Shared Active-Form Structures in the ATLAS Registry: Precursor Mapping and Predicted Susceptibility Profiles

This working paper (with FAIR) examines shared active-form structures in the ATLAS molecular registry and their relationship to existing predicted susceptibility profiles. It analyses the recorded correspondence between pharmacological precursor pairs (AF/CF) and assigned intracellular active-form pairs (AAF/ACF), distinguishing molecular identity, precursor multiplicity, and susceptibility agreement. Across 97 AACF entries, stereochemistry-aware SMILES normalisation identifies 95 distinct precursor pairs and 61 distinct active-form pairs. Fourteen convergence classes contain 48 AACF entries and generate 88 unordered within-class comparisons. No pure partial convergence is observed: the AAF and ACF branch groupings reproduce the complete active-pair convergence classes. The susceptibility analysis uses the supplied integrated ATLAS predictions for 1,999 cell-line records. Within convergence classes, mean six-zone agreement is approximately 47.99%, and favourable-domain agreement is 68.45%, while numerical coincidence remains rare. Conditional permutation comparisons show greater coherence than alternative groupings of the same 48 profiles with the same class sizes. All four susceptibility endpoints retain statistical support after within-family Holm adjustment. Because the observed classes align strongly with AACF family labels, this comparison does not isolate an active-form-specific contribution to coherence.Two precursor graph-topological representations—degree distributions and unweighted shortest-path distributions—are compared using Jensen–Shannon divergence. Their associations with susceptibility agreement are small, and conditional within-class permutation analyses do not establish additional positive topology–susceptibility alignment. Consequently, the joint coherence-and-topology claim is not supported by the intersection–union assessment. A sensitivity analysis excluding one extreme source value leaves the principal categorical conclusions unchanged. The accompanying computational resources include source tables, Wolfram Language notebooks and text exports, diagnostic audits, molecular mapping and class-membership tables, pairwise analytical results, permutation null distributions, a deterministic synthesis script, and a README describing execution order and analytical conventions. The findings describe structural assignments and computational predictions within the supplied ATLAS dataset. They do not establish experimentally verified intracellular transformations, pharmacological equivalence, or therapeutic efficacy. Shared active-form identity does not imply interchangeable susceptibility profiles.