ACEGEN: Reinforcement Learning of Generative Chemical Agents for Drug Discovery.

In recent years, reinforcement learning (RL) has emerged as a valuable tool in drug design, offering the potential to propose and optimize molecules with desired properties. However, striking a balance between capabilities, flexibility, reliability, and efficiency remains challenging due to the complexity of advanced RL algorithms and the significant reliance on specialized code. In this work, we introduce ACEGEN, a comprehensive and streamlined toolkit tailored for generative drug design, built using TorchRL, a modern RL library that offers thoroughly tested reusable components. We validate ACEGEN by benchmarking against other published generative modeling algorithms and show comparable or improved performance. We also show examples of ACEGEN applied in multiple drug discovery case studies. ACEGEN is accessible at https://github.com/acellera/acegen-open and available for use under the MIT license.

Elucidating the energetic contributions to the binding free energy.

New exact equations are derived for the terms contributing to the binding free energy (ΔG0) of a ligand-receptor pair using our recently introduced formalism which we here call perturbation-divergence formalism (PDF). Specifically, ΔG0 equals the sum of the average of the perturbation (pertaining to new interactions) and additional dissipative terms. The average of the perturbation includes the sum of the average receptor-ligand interactions and the average of the change of solvation energies upon association. The Kullback-Leibler (KL) divergence quantifies the energetically dissipative terms, which are due to the configurational changes and, using the chain rule of KL divergence, can be decomposed into (i) dissipation due to limiting the external liberation (translation and rotation) of the ligand relative to the receptor and (ii) dissipation due to conformational (internal) changes inside the receptor and the ligand. We also identify all exactly canceling energetic terms which do not contribute to ΔG0. Furthermore, the PDF provides a new approach towards dimensionality reduction in the representation of the association process and towards relating the dynamic (high dimensional) with the thermodynamic (one-dimensional) changes.

Interaction of calmodulin with Sec61α limits Ca2+ leakage from the endoplasmic reticulum.

In eukaryotes, protein transport into the endoplasmic reticulum (ER) is facilitated by a protein-conducting channel, the Sec61 complex. The presence of large, water-filled pores with uncontrolled ion permeability, as formed by Sec61 complexes in the ER membrane, would seriously interfere with the regulated release of calcium from the ER lumen into the cytosol, an essential mechanism for intracellular signalling. We identified a calmodulin (CaM)-binding motif in the cytosolic N-terminus of mammalian Sec61α that bound CaM but not Ca2+-free apocalmodulin with nanomolar affinity and sequence specificity. In single-channel measurements, CaM potently mediated Sec61-channel closure in Ca2+-dependent manner. At the cellular level, two different CaM antagonists stimulated calcium release from the ER through Sec61 channels. However, protein transport into microsomes was not modulated by Ca2+-CaM. Molecular modelling of the ribosome/Sec61/CaM complexes supports the view that simultaneous ribosome and CaM binding to the Sec61 complex may be possible. Overall, CaM is involved in limiting Ca2+ leakage from the ER.

PromptSMILES: prompting for scaffold decoration and fragment linking in chemical language models.

SMILES-based generative models are amongst the most robust and successful recent methods used to augment drug design. They are typically used for complete de novo generation, however, scaffold decoration and fragment linking applications are sometimes desirable which requires a different grammar, architecture, training dataset and therefore, re-training of a new model. In this work, we describe a simple procedure to conduct constrained molecule generation with a SMILES-based generative model to extend applicability to scaffold decoration and fragment linking by providing SMILES prompts, without the need for re-training. In combination with reinforcement learning, we show that pre-trained, decoder-only models adapt to these applications quickly and can further optimize molecule generation towards a specified objective. We compare the performance of this approach to a variety of orthogonal approaches and show that performance is comparable or better. For convenience, we provide an easy-to-use python package to facilitate model sampling which can be found on GitHub and the Python Package Index.Scientific contributionThis novel method extends an autoregressive chemical language model to scaffold decoration and fragment linking scenarios. This doesn't require re-training, the use of a bespoke grammar, or curation of a custom dataset, as commonly required by other approaches.

Protein translocation across the ER membrane.

Protein translocation into the endoplasmic reticulum (ER) is the first and decisive step in the biogenesis of most extracellular and many soluble organelle proteins in eukaryotic cells. It is mechanistically related to protein export from eubacteria and archaea and to the integration of newly synthesized membrane proteins into the ER membrane and the plasma membranes of eubacteria and archaea (with the exception of tail anchored membrane proteins). Typically, protein translocation into the ER involves cleavable amino terminal signal peptides in precursor proteins and sophisticated transport machinery components in the cytosol, the ER membrane, and the ER lumen. Depending on the hydrophobicity and/or overall amino acid content of the precursor protein, transport can occur co- or posttranslationally. The respective mechanism determines the requirements for certain cytosolic transport components. The two mechanisms merge at the level of the ER membrane, specifically, at the heterotrimeric Sec61 complex present in the membrane. The Sec61 complex provides a signal peptide recognition site and forms a polypeptide conducting channel. Apparently, the Sec61 complex is gated by various ligands, such as signal peptides of the transport substrates, ribosomes (in cotranslational transport), and the ER lumenal molecular chaperone, BiP. Binding of BiP to the incoming polypeptide contributes to efficiency and unidirectionality of transport. Recent insights into the structure of the Sec61 complex and the comparison of the transport mechanisms and machineries in the yeast Saccharomyces cerevisiae, the human parasite Trypanosoma brucei, and mammals have various important mechanistic as well as potential medical implications. This article is part of a Special Issue entitled Protein translocation across or insertion into membranes.

