Adaptive language model training for molecular design.

The vast size of chemical space necessitates computational approaches to automate and accelerate the design of molecular sequences to guide experimental efforts for drug discovery. Genetic algorithms provide a useful framework to incrementally generate molecules by applying mutations to known chemical structures. Recently, masked language models have been applied to automate the mutation process by leveraging large compound libraries to learn commonly occurring chemical sequences (i.e., using tokenization) and predict rearrangements (i.e., using mask prediction). Here, we consider how language models can be adapted to improve molecule generation for different optimization tasks. We use two different generation strategies for comparison, fixed and adaptive. The fixed strategy uses a pre-trained model to generate mutations; the adaptive strategy trains the language model on each new generation of molecules selected for target properties during optimization. Our results show that the adaptive strategy allows the language model to more closely fit the distribution of molecules in the population. Therefore, for enhanced fitness optimization, we suggest the use of the fixed strategy during an initial phase followed by the use of the adaptive strategy. We demonstrate the impact of adaptive training by searching for molecules that optimize both heuristic metrics, drug-likeness and synthesizability, as well as predicted protein binding affinity from a surrogate model. Our results show that the adaptive strategy provides a significant improvement in fitness optimization compared to the fixed pre-trained model, empowering the application of language models to molecular design tasks.

tinyIFD: A High-Throughput Binding Pose Refinement Workflow Through Induced-Fit Ligand Docking.

A critical step in structure-based drug discovery is predicting whether and how a candidate molecule binds to a model of a therapeutic target. However, substantial protein side chain movements prevent current screening methods, such as docking, from accurately predicting the ligand conformations and require expensive refinements to produce viable candidates. We present the development of a high-throughput and flexible ligand pose refinement workflow, called "tinyIFD". The main features of the workflow include the use of specialized high-throughput, small-system MD simulation code mdgx.cuda and an actively learning model zoo approach. We show the application of this workflow on a large test set of diverse protein targets, achieving 66% and 76% success rates for finding a crystal-like pose within the top-2 and top-5 poses, respectively. We also applied this workflow to the SARS-CoV-2 main protease (Mpro) inhibitors, where we demonstrate the benefit of the active learning aspect in this workflow.

SARS-CoV2 billion-compound docking.

This dataset contains ligand conformations and docking scores for 1.4 billion molecules docked against 6 structural targets from SARS-CoV2, representing 5 unique proteins: MPro, NSP15, PLPro, RDRP, and the Spike protein. Docking was carried out using the AutoDock-GPU platform on the Summit supercomputer and Google Cloud. The docking procedure employed the Solis Wets search method to generate 20 independent ligand binding poses per compound. Each compound geometry was scored using the AutoDock free energy estimate, and rescored using RFScore v3 and DUD-E machine-learned rescoring models. Input protein structures are included, suitable for use by AutoDock-GPU and other docking programs. As the result of an exceptionally large docking campaign, this dataset represents a valuable resource for discovering trends across small molecule and protein binding sites, training AI models, and comparing to inhibitor compounds targeting SARS-CoV-2. The work also gives an example of how to organize and process data from ultra-large docking screens.

Hit Expansion of a Noncovalent SARS-CoV-2 Main Protease Inhibitor.

Inhibition of the SARS-CoV-2 main protease (Mpro) is a major focus of drug discovery efforts against COVID-19. Here we report a hit expansion of non-covalent inhibitors of Mpro. Starting from a recently discovered scaffold (The COVID Moonshot Consortium. Open Science Discovery of Oral Non-Covalent SARS-CoV-2 Main Protease Inhibitor Therapeutics. bioRxiv 2020.10.29.339317) represented by an isoquinoline series, we searched a database of over a billion compounds using a cheminformatics molecular fingerprinting approach. We identified and tested 48 compounds in enzyme inhibition assays, of which 21 exhibited inhibitory activity above 50% at 20 µM. Among these, four compounds with IC50 values around 1 µM were found. Interestingly, despite the large search space, the isoquinolone motif was conserved in each of these four strongest binders. Room-temperature X-ray structures of co-crystallized protein-inhibitor complexes were determined up to 1.9 Å resolution for two of these compounds as well as one of the stronger inhibitors in the original isoquinoline series, revealing essential interactions with the binding site and water molecules. Molecular dynamics simulations and quantum chemical calculations further elucidate the binding interactions as well as electrostatic effects on ligand binding. The results help explain the strength of this new non-covalent scaffold for Mpro inhibition and inform lead optimization efforts for this series, while demonstrating the effectiveness of a high-throughput computational approach to expanding a pharmacophore library.

