Diverse Hits in De Novo Molecule Design: Diversity-Based Comparison of Goal-Directed Generators.

Since the rise of generative AI models, many goal-directed molecule generators have been proposed as tools for discovering novel drug candidates. However, molecule generators often produce highly similar molecules and tend to overemphasize conformity to an imperfect scoring function rather than capturing the true underlying properties sought. We rectify these two shortcomings by offering diversity-based evaluations using the #Circles metric and considering constraints on scoring function calls or computation time. Our findings highlight the superior performance of SMILES-based autoregressive models in generating diverse sets of desired molecules compared to graph-based models or genetic algorithms.

DrugEx: Deep Learning Models and Tools for Exploration of Drug-Like Chemical Space.

The discovery of novel molecules with desirable properties is a classic challenge in medicinal chemistry. With the recent advancements of machine learning, there has been a surge of de novo drug design tools. However, few resources exist that are user-friendly as well as easily customizable. In this application note, we present the new versatile open-source software package DrugEx for multiobjective reinforcement learning. This package contains the consolidated and redesigned scripts from the prior DrugEx papers including multiple generator architectures, a variety of scoring tools, and multiobjective optimization methods. It has a flexible application programming interface and can readily be used via the command line interface or the graphical user interface GenUI. The DrugEx package is publicly available at https://github.com/CDDLeiden/DrugEx.

Large-Scale Modeling of Sparse Protein Kinase Activity Data.

Protein kinases are a protein family that plays an important role in several complex diseases such as cancer and cardiovascular and immunological diseases. Protein kinases have conserved ATP binding sites, which when targeted can lead to similar activities of inhibitors against different kinases. This can be exploited to create multitarget drugs. On the other hand, selectivity (lack of similar activities) is desirable in order to avoid toxicity issues. There is a vast amount of protein kinase activity data in the public domain, which can be used in many different ways. Multitask machine learning models are expected to excel for these kinds of data sets because they can learn from implicit correlations between tasks (in this case activities against a variety of kinases). However, multitask modeling of sparse data poses two major challenges: (i) creating a balanced train-test split without data leakage and (ii) handling missing data. In this work, we construct a protein kinase benchmark set composed of two balanced splits without data leakage, using random and dissimilarity-driven cluster-based mechanisms, respectively. This data set can be used for benchmarking and developing protein kinase activity prediction models. Overall, the performance on the dissimilarity-driven cluster-based split is lower than on random split-based sets for all models, indicating poor generalizability of models. Nevertheless, we show that multitask deep learning models, on this very sparse data set, outperform single-task deep learning and tree-based models. Finally, we demonstrate that data imputation does not improve the performance of (multitask) models on this benchmark set.

Accurate prediction of hydration free energies and solvation structures using molecular density functional theory with a simple bridge functional.

This paper assesses the ability of molecular density functional theory to predict efficiently and accurately the hydration free energies of molecular solutes and the surrounding microscopic water structure. A wide range of solutes were investigated, including hydrophobes, water as a solute, and the FreeSolv database containing 642 drug-like molecules having a variety of shapes and sizes. The usual second-order approximation of the theory is corrected by a third-order, angular-independent bridge functional. The overall functional is parameter-free in the sense that the only inputs are bulk water properties, independent of the solutes considered. These inputs are the direct correlation function, compressibility, liquid-gas surface tension, and excess chemical potential of the solvent. Compared to molecular simulations with the same force field and the same fixed solute geometries, the present theory is shown to describe accurately the solvation free energy and structure of both hydrophobic and hydrophilic solutes. Overall, the method yields a precision of order 0.5 kBT for the hydration free energies of the FreeSolv database, with a computer speedup of 3 orders of magnitude. The theory remains to be improved for a better description of the H-bonding structure and the hydration free energy of charged solutes.

Predicting Hydration Free Energies of the FreeSolv Database of Drug-like Molecules with Molecular Density Functional Theory.

We assess the performance of molecular density functional theory (MDFT) to predict hydration free energies of the small drug-like molecules benchmark, FreeSolv. The MDFT in the hypernetted chain approximation (HNC) coupled with a pressure correction predicts experimental hydration free energies of the FreeSolv database within 1 kcal/mol with an average computation time of 2 cpu·min per molecule. This is the same accuracy as for simulation-based free energy calculations that typically require hundreds of cpu·h or tens of gpu·h per molecule.

Pressure correction for solvation theories.

Liquid state theories such as integral equations and classical density functional theory often overestimate the bulk pressure of fluids because they require closure relations or truncations of functionals. Consequently, the cost to create a molecular cavity in the fluid is no longer negligible, and those theories predict incorrect solvation free energies. We show how to correct them simply by computing an optimized Van der Walls volume of the solute and removing the undue free energy to create such volume in the fluid. Given this versatile correction, we demonstrate that state-of-the-art solvation theories can predict, within seconds, hydration free energies of a benchmark of small neutral drug-like molecules with the same accuracy as day-long molecular simulations.