Relative Principal Components Analysis: Application to Analyzing Biomolecular Conformational Changes.

A new method termed "Relative Principal Components Analysis" (RPCA) is introduced that extracts optimal relevant principal components to describe the change between two data samples representing two macroscopic states. The method is widely applicable in data-driven science. Calculating the components is based on a physical framework that introduces the objective function (the Kullback-Leibler divergence) appropriate for quantifying the change of the macroscopic state affected by the changes in the microscopic features. To demonstrate the applicability of RPCA, we analyze the thermodynamically relevant conformational changes of the protein HIV-1 protease upon binding to different drug molecules. In this case, the RPCA method provides a sound thermodynamic foundation for analyzing the binding process and thus characterizing both the collective and the locally relevant conformational changes. Moreover, the relevant collective conformational changes can be reconstructed from the informative latent variables to exhibit both the enhanced and the restricted conformational fluctuations upon ligand association.

The Role of Conformational Changes in Molecular Recognition.

Conformational changes of molecules are crucial elements in many biochemical processes, and also in molecular recognition. Here, we present a novel exact mathematical equation for the binding free energy of a receptor-ligand pair. It shows that the energetic contribution due to conformational changes upon molecular recognition is defined by the so-called Kullback-Leibler (KL) divergence between the probability distributions of the conformational ensemble of the biomolecule in the bound and free states. We show that conformational changes always contribute positively to the change in free energy and therefore disfavor the association process. Using the example of ligands binding to a flexible cavity of T4 lysozyme, we illustrate that, due to enthalpy-entropy compensation, the conformational entropy is a misleading quantity for assessing the conformational contribution to the binding free energy, in contrast to the KL divergence, which is the correct quantity to use in this context.

Enthalpy-Entropy Compensation upon Molecular Conformational Changes.

The change in free energy is the dominant factor in all chemical processes; it usually encompasses enthalpy-entropy compensation (EEC). Here, we use the free energy perturbation formalism to show that EEC is influenced by the molecular conformational changes (CCs) of the entire system comprising the solute and by the already known solvent reorganization. The internal changes of enthalpy and the entropy due to CCs upon modifying the interactions (perturbation) cancel each other exactly. The CCs influence the dissipation of the modified interactions and their contributions to the free energy. Using molecular simulations, we show that, for solvation of six different HIV-1 protease inhibitors, CCs in the solute cause EEC as large as 10-30 kcal/mol. Moreover, the EEC due to CCs in HIV-1 protease is shown to vary significantly upon modifying its bound ligand. These findings have important implications for understanding of EEC phenomena and for interpretation of thermodynamic measurements.

How Molecular Conformational Changes Affect Changes in Free Energy.

A simple quantitative relationship between the molecular conformational changes and the corresponding changes in the free energy is presented. The change in free energy is the sum of that part of the enthalpic change that is due to the externally applied work (perturbation) and of that part of the entropic change, termed dissipative entropy, that is related to the conformational changes. The dissipative entropy is equivalent to the relative entropy, a concept from information theory, between the distributions of the conformations in the initial and the final states. The remaining change in entropy (nondissipative) cancels exactly with the remaining enthalpic change. The calculation of the dissipative entropy is demonstrated to pose the main difficulty in free energy computation. The straightforward decomposition of the dissipative entropy into contributions from different parts of the system promises to improve the understanding of the role of conformational changes in biochemical reactions.

Adhesive water networks facilitate binding of protein interfaces.

Water structure has an essential role in biological assembly. Hydrophobic dewetting has been documented as a general mechanism for the assembly of hydrophobic surfaces; however, the association mechanism of hydrophilic interfaces remains mysterious and cannot be explained by simple continuum water models that ignore the solvent structure. Here we study the association of two hydrophilic proteins using unbiased extensive molecular dynamics simulations that reproducibly recovered the native bound complex. The water in the interfacial gap forms an adhesive hydrogen-bond network between the interfaces stabilizing early intermediates before native contacts are formed. Furthermore, the interfacial gap solvent showed a reduced dielectric shielding up to distances of few nanometres during the diffusive phase. The interfacial gap solvent generates an anisotropic dielectric shielding with a strongly preferred directionality for the electrostatic interactions along the association direction.