Predicting Strategies for Lead Optimization via Learning to Rank

Abstract

Lead optimization is an essential step in drug discovery in which the chemical structures of compounds are modified to improve characteristics such as binding affinity, target selectivity, physicochemical properties, and toxicity. We present a concept for a computational compound optimization system that outputs optimized compounds from hit compounds by using previous lead optimization data from a pharmaceutical company. In this study, to predict the drug-likeness of compounds in the evaluation function of this system, we evaluated and compared the ability to correctly predict lead optimization strategies through learning to rank methods.

MolSHAP: Interpreting Quantitative Structure–Activity Relationships Using Shapley Values of R-Groups

“…With recent advances in machine learning-aided drug discovery, , many in silico methods have been proposed to help accelerate the drug discovery process due to their prediction power with high efficiency and low cost compared with wet-lab experiments, such as virtual screening, − compound property prediction, − molecule generation, − and molecule optimization. − There are also machine learning-based QSAR models developed for analyzing and predicting compound activities. − …”

libmolgrid: Graphics Processing Unit Accelerated Molecular Gridding for Deep Learning Applications

“…[6][7][8][9][10][11] In particular, machine learning has emerged as a novel area of study that holds great promise for unprecedented improvements in predictive capabilities for such problems as virtual screening, 12 binding affinity prediction, 13,14 pose prediction, 15,16 and lead optimization. [17][18][19][20] The representation of input data can fundamentally limit or enhance the performance and applicability of machine learning algorithms. [21][22][23] Standard approaches to data representation include performing initial feature selection based on various types of molecular descriptors/fingerprints, [24][25][26] including simple molecular properties, 27,28 molecular connectivity and shape, 29,30 electro-topological state, [31][32][33][34] quantum chemical properties, 35 and geometrical properties 31 (or a combination of multiple of these descriptor categories 36,37 ); summarizing inputs using representations that are amenable to direct algorithmic analysis while preserving as much relevant information as possible, such as pairwise distances between all or selected atom groups, 13,[38][39][40] using Coulomb matrices or representations derived from them, 41,42 or encoding information about local atomic environments that comprise a molecule;…”

“…Deep learning has emerged as an important area of research in computational chemistry. It holds great promise for unprecedented improvements in predictive capabilities for such problems as virtual screening, binding affinity prediction, , pose prediction, , and lead optimization. − The representation of input data can fundamentally limit or enhance the performance and applicability of machine learning algorithms. − Deep learning can derive class-defining features directly from training examples. Common input representations include molecular formats like SMILES and/or InChi strings, , molecular graphs, ,,− and voxelized spatial grids , representing the locations of atoms.…”