Open science discovery of potent noncovalent SARS-CoV-2 main protease inhibitors.

We report the results of the COVID Moonshot, a fully open-science, crowdsourced, and structure-enabled drug discovery campaign targeting the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) main protease. We discovered a noncovalent, nonpeptidic inhibitor scaffold with lead-like properties that is differentiated from current main protease inhibitors. Our approach leveraged crowdsourcing, machine learning, exascale molecular simulations, and high-throughput structural biology and chemistry. We generated a detailed map of the structural plasticity of the SARS-CoV-2 main protease, extensive structure-activity relationships for multiple chemotypes, and a wealth of biochemical activity data. All compound designs (>18,000 designs), crystallographic data (>490 ligand-bound x-ray structures), assay data (>10,000 measurements), and synthesized molecules (>2400 compounds) for this campaign were shared rapidly and openly, creating a rich, open, and intellectual property-free knowledge base for future anticoronavirus drug discovery.

Protein structure, amino acid composition and sequence determine proteome vulnerability to oxidation-induced damage.

Oxidative stress alters cell viability, from microorganism irradiation sensitivity to human aging and neurodegeneration. Deleterious effects of protein carbonylation by reactive oxygen species (ROS) make understanding molecular properties determining ROS susceptibility essential. The radiation-resistant bacterium Deinococcus radiodurans accumulates less carbonylation than sensitive organisms, making it a key model for deciphering properties governing oxidative stress resistance. We integrated shotgun redox proteomics, structural systems biology, and machine learning to resolve properties determining protein damage by γ-irradiation in Escherichia coli and D. radiodurans at multiple scales. Local accessibility, charge, and lysine enrichment accurately predict ROS susceptibility. Lysine, methionine, and cysteine usage also contribute to ROS resistance of the D. radiodurans proteome. Our model predicts proteome maintenance machinery, and proteins protecting against ROS are more resistant in D. radiodurans. Our findings substantiate that protein-intrinsic protection impacts oxidative stress resistance, identifying causal molecular properties.

Impact of Chemist-In-The-Loop Molecular Representations on Machine Learning Outcomes.

The development of molecular descriptors is a central challenge in cheminformatics. Most approaches use algorithms that extract atomic environments or end-to-end machine learning. However, a looming question is that how do these approaches compare with the critical eye of trained chemists. The CAS fingerprint engages expert chemists to curate chemical motifs, which they deem could influence bioactivity. In this paper, we benchmark the CAS fingerprint against commonly used fingerprints using a well-established benchmark set of 88 targets. We show that the CAS fingerprint outperforms most of the commonly used molecular fingerprints. Analysis of the CAS fingerprint reveals that experts tend to select features that are rarely reported in the literature, though not all rare features are selected. Our analysis also shows that the CAS fingerprint provides a different source of information compared to other commonly used fingerprints. These results suggest that anthropomorphic insights do have predictive power and highlight the importance of a chemist-in-the-loop approach in the era of machine learning.

Validating the validation: reanalyzing a large-scale comparison of deep learning and machine learning models for bioactivity prediction.

Machine learning methods may have the potential to significantly accelerate drug discovery. However, the increasing rate of new methodological approaches being published in the literature raises the fundamental question of how models should be benchmarked and validated. We reanalyze the data generated by a recently published large-scale comparison of machine learning models for bioactivity prediction and arrive at a somewhat different conclusion. We show that the performance of support vector machines is competitive with that of deep learning methods. Additionally, using a series of numerical experiments, we question the relevance of area under the receiver operating characteristic curve as a metric in virtual screening. We further suggest that area under the precision-recall curve should be used in conjunction with the receiver operating characteristic curve. Our numerical experiments also highlight challenges in estimating the uncertainty in model performance via scaffold-split nested cross validation.

Development and Testing of the OPLS-AA/M Force Field for RNA.

Significant improvements have been made to the OPLS-AA force field for modeling RNA. New torsional potentials were optimized based on density functional theory (DFT) scans at the ωB97X-D/6-311++G(d,p) level for potential energy surfaces of the backbone α and γ dihedral angles. In combination with previously reported improvements for the sugar puckering and glycosidic torsion terms, the new force field was validated through diverse molecular dynamics simulations for RNAs in aqueous solution. Results for dinucleotides and tetranucleotides revealed both accurate reproduction of 3 J couplings from NMR and the avoidance of several unphysical states observed with other force fields. Simulations of larger systems with noncanonical motifs showed significant structural improvements over the previous OPLS-AA parameters. The new force field, OPLS-AA/M, is expected to perform competitively with other recent RNA force fields and to be compatible with OPLS-AA models for proteins and small molecules.

Genetic Background, Maternal Age, and Interaction Effects Mediate Rates of Crossing Over in Drosophila melanogaster Females.

Meiotic recombination is a genetic process that is critical for proper chromosome segregation in many organisms. Despite being fundamental for organismal fitness, rates of crossing over vary greatly between taxa. Both genetic and environmental factors contribute to phenotypic variation in crossover frequency, as do genotype-environment interactions. Here, we test the hypothesis that maternal age influences rates of crossing over in a genotypic-specific manner. Using classical genetic techniques, we estimated rates of crossing over for individual Drosophila melanogaster females from five strains over their lifetime from a single mating event. We find that both age and genetic background significantly contribute to observed variation in recombination frequency, as do genotype-age interactions. We further find differences in the effect of age on recombination frequency in the two genomic regions surveyed. Our results highlight the complexity of recombination rate variation and reveal a new role of genotype by maternal age interactions in mediating recombination rate.

Population genomic analysis reveals no evidence for GC-biased gene conversion in Drosophila melanogaster.

Gene conversion is the nonreciprocal exchange of genetic material between homologous chromosomes. Multiple lines of evidence from a variety of taxa strongly suggest that gene conversion events are biased toward GC-bearing alleles. However, in Drosophila, the data have largely been indirect and unclear, with some studies supporting the predictions of a GC-biased gene conversion model and other data showing contradictory findings. Here, we test whether gene conversion events are GC-biased in Drosophila melanogaster using whole-genome polymorphism and divergence data. Our results provide no support for GC-biased gene conversion and thus suggest that this process is unlikely to significantly contribute to patterns of polymorphism and divergence in this system.