Cheminformatics tools for analyzing and designing optimized small molecule libraries

Abstract

Libraries of highly annotated small molecules have many uses in chemical genetics, drug discovery and drug repurposing. Many such libraries have become available, but few data-driven approaches exist to compare these libraries and design new ones. In this paper, we describe such an approach that makes use of data on binding selectivity, target coverage and induced cellular phenotypes as well as chemical structure and stage of clinical development. We implement the approach as R software and a Web-accessible tool (http://www.smallmoleculesuite.org) that uses incomplete and often confounded public data in combination with user preferences to score and create libraries. Analysis of six kinase inhibitor libraries using our approach reveals dramatic differences among them, leading us to design a new LSP-OptimalKinase library that outperforms all previous collections in terms of target coverage and compact size. We also assemble a mechanism of action library that optimally covers 1852 targets of the liganded genome. Using our tools, individual research groups and companies can quickly analyze private compound collections and public libraries can be progressively improved using the latest data.

Pharmacological approaches to understanding protein kinase signaling networks

“…Developments in computational approaches are also likely to be key to unlocking the potential of pharmacological approaches to understand protein kinase signaling networks. Efforts to rationalize the inhibitor libraries used for phenotypic studies, either to create libraries of high selectivity for each target kinase (Moret et al, 2019;Wells et al, 2021) or suitable for deconvolution approaches (Gujral et al, 2014b;Rata et al, 2020;Watson et al, 2020;Golkowski et al, 2023), will help optimize the trade-off between library size and experimental practicality. Further work should help to determine which algorithms are most effective for identifying relevant kinases and pathways from inhibitor-induced perturbations in the phosphoproteome, for target deconvolution based on phenotypic screens, and for rational design of polypharmacological agents and drug combinations (Gujral et al, 2014b;Hernandez-Armenta et al, 2017;Tang, 2017;Rocca and Kholodenko, 2021) Finally, machine learning is poised to provide notable advances in determining features of kinase networks that predict drug efficacy for personalized medicine, as well as in these other areas.…”

“…Methods that use kinase inhibitor libraries with the aim of mimicking one-agent-one-kinase genetic screens allow more acute inhibition of kinases and, with appropriate experimental design, can lower (but not eliminate) the potential for indirect effects. For such approaches to be broadly applicable, libraries with a high coverage of the kinome and the most selective inhibitors possible for each kinase must be assembled, and tools to facilitate design of such libraries have been developed (Drewry et al, 2017;Moret et al, 2019). An example of such an open source library is the kinase chemogenomic set (KCGS) assembled by the Structural Genomics Consortium (Elkins et al, 2016;Wells et al, 2021).…”

Identification of Hsp90 inhibitors as potential drugs for the treatment of TSC1/TSC2 deficient cancer

“…Rapamycin, an mTORC1 allosteric inhibitor, did not achieve IC50 over this dose range (S2 previous studies (39). This may relate to its intrinsic potency, or to its additional inhibitory effects on DNA-PKcs, PIK3C, PIK3R, and PI4KB (36). The other mTOR inhibitors studied also have inhibitory effects on other kinases (S1 Table) (37)(38)(39).…”

Systematic Chemogenetic Library Assembly

“…The computational approach was extremely powerful for the initial pre-2014 library assembly (Figure 2A) by large-scale identification and ranking of potential tool compounds, but was limited in the dependence on accessible, warehoused data. Identifying tool compounds from diverse and heterogeneous bioactivity databases was challenging from strictly an informatics computational library design (Moret et al, 2019). Aside from incorrect or inconcise chemical structure assignments, assay target annotations may also be incorrect or misleading (e.g., a protein that is an assay readout or stimulant may be annotated as the assay target) (Tang et al, 2018).…”