Frequently Asked Questions

Everything you need to know about FluxMateria capabilities, accuracy, and how to get started.

Product

What makes FluxMateria different from ML-based screening tools?
FluxMateria uses no machine-learning training — no neural networks, no backpropagation. Predictions are grounded in molecular physics rather than statistical training. Some modules augment the Deterministic Physics Kernel with structural similarity to curated reference databases, but no model weights are ever learned. This improves extrapolation in some regimes (see Benchmarks) and provides clearer confidence signals.
Is FluxMateria faster than DFT?
Yes. The specified GPAW-PBE comparison shows acceleration of order 104 per material call, and approximately 950× for the complete 15-material batch including initialization. The comparison is a FluxMateria prediction call versus a seven-point PBE equation-of-state scan; speed depends on the workload. See the timing scope. DFT remains useful for final-stage validation.
How accurate is FluxMateria compared to DFT?
See our Benchmarks page for quantitative comparisons. Generally: accuracy suitable for screening prioritization, with clear confidence indicators when predictions are less reliable.
Do I need all modules?
The focused pilot covers one workflow. Additional workflows can be scoped separately.
What is "physics-only prediction"?
FluxMateria derives molecular and material properties from fundamental physics — specifically, from molecular geometry and physical constants. There are no trained machine-learning models and no statistical correlations. Every prediction is a deterministic calculation from first principles, which means the same input always produces the same output, and every result can be traced back to the underlying physics.
What molecules and materials are supported?
Molecules: Any organic molecule expressible as a SMILES string or SDF file. The ADMET panel covers standard drug-like chemical space (MW < 1000, organic elements). Novel scaffolds are supported but may receive lower confidence scores.

Materials: 73 elements (H through Bi, excluding lanthanides). Binary and ternary compounds are best supported. The materials engine covers 14 categories including III-V, II-VI, oxides, perovskites, chalcopyrites, and more. See Materials for the full list.
What are the known limitations?
We document limitations honestly:
  • Novel scaffolds far from validated chemical space may have lower confidence
  • Some toxicity endpoints have limited experimental validation data
  • Quaternary+ material compositions may have lower accuracy
  • Lanthanide elements are not yet supported
  • Predictions are for screening prioritization, not regulatory submission
Each module page includes a "Scope notes" section with specific limitations. See Benchmarks for quantitative accuracy data.
Can I use FluxMateria results for regulatory submissions?
FluxMateria is designed for screening and prioritization, not direct regulatory submission. Results can inform which candidates to advance to experimental validation, but regulatory filings should rely on experimental data. We provide full provenance and methodology documentation to support internal decision-making and audit trails.

Technical

What inputs does FluxMateria accept?
SMILES strings, SDF files, batch CSV/JSON, and material compositions (module-dependent). See Docs → Inputs.
How should I interpret confidence indicators?
High = reliable for this chemical space. Medium = informative but verify for critical decisions. Low = outside reliable prediction space. See individual module docs for endpoint-specific guidance.
Can I integrate FluxMateria with existing tools?
Yes. Export formats include CSV, JSON, SDF, and PDF. API access is available for programmatic integration. See Integrations.
How do I interpret low-confidence predictions?
A low-confidence prediction means the molecule or material is outside the region where FluxMateria's physics model is most reliable. This does not mean the prediction is wrong — it means you should treat it as a hypothesis rather than a screening decision. Recommended actions:
  • Do not use low-confidence results as pass/fail gates
  • Flag the compound for experimental validation
  • Check which specific endpoints are low-confidence (others may still be reliable)
  • Consider whether the molecule has unusual structural features (macrocycles, metallocenes, etc.) that may be outside current coverage
See Troubleshooting for more guidance.
What file formats are supported for input and output?
Input: SMILES strings, SDF/MOL files, batch CSV (one SMILES per row), JSON, and material compositions (element symbols + stoichiometry).

Output: CSV, JSON, SDF with annotations, and PDF decision packets with full provenance. All exports include confidence indicators and methodology metadata. See Getting Started for format details.

Demos

How can we evaluate FluxMateria?
The interactive ADMET demo requires no account — input a SMILES and see results immediately. For batch evaluation and full platform access, contact us to schedule a pilot discussion.
What's included in the email-gated demos?
Three video walkthroughs (Inverse Search, Materials, Workspace) plus an interactive batch ADMET demo where you can upload your own molecules. See Demos.

Commercial

How do pilots work?
We agree on one workflow, one dataset, and the success criteria for an upcoming R&D decision. A typical pilot takes two to three weeks and delivers property predictions or a ranked shortlist, a comparison with your current method where available, and a technical readout.
What does a pilot cost?
The focused pilot costs €5,000 as a fixed fee. We quote broader validation work, additional workflows, and annual access separately. See pricing for details.
What if the pilot doesn't prove value?
Then you've learned quickly with bounded cost. We define success metrics upfront so outcomes are unambiguous, and we provide a written recommendation for next steps.
How do academic plans work with grant funding?
Review the academic access options for individual researchers, labs, and institutions. Contact us to discuss a quote for your grant application.

Security

Is my data used to train models?
No. FluxMateria is physics-only and does not use customer data for ML/AI training, and data is not shared between customers. See Security.
What compliance certifications do you have?
We are early-stage; a SOC 2 roadmap is available on request. During pilot and enterprise evaluations, we support security questionnaires and share architecture/data-handling details needed for internal review.
Can procurement and security review run in parallel with the pilot?
Yes. We can run legal/security review in parallel with technical evaluation to reduce cycle time. Typical workflow includes NDA/DPA review, security Q&A, and a technical architecture discussion.

Still have questions?

We're happy to discuss your specific needs and answer any other questions.

Discuss a pilot Contact Us