Multi-task generative topographic mapping in virtual screening

Abstract

The previously reported procedure to generate "universal" Generative Topographic Maps (GTMs) of the drug-like chemical space is in practice a multi-task learning process, in which both operational GTM parameters (example: map grid size) and hyperparameters (key example: the molecular descriptor space to be used) are being chosen by an evolutionary process in order to fit/select "universal" GTM manifolds. After selection (a one-time task aimed at optimizing the compromise in terms of neighborhood behavior compliance, over a large pool of various biological targets), for any further use the manifolds are ready to provide "fit-free" predictive models. Using any structure-activity set-irrespectively whether the associated target served at map fitting stage or not-the generation or "coloring" a property landscape enables predicting the property for any external molecule, with zero additional fitable parameters involved. While previous works have signaled the excellent behavior of such models in aggressive threefold cross-validation assessments of their predictive power, the present work wished to explore their behavior in Virtual Screening (VS), here simulated on hand of external DUD ligand and decoy series that are fully disjoint from the ChEMBL-extracted landscape coloring sets. Beyond the rather robust results of the universal GTM manifolds in this challenge, it could be shown that the descriptor spaces selected by the evolutionary multi-task learner were intrinsically able to serve as an excellent support for many other VS procedures, starting from parameter-free similarity searching, to local (target-specific) GTM models, to parameter-rich, nonlinear Random Forest and Neural Network approaches.

Diversifying chemical libraries with generative topographic mapping

“…The most of interest represented some zones exclusively populated by AMS compounds. The latter were extracted and profiled using universal GTMs described in our previous papers [14,15]. To this purpose, the publicly available virtual screening webserver of the Laboratory of Chemoinformatics (http://infoc him.u-stras bg.fr/webse rv/ VSEng ine.html) was employed.…”

“…The developed tool was used for enrichment of the in-house collection of Boehringer Ingelheim (further on referred to as the "BI Pool") by the compounds from the commercial Aldrich-Market Select (AMS) database. A drug-likeness and an activity profile of selected AMS compounds against 749 biological targets were assessed using the ChEMBL data-driven predictor based on Universal GTMs [14,15].…”

A Chemographic Audit of anti‐Coronavirus Structure‐activity Information from Public Databases (ChEMBL)

“…They were hence used for antiviral chemical space analysis and antiviral compound repurposing, as already mentioned in Introduction. Here, all the seven [16,20] UMs -each based on different descriptor spaces, capturing complementary chemical informationwere used for quantitative assessments (vide infra) but most of the displayed landscapes were shown on UM#1 (the one of best average predictive propensity over the battery of selection targets).…”

Parallel Generative Topographic Mapping: An Efficient Approach for Big Data Handling

“…[14] This was estimated in terms of threefold cross-validated classification challenges of active versus inactive compounds associated with a large profile of ChEMBL biological targets, following the "universal map" paradigm. [18,21,28] Within the cross-validation procedure, a target-specific data set was split into three folds, and a GTM class landscape (not a manifold) was trained on two folds and evaluated by the third one. Balanced Accuracy (BA) was applied in this study to assess the predictive performance of the maps.…”

“…Thus, a question on the optimal FS size arises. In the previous studies, [18,27,28] the size of the FS was either optimized by the Genetic Algorithm [29] (GA) as one of the hyper-parameters of the GTM model or specified manually based on the researcher's experience. [17] Intuitively, one can assume that a larger chemical collection may need a larger FS to represent a given chemical space, whereas the GA was often selecting FSs of few thousands (5 K-25 K structures).…”