Integrated Molecular Modeling and Machine Learning for Drug Design

Abstract

Modern therapeutic development often involves several stages that are interconnected, and multiple iterations are usually required to bring a new drug to the market. Computational approaches have increasingly become an indispensable part of helping reduce the time and cost of the research and development of new drugs. In this Perspective, we summarize our recent efforts on integrating molecular modeling and machine learning to develop computational tools for modulator design, including a pocket-guided rational design approach based on AlphaSpace to target protein−protein interactions, delta machine learning scoring functions for protein−ligand docking as well as virtual screening, and state-of-the-art deep learning models to predict calculated and experimental molecular properties based on molecular mechanics optimized geometries. Meanwhile, we discuss remaining challenges and promising directions for further development and use a retrospective example of FDA approved kinase inhibitor Erlotinib to demonstrate the use of these newly developed computational tools.

Towards the “Eldorado” of pKa Determination: A Reliable and Rapid DFT Model

“…Indeed, it was previously demonstrated that coarse models of the first solvation shell could cause huge inaccuracy in determining pKa [ 5 ]. Further, several attempts were made in re-shaping the “reaction centre” with encouraging results [ 1 , 2 , 3 , 4 , 5 , 8 ]. Therefore, to better understand if the difference in pKa values obtained with different functionals could be related with a significant modification of the solvation cavity, an extensive analysis was performed.…”

“…The computational determination of pKa is usually performed through several methodologies, varying from the pioneering “machine learning” approach, to the more traditional physicochemical-inspired models. The former is focused on the quantitative structure–activity relationship (thereafter QSAR), but although it generally requires huge computational costs, it has not yet shown good reliability [ 4 ]. The latter relies on solid computational methods [ 5 , 6 ] that can be classified into two categories: (i) the “indirect” approach, which exploits a thermodynamic cycle that takes into consideration the energy at equilibrium of the deprotonated species in the gas and in the solution phase; (ii) the “direct” approach, in which the ionogenic equation in water is explicitly considered [ 5 , 6 ].…”

MolFeSCue: enhancing molecular property prediction in data-limited and imbalanced contexts using few-shot and contrastive learning

“…Its success underscores the need for further research on MolFeSCues compatibility with a variety of pretrained models, including those from different paradigms. Moreover, the integration of molecular dynamics simulations and machine learning, as evidenced by recent advances in computational tools for drug design, suggests an alternative avenue for improving model accuracy in specific scenarios ( Xia et al 2023 ). These methods combine physical principles with machine learning, and may provide superior outcomes by enhancing small molecule representation learning and efficient exploration of chemical space.…”

Harvesting Chemical Understanding with Machine Learning and Quantum Computers

“…We have observed a skyrocketing increase of ML applications in theoretical and computational chemistry in the past decade, − involving all space-time domains in Figure a. To most people, applying ML to theoretical and computational chemistry is merely taking advantage of a new tool to expediate the simulation and improve the accuracy.…”

“…In the recent theoretical and computational chemistry literature, − we have witnessed a gigantic growth of applications of artificial intelligence, machine learning (ML), and deep learning (hereafter, we do not distinguish these terminologies from each other and generally refer to them as ML). We also started noticing booming theoretical and computational chemistry publications using quantum computers (QC). − These newly developed methodologies are fascinating, and their impacts could be far-reaching.…”