Hydration free energies and solvation structures with molecular density functional theory in the hypernetted chain approximation.

The capability of molecular density functional theory in its lowest, second-order approximation, equivalent to the hypernetted chain approximation in integral equations, to predict accurately the hydration free-energies and microscopic structure of molecular solutes is explored for a variety of systems: spherical hydrophobic solutes, ions, water as a solute, and the Mobley's dataset of organic molecules. The successes and the caveats of the approach are carefully pinpointed. Compared to molecular simulations with the same force field and the same fixed solute geometries, the theory describes accurately the solvation of cations, less so that of anions or generally H-bond acceptors. Overall, the electrostatic contribution to solvation free-energies of neutral molecules is correctly reproduced. On the other hand, the cavity contribution is poorly described but can be corrected using scaled-particle theory ideas. Addition of a physically motivated, one-parameter cavity correction accounting for both pressure and surface effects in the nonpolar solvation contribution yields a precision of 0.8 kcal/mol for the overall hydration free energies of the whole Mobley's dataset. Inclusion of another one-parameter cavity correction for the electrostatics brings it to 0.6 kcal/mol, that is, kBT. This is accomplished with a three-orders of magnitude numerical speed-up with respect to molecular simulations.

Ability of the Poisson-Boltzmann equation to capture molecular dynamics predicted ion distribution around polyelectrolytes.

Here, we examine polyelectrolyte (PE) and ion chemistry specificity in ion condensation via all-atom molecular dynamics (MD) simulations and assess the ability of the Poisson-Boltzmann (PB) equation to describe the ion distribution predicted by the MD simulations. The PB model enables the extraction of parameters characterizing ion condensation. We find that the modified PB equation which contains the effective PE radius and the energy of the ion-specific interaction as empirical fitting parameters describes ion distribution accurately at large distances but close to the PE, especially when strongly localized charge or specific ion binding sites are present, the mean field description of PB fails. However, the PB model captures the MD predicted ion condensation in terms of the Manning radius and fraction of condensed counterions for all the examined PEs and ion species. We show that the condensed ion layer thickness in our MD simulations collapses on a single master curve for all the examined simple, monovalent ions (Na+, Br+, Cs+, Cl-, and Br-) and PEs when plotted against the Manning parameter (and consequently the PE line charge density). The significance of this finding is that, contrary to the Manning radius extracted from the mean field PB model, the condensed layer thickness in the all atom detail MD modelling does not depend on the PE chemistry or counterion type. Furthermore, the fraction of condensed counterions in the MD simulations exceeds the PB theory prediction. The findings contribute toward understanding and modelling ion distribution around PEs and other charged macromolecules in aqueous solutions, such as DNA, functionalized nanotubes, and viruses.

Artificial intelligence in multi-objective drug design.

The factors determining a drug's success are manifold, making de novo drug design an inherently multi-objective optimisation (MOO) problem. With the advent of machine learning and optimisation methods, the field of multi-objective compound design has seen a rapid increase in developments and applications. Population-based metaheuris-tics and deep reinforcement learning are the most commonly used artificial intelligence methods in the field, but recently conditional learning methods are gaining popularity. The former approaches are coupled with a MOO strat-egy which is most commonly an aggregation function, but Pareto-based strategies are widespread too. Besides these and conditional learning, various innovative approaches to tackle MOO in drug design have been proposed. Here we provide a brief overview of the field and the latest innovations.

Simple Parameter-Free Bridge Functionals for Molecular Density Functional Theory. Application to Hydrophobic Solvation.

Computer simulations have been fundamental in understanding the fine details of hydrophobic solvation and hydrophobic interactions. Alternative approaches based on liquid-state theories have been proposed, but are not yet at the same degree of completeness and accuracy. In this vein, a classical, molecular density functional theory approach to hydrophobic solvation is introduced. The lowest, second-order approximation of the theory, equivalent to the hypernetted chain approximation in integral equations, fails in describing correctly cavitation free-energies. It is corrected here by two simple, angular-independent, so-called bridge functionals; they are parameter-free in the sense that all variables can be fixed unambiguously from the water bulk properties, including pressure, isothermal compressibility, and liquid-gas surface tension. A hard-sphere bridge functional, based on the known functional of a reference hard fluid system, turns out to face strong limitations for water. A simpler weighted density approximation is shown to properly reproduce the solvation free energy of hydrophobes of various sizes, from microscopic ones to the nanoscale, and predicting the solvation free energy of a data set of more than 600 model hydrophobic molecules having a variety of shapes and sizes with an accuracy of a quarter of kBT compared to Monte Carlo simulations values. It constitutes an excellent starting point for a general functional describing accurately both hydrophobic and hydrophilic solvation, and making it possible to study nonidealized hydrophobic interactions.