Mining for Potent Inhibitors through Artificial Intelligence and Physics: A Unified Methodology for Ligand Based and Structure Based Drug Design.

Determining the viability of a new drug molecule is a time- and resource-intensive task that makes computer-aided assessments a vital approach to rapid drug discovery. Here we develop a machine learning algorithm, iMiner, that generates novel inhibitor molecules for target proteins by combining deep reinforcement learning with real-time 3D molecular docking using AutoDock Vina, thereby simultaneously creating chemical novelty while constraining molecules for shape and molecular compatibility with target active sites. Moreover, through the use of various types of reward functions, we have introduced novelty in generative tasks for new molecules such as chemical similarity to a target ligand, molecules grown from known protein bound fragments, and creation of molecules that enforce interactions with target residues in the protein active site. The iMiner algorithm is embedded in a composite workflow that filters out Pan-assay interference compounds, Lipinski rule violations, uncommon structures in medicinal chemistry, and poor synthetic accessibility with options for cross-validation against other docking scoring functions and automation of a molecular dynamics simulation to measure pose stability. We also allow users to define a set of rules for the structures they would like to exclude during the training process and postfiltering steps. Because our approach relies only on the structure of the target protein, iMiner can be easily adapted for the future development of other inhibitors or small molecule therapeutics of any target protein.

D3R grand challenge 4: blind prediction of protein-ligand poses, affinity rankings, and relative binding free energies.

The Drug Design Data Resource (D3R) aims to identify best practice methods for computer aided drug design through blinded ligand pose prediction and affinity challenges. Herein, we report on the results of Grand Challenge 4 (GC4). GC4 focused on proteins beta secretase 1 and Cathepsin S, and was run in an analogous manner to prior challenges. In Stage 1, participant ability to predict the pose and affinity of BACE1 ligands were assessed. Following the completion of Stage 1, all BACE1 co-crystal structures were released, and Stage 2 tested affinity rankings with co-crystal structures. We provide an analysis of the results and discuss insights into determined best practice methods.

D3R Grand Challenge 3: blind prediction of protein-ligand poses and affinity rankings.

The Drug Design Data Resource aims to test and advance the state of the art in protein-ligand modeling by holding community-wide blinded, prediction challenges. Here, we report on our third major round, Grand Challenge 3 (GC3). Held 2017-2018, GC3 centered on the protein Cathepsin S and the kinases VEGFR2, JAK2, p38-α, TIE2, and ABL1, and included both pose-prediction and affinity-ranking components. GC3 was structured much like the prior challenges GC2015 and GC2. First, Stage 1 tested pose prediction and affinity ranking methods; then all available crystal structures were released, and Stage 2 tested only affinity rankings, now in the context of the available structures. Unique to GC3 was the addition of a Stage 1b self-docking subchallenge, in which the protein coordinates from all of the cocrystal structures used in the cross-docking challenge were released, and participants were asked to predict the pose of CatS ligands using these newly released structures. We provide an overview of the outcomes and discuss insights into trends and best-practices.

Nanocrystal Dissolution Kinetics and Solubility Increase Prediction from Molecular Dynamics: The Case of α-, ß-, and γ-Glycine.

Nanocrystals are receiving increased attention for pharmaceutical applications due to their enhanced solubility relative to their micron-sized counterpart and, in turn, potentially increased bioavailability. In this work, a computational method is proposed to predict the following: (1) polymorph specific dissolution kinetics and (2) the multiplicative increase in the polymorph specific nanocrystal solubility relative to the bulk solubility. The method uses a combination of molecular dynamics and a parametric particle size dependent mass transfer model. The method is demonstrated using a case study of α-, ß-, and γ-glycine. It is shown that only the α-glycine form is predicted to have an increasing dissolution rate with decreasing particle size over the range of particle sizes simulated. On the contrary, γ-glycine shows a monotonically increasing dissolution rate with increasing particle size and dissolves at a rate 1.5 to 2 times larger than α- or ß-glycine. The accelerated dissolution rate of γ-glycine relative to the other two polymorphs correlates directly with the interfacial energy ranking of γ > ß > α obtained from the dissolution simulations, where γ- is predicted to have an interfacial energy roughly four times larger than either α- or ß-glycine. From the interfacial energies, α- and ß-glycine nanoparticles were predicted to experience modest solubility increases of up to 1.4 and 1.8 times the bulk solubility, where as γ-glycine showed upward of an 8 times amplification in the solubility. These MD simulations represent a first attempt at a computational (pre)screening method for the rational design of experiments for future engineering of nanocrystal API formulations.

Solubility curves and nucleation rates from molecular dynamics for polymorph prediction - moving beyond lattice energy minimization.

