å) cryo-EM structures and a popular docking tool Autodock-Vina, only 20% of docking succeeded, when the success rate doubles in the same kind of cross-docking but using high-resolution (<2 Å) crystal structures instead. We decipher the reason for failures by decomposing the contribution from resolution-dependent and independent factors. The heterogeneity in the protein side-chain and backbone conformations is identified as the major resolution-dependent factor causing docking difficulty from our analysis, while intrinsic receptor flexibility mainly comprises the resolution-independent factor. We demonstrate the flexibility implementation in current ligand docking tools is able to rescue only a portion of failures (10%), and the limited performance was majorly due to potential structural errors than conformational changes. Our work suggests the strong necessity of more robust method developments on ligand docking and EM modeling techniques in order to fully utilize cryo-EM structures for in silico drug design.

Evaluating GPCR modeling and docking strategies in the era of deep learning-based protein structure prediction.

While deep learning (DL) has brought a revolution in the protein structure prediction field, still an important question remains how the revolution can be transferred to advances in structure-based drug discovery. Because the lessons from the recent GPCR dock challenge were inconclusive primarily due to the size of the dataset, in this work we further elaborated on 70 diverse GPCR complexes bound to either small molecules or peptides to investigate the best-practice modeling and docking strategies for GPCR drug discovery. From our quantitative analysis, it is shown that substantial improvements in docking and virtual screening have been possible by the advance in DL-based protein structure predictions with respect to the expected results from the combination of best pre-DL tools. The success rate of docking on DL-based model structures approaches that of cross-docking on experimental structures, showing over 30% improvement from the best pre-DL protocols. This amount of performance could be achieved only when two modeling points were considered properly: 1) correct functional-state modeling of receptors and 2) receptor-flexible docking. Best-practice modeling strategies and the model confidence estimation metric suggested in this work may serve as a guideline for future computer-aided GPCR drug discovery scenarios.

Pseudo-Isolated α-Helix Platform for the Recognition of Deep and Narrow Targets.

Although interest in stabilized α-helical peptides as next-generation therapeutics for modulating biomolecular interfaces is increasing, peptides have limited functionality and stability due to their small size. In comparison, α-helical ligands based on proteins can make steric clash with targets due to their large size. Here, we report the design of a monomeric pseudo-isolated α-helix (mPIH) system in which proteins behave as if they are peptides. The designed proteins contain α-helix ligands that do not require any covalent chemical modification, do not have frayed ends, and importantly can make sterically favorable interactions similar to isolated peptides. An optimal mPIH showed a more than 100-fold increase in target selectivity, which might be related to the advantages in conformational selection due to the absence of frayed ends. The α-helical ligand in the mPIH displayed high thermal stability well above human body temperature and showed reversible and rapid folding/unfolding transitions. Thus, mPIH can become a promising protein-based platform for developing stabilized α-helix pharmaceuticals.

Divergent acyl carrier protein decouples mitochondrial Fe-S cluster biogenesis from fatty acid synthesis in malaria parasites.

Most eukaryotic cells retain a mitochondrial fatty acid synthesis (FASII) pathway whose acyl carrier protein (mACP) and 4-phosphopantetheine (Ppant) prosthetic group provide a soluble scaffold for acyl chain synthesis and biochemically couple FASII activity to mitochondrial electron transport chain (ETC) assembly and Fe-S cluster biogenesis. In contrast, the mitochondrion of Plasmodium falciparum malaria parasites lacks FASII enzymes yet curiously retains a divergent mACP lacking a Ppant group. We report that ligand-dependent knockdown of mACP is lethal to parasites, indicating an essential FASII-independent function. Decyl-ubiquinone rescues parasites temporarily from death, suggesting a dominant dysfunction of the mitochondrial ETC. Biochemical studies reveal that Plasmodium mACP binds and stabilizes the Isd11-Nfs1 complex required for Fe-S cluster biosynthesis, despite lacking the Ppant group required for this association in other eukaryotes, and knockdown of parasite mACP causes loss of Nfs1 and the Rieske Fe-S protein in ETC complex III. This work reveals that Plasmodium parasites have evolved to decouple mitochondrial Fe-S cluster biogenesis from FASII activity, and this adaptation is a shared metabolic feature of other apicomplexan pathogens, including Toxoplasma and Babesia. This discovery unveils an evolutionary driving force to retain interaction of mitochondrial Fe-S cluster biogenesis with ACP independent of its eponymous function in FASII.

