QSAR modeling and chemical space analysis of antimalarial compounds

Abstract

Generative topographic mapping (GTM) has been used to visualize and analyze the chemical space of antimalarial compounds as well as to build predictive models linking structure of molecules with their antimalarial activity. For this, a database, including ~3000 molecules tested in one or several of 17 anti-Plasmodium activity assessment protocols, has been compiled by assembling experimental data from in-house and ChEMBL databases. GTM classification models built on subsets corresponding to individual bioassays perform similarly to the earlier reported SVM models. Zones preferentially populated by active and inactive molecules, respectively, clearly emerge in the class landscapes supported by the GTM model. Their analysis resulted in identification of privileged structural motifs of potential antimalarial compounds. Projection of marketed antimalarial drugs on this map allowed us to delineate several areas in the chemical space corresponding to different mechanisms of antimalarial activity. This helped us to make a suggestion about the mode of action of the molecules populating these zones.

AntiMalarial Mode of Action (AMMA) Database: Data Selection, Verification and Chemical Space Analysis

“…In addition to cross‐validation, the maps are exposed to the validation of their competence to predict the compounds that are active against the Plasmodium parasite. Previously, we have built a SAR‐dedicated map specifically for this purpose, and it has shown robust classification performances, on par with classical SVM models . For the four current maps, this truly ‘external’ validation is more challenging, because, unlike in previous work, they are specialized in the separation of compounds by their respective target and were never presented with anti‐ Plasmodium SAR information at their construction stage.…”

“…This paper presents our to‐date effort of collecting and curating of a robust set of 15461 compounds tested against malaria parasite, with both activity and MoA annotations. In continuation of the work already published, using the so‐far curated data at the corresponding moments in time, we extended here the scope of our collection to include mechanistic annotations, as far as available. On the occasion of this publication, we also wish to open access to our curated and annotated compound collection, given as Excel spreadsheet in Supplementary Material.…”

A Chemographic Audit of Anti-Coronavirus Structure-Activity Information from Public Databases (ChEMBL)

“…it results in many very small compound sets labeled as "active", but each label has ultimately a different meaning. That is detrimental [24] to statistics-based chemoinformatics methods and machine learning, as the small sets cannot be merged (too heterogeneous), nor used individually (too small).…”

Diversifying chemical libraries with generative topographic mapping

“…In turn, the entire data set can be characterized by a vector of cumulative responsibilities. This enables the user to perform an efficient data sets comparison as well as QSAR/QSPR studies [10,11,19].…”

“…Responsibility patterns (RP) have been used to identify the shared underlying features (scaffolds, substructures, pharmacophore patterns) for a chosen area on the map [19,30]. Compounds sharing a same RP will typically share some common structural features that are further manually processed to annotate the map.…”