We have published a new enterprise-TCO case study examining the architecture and economics of ADMET prediction across the four pathways a pharma discovery program faces today: stitched commercial ADMET stacks, in-house ML pipelines, DFT-based mechanistic ADMET, and free or sanity-check tooling.
The headline result. FluxMateria profiled 245 FDA-approved drugs across all 8 ADMET endpoints — 1,960 mechanism-aware predictions — in ~51 seconds of end-to-end wall-clock time, in full mechanistic mode (DILI exposure-aware logic, CYP isoform gating, transporter inference, and reactive-metabolite alerts all active). Three of the eight endpoints are strict #1 SOTA on the public leaderboards: aqueous solubility (logS MAE 0.06 vs MiniMol 0.741), metabolism (Spearman 0.692 vs TDC SOTA 0.536), and PPB HIGH-tier (MAE 3.65%, at the inter-laboratory experimental noise floor on a 14,288-compound leave-one-out validation). The DILI engine now reaches AUROC 0.9597 on the comparable TDC binary task versus the MiniMol reference around 0.956, while returning mechanism and exposure context rather than only a binary label. Caco-2 permeability has now also joined the SOTA tier: MAE 0.277 on the TDC caco2_wang scaffold-stratified test set, matching the public reference SOTA at 0.276 from pure physics with zero training labels consumed.
A typical lead-optimization workflow today integrates four to six commercial and internal systems — Schrödinger QikProp for one set of endpoints, Simulations Plus ADMET Predictor for another, OpenEye toolkits for a third, internal ML models for hERG and DILI, plus free tools as sanity checks. Each carries its own license, output schema, and chemical-space coverage; reconciling them into a unified decision-grade output adds non-trivial integration overhead per workflow. Multi-vendor ADMET workflows reflect the assay-by-assay history of the field rather than the structure of the underlying physics.
A unified first-principles physics model returns eight endpoints, per-prediction confidence, and mechanism-evidence trail as a single output document, with no cross-tool reconciliation required. The case study walks through the four enterprise pathways with annualized capability cost ranges and the structural limitations of each.
The case study covers operational implications across the discovery workflow: real-time integration with lead optimization (~210 ms per compound, full panel), portfolio-scale safety triage (10,000 compounds across all eight endpoints in approximately 35 minutes), mechanism-aware DILI assessment (CYP isoform attribution, transporter substrate flags, hepatic exposure context, reactive-metabolite alerts in the same call as the risk score), coverage of novel chemistry by construction (PROTACs, peptidomimetics, macrocycles handled within scope), unified output schema, and audit-grade reproducibility (frozen JSON manifests with commit hash for every screen).
The public materials include the case-study page with the eight-endpoint head-to-head accuracy table (FluxMateria vs MiniMol vs TDC SOTA, same compounds and metrics), the annualized TCO comparison across the four enterprise pathways, the technical specifications block, the previously published ADMET clinical-failures retrospective (88.2% sensitivity on 34 withdrawn drugs, zero false positives on 16 safe controls), and the publicly audited ADMET benchmark with full per-tier and per-class breakdowns.
This article describes a computational ADMET screening capability and total-cost-of-ownership analysis. It is not a regulatory sign-off, clinical efficacy claim, or replacement for in-vitro / in-vivo safety studies. TCO ranges reflect typical industry benchmarks for ongoing pharma ADMET capability and are independently verifiable from publicly cited license and personnel cost models.