Protein tertiary structure prediction and refinement using deep learning and Rosetta in CASP14.

The trRosetta structure prediction method employs deep learning to generate predicted residue-residue distance and orientation distributions from which 3D models are built. We sought to improve the method by incorporating as inputs (in addition to sequence information) both language model embeddings and template information weighted by sequence similarity to the target. We also developed a refinement pipeline that recombines models generated by template-free and template utilizing versions of trRosetta guided by the DeepAccNet accuracy predictor. Both benchmark tests and CASP results show that the new pipeline is a considerable improvement over the original trRosetta, and it is faster and requires less computing resources, completing the entire modeling process in a median < 3 h in CASP14. Our human group improved results with this pipeline primarily by identifying additional homologous sequences for input into the network. We also used the DeepAccNet accuracy predictor to guide Rosetta high-resolution refinement for submissions in the regular and refinement categories; although performance was quite good on a CASP relative scale, the overall improvements were rather modest in part due to missing inter-domain or inter-chain contacts.

Protein oligomer modeling guided by predicted interchain contacts in CASP14.

For CASP14, we developed deep learning-based methods for predicting homo-oligomeric and hetero-oligomeric contacts and used them for oligomer modeling. To build structure models, we developed an oligomer structure generation method that utilizes predicted interchain contacts to guide iterative restrained minimization from random backbone structures. We supplemented this gradient-based fold-and-dock method with template-based and ab initio docking approaches using deep learning-based subunit predictions on 29 assembly targets. These methods produced oligomer models with summed Z-scores 5.5 units higher than the next best group, with the fold-and-dock method having the best relative performance. Over the eight targets for which this method was used, the best of the five submitted models had average oligomer TM-score of 0.71 (average oligomer TM-score of the next best group: 0.64), and explicit modeling of inter-subunit interactions improved modeling of six out of 40 individual domains (ΔGDT-TS > 2.0).

Accurate prediction of protein structures and interactions using a three-track neural network.

DeepMind presented notably accurate predictions at the recent 14th Critical Assessment of Structure Prediction (CASP14) conference. We explored network architectures that incorporate related ideas and obtained the best performance with a three-track network in which information at the one-dimensional (1D) sequence level, the 2D distance map level, and the 3D coordinate level is successively transformed and integrated. The three-track network produces structure predictions with accuracies approaching those of DeepMind in CASP14, enables the rapid solution of challenging x-ray crystallography and cryo-electron microscopy structure modeling problems, and provides insights into the functions of proteins of currently unknown structure. The network also enables rapid generation of accurate protein-protein complex models from sequence information alone, short-circuiting traditional approaches that require modeling of individual subunits followed by docking. We make the method available to the scientific community to speed biological research.

Force Field Optimization Guided by Small Molecule Crystal Lattice Data Enables Consistent Sub-Angstrom Protein-Ligand Docking.

Accurate and rapid calculation of protein-small molecule interaction free energies is critical for computational drug discovery. Because of the large chemical space spanned by drug-like molecules, classical force fields contain thousands of parameters describing atom-pair distance and torsional preferences; each parameter is typically optimized independently on simple representative molecules. Here, we describe a new approach in which small molecule force field parameters are jointly optimized guided by the rich source of information contained within thousands of available small molecule crystal structures. We optimize parameters by requiring that the experimentally determined molecular lattice arrangements have lower energy than all alternative lattice arrangements. Thousands of independent crystal lattice-prediction simulations were run on each of 1386 small molecule crystal structures, and energy function parameters of an implicit solvent energy model were optimized, so native crystal lattice arrangements had the lowest energy. The resulting energy model was implemented in Rosetta, together with a rapid genetic algorithm docking method employing grid-based scoring and receptor flexibility. The success rate of bound structure recapitulation in cross-docking on 1112 complexes was improved by more than 10% over previously published methods, with solutions within <1 Å in over half of the cases. Our results demonstrate that small molecule crystal structures are a rich source of information for guiding molecular force field development, and the improved Rosetta energy function should increase accuracy in a wide range of small molecule structure prediction and design studies.

Improved protein structure refinement guided by deep learning based accuracy estimation.