Language models for the prediction of SARS-CoV-2 inhibitors.

The COVID-19 pandemic highlights the need for computational tools to automate and accelerate drug design for novel protein targets. We leverage deep learning language models to generate and score drug candidates based on predicted protein binding affinity. We pre-trained a deep learning language model (BERT) on ∼9.6 billion molecules and achieved peak performance of 603 petaflops in mixed precision. Our work reduces pre-training time from days to hours, compared to previous efforts with this architecture, while also increasing the dataset size by nearly an order of magnitude. For scoring, we fine-tuned the language model using an assembled set of thousands of protein targets with binding affinity data and searched for inhibitors of specific protein targets, SARS-CoV-2 Mpro and PLpro. We utilized a genetic algorithm approach for finding optimal candidates using the generation and scoring capabilities of the language model. Our generalizable models accelerate the identification of inhibitors for emerging therapeutic targets.

The role of complementary shape in protein dimerization.

Shape guides colloidal nanoparticles to form complex assemblies, but its role in defining interfaces in biomolecular complexes is less clear. In this work, we isolate the role of shape in protein complexes by studying the reversible binding processes of 46 protein dimer pairs, and investigate when entropic effects from shape complementarity alone are sufficient to predict the native protein binding interface. We employ depletants using a generic, implicit depletion model to amplify the magnitude of the entropic forces arising from lock-and-key binding and isolate the effect of shape complementarity in protein dimerization. For 13% of the complexes studied here, protein shape is sufficient to predict native complexes as equilibrium assemblies. We elucidate the results by analyzing the importance of competing binding configurations and how it affects the assembly. A machine learning classifier, with a precision of 89.14% and a recall of 77.11%, is able to identify the cases where shape alone predicts the native protein interface.

Supercomputing Pipelines Search for Therapeutics Against COVID-19.

The urgent search for drugs to combat SARS-CoV-2 has included the use of supercomputers. The use of general-purpose graphical processing units (GPUs), massive parallelism, and new software for high-performance computing (HPC) has allowed researchers to search the vast chemical space of potential drugs faster than ever before. We developed a new drug discovery pipeline using the Summit supercomputer at Oak Ridge National Laboratory to help pioneer this effort, with new platforms that incorporate GPU-accelerated simulation and allow for the virtual screening of billions of potential drug compounds in days compared to weeks or months for their ability to inhibit SARS-COV-2 proteins. This effort will accelerate the process of developing drugs to combat the current COVID-19 pandemic and other diseases.

Supercharging enables organized assembly of synthetic biomolecules.

Symmetrical protein oligomers are ubiquitous in biological systems and perform key structural and regulatory functions. However, there are few methods for constructing such oligomers. Here we have engineered completely synthetic, symmetrical oligomers by combining pairs of oppositely supercharged variants of a normally monomeric model protein through a strategy we term 'supercharged protein assembly' (SuPrA). We show that supercharged variants of green fluorescent protein can assemble into a variety of architectures including a well-defined symmetrical 16-mer structure that we solved using cryo-electron microscopy at 3.47 Å resolution. The 16-mer is composed of two stacked rings of octamers, in which the octamers contain supercharged proteins of alternating charges, and interactions within and between the rings are mediated by a variety of specific electrostatic contacts. The ready assembly of this structure suggests that combining oppositely supercharged pairs of protein variants may provide broad opportunities for generating novel architectures via otherwise unprogrammed interactions.

Using depletion to control colloidal crystal assemblies of hard cuboctahedra.

Depletion interactions arise from entropic forces, and their ability to induce aggregation and even ordering of colloidal particles through self-assembly is well established, especially for spherical colloids. We vary the size and concentration of penetrable hard sphere depletants in a system of cuboctahedra, and we show how depletion changes the preferential facet alignment of the colloids and thereby selects different crystal structures. Moreover, we explain the cuboctahedra phase behavior using perturbative free energy calculations. We find that cuboctahedra can form a stable simple cubic phase, and, remarkably, that the stability of this phase can be rationalized only by considering the effects of both the colloid and depletant entropy. We corroborate our results by analyzing how the depletant concentration and size affect the emergent directional entropic forces and hence the effective particle shape. We propose the use of depletants as a means of easily changing the effective shape of self-assembling anisotropic colloids.

