Minisymposium

Molecular docking computationally predicts the conformation of a small molecule when binding to a receptor. Scoring functions are a vital piece of any molecular docking pipeline as they determine the fitness of sampled poses.We will describe the training and development of convolutional neural networks for protein-ligand scoring and how these deep learning models are integrated into the GNINA molecular docking open source software. We will describe the role high performance computing played in the training and optimzation of these networks and how high throughput docking can be performed at scale. Successful prospective evaluations of GNINA will be discussed, including recent top performance in the Critical Assessment of Computational Hit-Finding Experiments (CACHE).

Computational approaches for estimating protein-ligand binding affinities are crucial in modern drug discovery. All-atom explicit-solvent molecular dynamics (MD) simulations use the foundational principles of classical mechanics and statistical thermodynamics to rigorously calculate binding free energies. They are the most accurate affinity prediction methods that are still tractable when applied to biological systems, but they are also among the most computationally demanding. To mitigate costs, efficient sampling protocols such as multistate methods have been developed, enabling calculations of relative binding free energy across multiple ligands from a single simulation. One such technique, replica-exchange EDS (RE-EDS), leverages similarities in ligand environments to achieve improved sampling performance compared to pairwise methods. Its recent GPU-accelerated implementation in the OpenMM toolkit for molecular simulation has been benchmarked on a set of solvation free energies from the FreeSolv dataset and forms the basis of a high throughput framework for free energy calculations.

Estimating protein targets of compounds based on the similarity principle is a long-standing strategy in drug discovery. Building upon prior quantification of this principle, the large-scale assessment of its predictive power was performed using an unprecedented vast external test set of more than 300’000 active small molecules against another bioactivity set of more than 500’000 compounds. It was found that machine-learning can predict the correct targets, with the highest probability among 2069 proteins, for more than 51% of the external molecules. The strong enrichment thus obtained demonstrates its usefulness in supporting phenotypic screens, polypharmacology, or repurposing. Moreover, the impact of the bioactivity knowledge available for proteins in terms of number and diversity of actives was investigated. This study advocates for the adoption of application-oriented benchmarking strategies to prevent accidental overestimation of their predictive ability, and the use of large, high-quality, non-overlapping datasets.

In virtual drug discovery, the reproducibility of results across different practitioners is a frequent concern, and the roles of human intervention, e.g., through visual inspection, are hard to quantify and recapitulate. This is in part a result of the nature of the task, which is to find, rather than explain, promising compounds. Nevertheless, the standardization of workflows is a concern, both in general and especially so when the results of screening molecules have a stochastic component. The abstraction of tasks through sophisticated user interfaces such as Maestro (Schrodinger) is one route toward standardization, which has the additional benefit of flattening the learning curves for beginners. In my talk, I will present a unified, open-source graphical user interface developed in our group, ACGui, which serves as a frontend for different popular docking softwares. I will explain the SQL database we use for compound management, including topics such as parameterization, conformer generation, and visualization of results. ACGui also abstracts an interface to HPC resources through SLURM to tackle large-scale computations as well as the setup of bulk molecular dynamics simulations using GROMACS to study promising hits in more detail. I will briefly present the main workflows of both of these components.