We develop a deep learning framework (DeepAccNet) that estimates per-residue accuracy and residue-residue distance signed error in protein models and uses these predictions to guide Rosetta protein structure refinement. The network uses 3D convolutions to evaluate local atomic environments followed by 2D convolutions to provide their global contexts and outperforms other methods that similarly predict the accuracy of protein structure models. Overall accuracy predictions for X-ray and cryoEM structures in the PDB correlate with their resolution, and the network should be broadly useful for assessing the accuracy of both predicted structure models and experimentally determined structures and identifying specific regions likely to be in error. Incorporation of the accuracy predictions at multiple stages in the Rosetta refinement protocol considerably increased the accuracy of the resulting protein structure models, illustrating how deep learning can improve search for global energy minima of biomolecules.

Prediction of Protein Mutational Free Energy: Benchmark and Sampling Improvements Increase Classification Accuracy.

Software to predict the change in protein stability upon point mutation is a valuable tool for a number of biotechnological and scientific problems. To facilitate the development of such software and provide easy access to the available experimental data, the ProTherm database was created. Biases in the methods and types of information collected has led to disparity in the types of mutations for which experimental data is available. For example, mutations to alanine are hugely overrepresented whereas those involving charged residues, especially from one charged residue to another, are underrepresented. ProTherm subsets created as benchmark sets that do not account for this often underrepresent tense certain mutational types. This issue introduces systematic biases into previously published protocols' ability to accurately predict the change in folding energy on these classes of mutations. To resolve this issue, we have generated a new benchmark set with these problems corrected. We have then used the benchmark set to test a number of improvements to the point mutation energetics tools in the Rosetta software suite.

Efficient consideration of coordinated water molecules improves computational protein-protein and protein-ligand docking discrimination.

Highly coordinated water molecules are frequently an integral part of protein-protein and protein-ligand interfaces. We introduce an updated energy model that efficiently captures the energetic effects of these ordered water molecules on the surfaces of proteins. A two-stage method is developed in which polar groups arranged in geometries suitable for water placement are first identified, then a modified Monte Carlo simulation allows highly coordinated waters to be placed on the surface of a protein while simultaneously sampling amino acid side chain orientations. This "semi-explicit" water model is implemented in Rosetta and is suitable for both structure prediction and protein design. We show that our new approach and energy model yield significant improvements in native structure recovery of protein-protein and protein-ligand docking discrimination tests.

Improved protein structure prediction using predicted interresidue orientations.

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.

High-accuracy refinement using Rosetta in CASP13.

Because proteins generally fold to their lowest free energy states, energy-guided refinement in principle should be able to systematically improve the quality of protein structure models generated using homologous structure or co-evolution derived information. However, because of the high dimensionality of the search space, there are far more ways to degrade the quality of a near native model than to improve it, and hence, refinement methods are very sensitive to energy function errors. In the 13th Critial Assessment of techniques for protein Structure Prediction (CASP13), we sought to carry out a thorough search for low energy states in the neighborhood of a starting model using restraints to avoid straying too far. The approach was reasonably successful in improving both regions largely incorrect in the starting models as well as core regions that started out closer to the correct structure. Models with GDT-HA over 70 were obtained for five targets and for one of those, an accuracy of 0.5 å backbone root-mean-square deviation (RMSD) was achieved. An important current challenge is to improve performance in refining oligomers and larger proteins, for which the search problem remains extremely difficult.

De novo design of a fluorescence-activating ß-barrel.

The regular arrangements of ß-strands around a central axis in ß-barrels and of α-helices in coiled coils contrast with the irregular tertiary structures of most globular proteins, and have fascinated structural biologists since they were first discovered. Simple parametric models have been used to design a wide range of α-helical coiled-coil structures, but to date there has been no success with ß-barrels. Here we show that accurate de novo design of ß-barrels requires considerable symmetry-breaking to achieve continuous hydrogen-bond connectivity and eliminate backbone strain. We then build ensembles of ß-barrel backbone models with cavity shapes that match the fluorogenic compound DFHBI, and use a hierarchical grid-based search method to simultaneously optimize the rigid-body placement of DFHBI in these cavities and the identities of the surrounding amino acids to achieve high shape and chemical complementarity. The designs have high structural accuracy and bind and fluorescently activate DFHBI in vitro and in Escherichia coli, yeast and mammalian cells. This de novo design of small-molecule binding activity, using backbones custom-built to bind the ligand, should enable the design of increasingly sophisticated ligand-binding proteins, sensors and catalysts that are not limited by the backbone geometries available in known protein structures.