Articles
Technical deep-dives on molecular screening workflows, benchmark methodology, and decision-making under uncertainty.
Technical deep-dives on molecular screening workflows, benchmark methodology, and decision-making under uncertainty.
Modern computational science is betting heavily on scale: more data, more parameters, more GPUs. FluxMateria is following a different path — using physics-native computation to move closer to the mechanisms beneath chemistry, materials, and life sciences.
FluxMateria is opening external validation tracks so researchers can choose blind datasets, define metrics, and score frozen predictions independently.
FluxMateria reaches area under receiver operating characteristic curve (AUROC) 0.9597 on the comparable Therapeutics Data Commons (TDC) binary drug-induced liver injury (DILI) task versus the MiniMol public reference around 0.956, while returning exposure, mechanism, dose, confidence, and score-trace outputs for enterprise safety review.
A unified mechanism-aware ADMET pipeline that replaces the 4–6-tool stitched stack pharma teams currently maintain. Three strict #1 SOTA endpoints, plus DILI AUROC 0.9597 on the comparable TDC binary task and Caco-2 MAE 0.277 matching the public TDC reference SOTA from pure physics. Annualized TCO materially below Schrödinger, Simulations Plus, and in-house ML.
313 materials × 16 properties, end-to-end. 1.17% MAPE on family holdout — 9 to 31 times more accurate than AFLOW, JARVIS, and MatBench on the same hard split. Annualized TCO materially below DFT HPC, in-house ML, and stitched commercial stacks.
A public explanation of why splitting absorber choice, contact engineering, and build planning across separate workflows is hard to defend, and why FluxMateria kept the whole solar decision structure in one local run.
A public comparison of current battery workflows, where market solutions help, and why FluxMateria handled scoring, transport, interphase, degradation, uncertainty, and build handoff in one local pass.
One local FluxMateria workflow produced four different decision winners, a prototype handoff, and literature-backed convergence on real battery-material families.
707 candidates, 48 strict passes, two BACE1-native chemotypes, and a full white paper. Why this case study matters, and what FluxMateria actually did before wet-lab validation begins.
When screening costs drop to near zero, the order of operations changes. The economics of drug discovery shift with it.
How to combine ADMET screening with spec-driven candidate discovery for efficient lead triage.
A practical walkthrough of materials property screening and candidate prioritization.
Why capturing decision context matters and how to use decision packets effectively.
A standard PBE screening baseline (GPAW, 200 eV, 6³ k-points) vs FluxMateria on 15 canonical materials. Lattice 0.1% median off experiment, band gap 7.6% MAPE vs PBE 45.1%, magnetic moment 3.6% vs PBE 9.0%, all from chemical formula alone, at ~25,000× the per-material wall time.
How FLUX Theory calculates ΔH for any chemical equation using bond energies and Hess’s law, achieving 3.5% MAPE across 157 reactions.
A practical guide for R&D teams evaluating computational screening tools. No sales pitch — just trade-offs.
A guide to interpreting confidence signals and avoiding common pitfalls.
Understanding why pure ML models struggle with novel chemistry and what to do about it.
How to evaluate screening tools and avoid misleading performance claims. Seven questions to ask any vendor.
The reasoning behind our validation approach and what it tells you about prediction reliability.
Honest assessment of where FluxMateria works well and where to be cautious.
Follow our work or get in touch to discuss how FluxMateria fits your workflow.