KINOTECA: A Navigator of Chemical Datasets with Reported Kinase Inhibitory Activity

Abstract

Protein kinases are the most intensively targeted protein family for small-molecule drugs, and the large volume of inhibitory-assay data reported for them makes the family particularly well suited to data-driven and machine-learning approaches to drug discovery. This data, however, is scattered across heterogeneous assays, reports and publications, and its reuse for modeling requires substantial curation. Here, we present Kinoteca, a curated database of kinase inhibitory-activity data derived from ChEMBL and accessible through a web interface for browsing, filtering, visualization and download. Kinoteca reconciles the multiple raw independent measurements reported for each combination of compound, kinase and activity type into a single fused activity value, together with a dispersion score based on the mean unsigned error (MUE) that quantifies how consistent the underlying measurements were. The database currently organizes curated data for 585 kinases and arranges them along biologically meaningful groupings such as kinome family, pathway and source organism. Curated activities can be filtered by the physicochemical properties of the assayed molecules or by measurement quality, explored through summary statistics, activity distributions and structural clustering, and exported either as curated or semi-raw data tables or as molecular structure files. These data formats are suitable for the direct training of machine learning models, such as activity prediction models. By delivering reconciled, analysis-ready data with its provenance preserved, Kinoteca aims to lower the barrier to reproducible, data-driven kinase drug discovery.