Quantum Mechanical-Based Strategies in Drug Discovery

Quantum mechanical (QM) methods provide a robust and versatile framework for exploring chemical reactivity in biomolecular systems. Over the past decades, these methods have permeated drug discovery, offering chemically accurate properties crucial for drug design. Despite the high computational cost, the development of novel algorithms, the combination of QM with machine learning (ML), and advances in computer performance have enabled the widespread adoption of QM-based strategies. This review summarizes recent advancements in QM-based methods for characterizing bioactive species, structure-guided hit-to-lead optimization, and identifying molecular features of bioactivity.

Conformational Sampling of Ligands

Conformational sampling is vital for understanding the physicochemical properties of drug-like compounds. By assessing the torsional preferences and identifying low-energy rotamers, researchers can rationalize the activities of these compounds. Generating accessible conformer ensembles helps identify bioactive species, define pharmacophore models, and assist in 3D ligand-based screenings. The analysis of conformer populations in different solvent environments is crucial for predicting the solubility and permeability of compounds beyond the rule-of-5. QM methods, such as Density Functional Theory (DFT) combined with implicit solvation calculations, are employed to discern the propensities of small molecules to adopt specific structural elements.

Key Achievements:

• Development of dedicated software like the CREST program for efficient conformational exploration.
• Implementation of ML tools trained with accurate QM data, such as the Auto3D package, to generate stereoisomeric conformational spaces.
• Introduction of deep neural network models like TorsionNet to predict torsion energy profiles with QM-level accuracy.

Refinement of Experimental Binding Poses

Defining the ligand pose within the binding pocket is crucial for medicinal chemistry studies. QM and QM/MM methods are well-suited to refine ligand geometry under the protein environment's electrostatic field, improving the description of internal parameters and intermolecular distances. This approach assists in distinguishing native poses from decoys, enhancing the reliability of drug binding predictions.

Notable Examples:

• Identification of productive inhibitor binding orientation in fatty acid amide hydrolase (FAAH) using QM/MM modeling.
• Elucidation of covalent adduct formation mechanisms between dipeptidyl nitriles and cruzain.

Advanced Techniques:

• The Quantum Mechanical Restraints (QMR) procedure for optimizing ligand geometry during crystallographic refinement.
• Automated Fragmentation QM/MM (AF-QM/MM) method for predicting protein-ligand binding structures based on NMR data.

Prediction of Binding Affinity

Accurately predicting binding free energy between a ligand and its target remains a significant challenge in drug discovery. Various QM methods, ranging from semiempirical calculations to DFT and coupled-cluster calculations, have been employed to develop robust models.

Innovative Scoring Functions:

• SQM2.20: A semiempirical QM-based scoring function that combines gas-phase interaction energy, solvation free energy changes, conformational contributions, proton transfer free energy, and ligand entropy loss.
• SophosQM: Utilizes the fragment molecular orbital (FMO) method to compute interaction energy and logP for binding affinity predictions.

Alternative Strategies:

• Indirect QM/MM approaches for correcting binding free energy determined from classical simulations.
• Development of tailored force fields like ARROW to improve the accuracy of electrostatic and exchange interactions in ligand-target binding simulations.

Molecular Determinants of Bioactivity

Understanding the structural and chemical features that determine a ligand's bioactivity profile is essential for rational drug design. QM methods offer insights into specific interactions that MM force fields cannot accurately assess, such as halogen bonds, polarized CH/π, and peptide amide-π interactions. These methods also help identify reactivity indexes for covalent inhibitors, aiding in structure-activity relationship studies.

Applications:

• Use of QM-derived properties, such as reactivity indexes, in virtual screening campaigns to prioritize top-ranked compounds.
• Identification of protein-protein interaction hotspots using QM FMO calculations for designing protein-protein modulators.

Datasets and ML Integration:

• Development of datasets like QMugs, containing QM properties of over 665,000 molecules, to facilitate ML model training.
• Integration of QM data with ML models, such as Chemprop, for predicting molecular properties like octanol/water partition coefficients.

Conclusion

The success of QM-based strategies in drug discovery hinges on continuous advancements in both algorithms and hardware. Current efforts focus on developing accurate semiempirical methods, refining QM/MM methods, and generating QM-assisted ML models. The advent of novel supercomputing architectures and quantum computing promises significant enhancements in QM simulations. These advancements, coupled with algorithmic progress, suggest an exciting future for QM-based drug discovery.