A parallel algorithm for implicit depletant simulations.

We present an algorithm to simulate the many-body depletion interaction between anisotropic colloids in an implicit way, integrating out the degrees of freedom of the depletants, which we treat as an ideal gas. Because the depletant particles are statistically independent and the depletion interaction is short-ranged, depletants are randomly inserted in parallel into the excluded volume surrounding a single translated and/or rotated colloid. A configurational bias scheme is used to enhance the acceptance rate. The method is validated and benchmarked both on multi-core processors and graphics processing units for the case of hard spheres, hemispheres, and discoids. With depletants, we report novel cluster phases in which hemispheres first assemble into spheres, which then form ordered hcp/fcc lattices. The method is significantly faster than any method without cluster moves and that tracks depletants explicitly, for systems of colloid packing fraction ϕc < 0.50, and additionally enables simulation of the fluid-solid transition.

Metastable orientational order of colloidal discoids.

The interplay between phase separation and kinetic arrest is important in supramolecular self-assembly, but their effects on emergent orientational order are not well understood when anisotropic building blocks are used. Contrary to the typical progression from disorder to order in isotropic systems, here we report that colloidal oblate discoids initially self-assemble into short, metastable strands with orientational order­regardless of the final structure. The model discoids are suspended in a refractive index and density-matched solvent. Then, we use confocal microscopy experiments and Monte Carlo simulations spanning a broad range of volume fractions and attraction strengths to show that disordered clusters form near coexistence boundaries, whereas oriented strands persist with strong attractions. We rationalize this unusual observation in light of the interaction anisotropy imparted by the discoids. These findings may guide self-assembly for anisotropic systems in which orientational order is desired, such as when tailored mechanical properties are sought.

Universality of block copolymer melts.

Simulations of five different coarse-grained models of symmetric diblock copolymers are compared to demonstrate a universal (i.e., model-independent) dependence of the free energy and order-disorder transition (ODT) on the invariant degree of polymerization N̄. The actual values of χN at the ODT approach predictions of the Fredrickson-Helfand (FH) theory for N̄ ≳ 10(4) but significantly exceed FH predictions at lower values characteristic of most experiments. The FH theory fails for modest N̄ because the competing phases become strongly segregated near the ODT, violating an underlying assumption of weak segregation.

Tube-width fluctuations of entangled stiff polymers.

The tubelike cages of stiff polymers in entangled solutions have been shown to exhibit characteristic spatial heterogeneities. We explain these observations by a systematic theory generalizing previous work by Morse [Phys. Rev. E 63, 031502 (2001)]. With a local version of the binary collision approximation, the distribution of confinement strengths is calculated, and the magnitude and the distribution function of tube radius fluctuations are predicted. Our main result is a unique scaling function for the tube radius distribution, in good agreement with experimental and simulation data.

Glass transition and rheological redundancy in F-actin solutions.

The unique mechanical performance of animal cells and tissues is attributed mostly to their internal biopolymer meshworks. Its perplexing universality and robustness against structural modifications by drugs and mutations is an enigma in cell biology and provides formidable challenges to materials science. Recent investigations could pinpoint highly universal patterns in the soft glassy rheology and nonlinear elasticity of cells and reconstituted networks. Here, we report observations of a glass transition in semidilute F-actin solutions, which could hold the key to a unified explanation of these phenomena. Combining suitable rheological protocols with high-precision dynamic light scattering, we can establish a remarkable rheological redundancy and trace it back to a highly universal exponential stretching of the single-polymer relaxation spectrum of a "glassy wormlike chain." By exploiting the ensuing generalized time-temperature superposition principle, the time domain accessible to microrheometry can be extended by several orders of magnitude, thus opening promising new metrological opportunities.

High-Yielding Rotaxane Synthesis with an Anion Template.

An electrophile caught like a mouse in a trap! An anionic stopper-wheel complex acts as a supramolecular nucleophile in an almost quantitative synthesis of a phenyl ether rotaxane. The electrophilic semiaxle has to thread through the macrocycle in order to contact the bulky phenolate group that is positioned on the other side, and probably tightly held in place by hydrogen bonds.