Current polymorph prediction methods, known as lattice energy minimization, seek to determine the crystal lattice with the lowest potential energy, rendering it unable to predict solvent dependent metastable form crystallization. Facilitated by embarrassingly parallel, multiple replica, large-scale molecular dynamics simulations, we report on a new method concerned with predicting crystal structures using the kinetics and solubility of the low energy polymorphs predicted by lattice energy minimization. The proposed molecular dynamics simulation methodology provides several new predictions to the field of crystallization. (1) The methodology is shown to correctly predict the kinetic preference for ß-glycine nucleation in water relative to α- and γ-glycine. (2) Analysis of nanocrystal melting temperatures show γ- nanocrystals have melting temperatures up to 20 K lower than either α- or ß-glycine. This provides a striking explanation of how an energetically unstable classical nucleation theory (CNT) transition state complex leads to kinetic inaccessibility of γ-glycine in water, despite being the thermodynamically preferred polymorph predicted by lattice energy minimization. (3) The methodology also predicts polymorph-specific solubility curves, where the α-glycine solubility curve is reproduced to within 19% error, over a 45 K temperature range, using nothing but atomistic-level information provided from nucleation simulations. (4) Finally, the methodology produces the correct solubility ranking of ß- > α-glycine. In this work, we demonstrate how the methodology supplements lattice energy minimization with molecular dynamics nucleation simulations to give the correct polymorph prediction, at different length scales, when lattice energy minimization alone would incorrectly predict the formation of γ-glycine in water from the ranking of lattice energies. Thus, lattice energy minimization optimization algorithms are supplemented with the necessary solvent/solute dependent solubility and nucleation kinetics of polymorphs to predict which structure will come out of solution, and not merely which structure has the most stable lattice energy.

Optimal experimental design for efficient toxicity testing in microphysiological systems: A bone marrow application.

Introduction: Microphysiological systems (MPS; organ-on-a-chip) aim to recapitulate the 3D organ microenvironment and improve clinical predictivity relative to previous approaches. Though MPS studies provide great promise to explore treatment options in a multifactorial manner, they are often very complex. It is therefore important to assess and manage technical confounding factors, to maximise power, efficiency and scalability. Methods: As an illustration of how MPS studies can benefit from a systematic evaluation of confounders, we developed an experimental design approach for a bone marrow (BM) MPS and tested it for a specified context of use, the assessment of lineage-specific toxicity. Results: We demonstrated the accuracy of our multicolour flow cytometry set-up to determine cell type and maturity, and the viability of a "repeated measures" design where we sample from chips repeatedly for increased scalability and robustness. Importantly, we demonstrated an optimal way to arrange technical confounders. Accounting for these confounders in a mixed-model analysis pipeline increased power, which meant that the expected lineage-specific toxicities following treatment with olaparib or carboplatin were detected earlier and at lower doses. Furthermore, we performed a sample size analysis to estimate the appropriate number of replicates required for different effect sizes. This experimental design-based approach will generalise to other MPS set-ups. Discussion: This design of experiments approach has established a groundwork for a reliable and reproducible in vitro analysis of BM toxicity in a MPS, and the lineage-specific toxicity data demonstrate the utility of this model for BM toxicity assessment. Toxicity data demonstrate the utility of this model for BM toxicity assessment.

Transcriptional signatures associated with persisting CD19 CAR-T cells in children with leukemia.

In the context of relapsed and refractory childhood pre-B cell acute lymphoblastic leukemia (R/R B-ALL), CD19-targeting chimeric antigen receptor (CAR)-T cells often induce durable remissions, which requires the persistence of CAR-T cells. In this study, we systematically analyzed CD19 CAR-T cells of 10 children with R/R B-ALL enrolled in the CARPALL trial via high-throughput single-cell gene expression and T cell receptor sequencing of infusion products and serial blood and bone marrow samples up to 5 years after infusion. We show that long-lived CAR-T cells developed a CD4/CD8 double-negative phenotype with an exhausted-like memory state and distinct transcriptional signature. This persistence signature was dominant among circulating CAR-T cells in all children with a long-lived treatment response for which sequencing data were sufficient (4/4, 100%). The signature was also present across T cell subsets and clonotypes, indicating that persisting CAR-T cells converge transcriptionally. This persistence signature was also detected in two adult patients with chronic lymphocytic leukemia with decade-long remissions who received a different CD19 CAR-T cell product. Examination of single T cell transcriptomes from a wide range of healthy and diseased tissues across children and adults indicated that the persistence signature may be specific to long-lived CAR-T cells. These findings raise the possibility that a universal transcriptional signature of clinically effective, persistent CD19 CAR-T cells exists.

An Analysis of Proteochemometric and Conformal Prediction Machine Learning Protein-Ligand Binding Affinity Models.

Protein-ligand binding affinity is a key pharmacodynamic endpoint in drug discovery. Sole reliance on experimental design, make, and test cycles is costly and time consuming, providing an opportunity for computational methods to assist. Herein, we present results comparing random forest and feed-forward neural network proteochemometric models for their ability to predict pIC50 measurements for held out generic Bemis-Murcko scaffolds. In addition, we assess the ability of conformal prediction to provide calibrated prediction intervals in both a retrospective and semi-prospective test using the recently released Grand Challenge 4 data set as an external test set. In total, random forest and deep neural network proteochemometric models show quality retrospective performance but suffer in the semi-prospective setting. However, the conformal predictor prediction intervals prove to be well-calibrated both retrospectively and semi-prospectively showing that they can be used to guide hit discovery and lead optimization campaigns.