TROOPIR scores folding, stability, variant impact, and binding compatibility on one deterministic axis, in minutes on a single CPU. No training data. No random seed on the score; the lone Monte Carlo structure step is isolated and reported separately. The fast first pass before you commit GPU-hours to FEP+ or a wet lab.
Illustrative output on the TTR system: the engine separates the FDA-approved stabilizer tafamidis from a glucose distractor by a 0.38 compatibility gap. Accuracy benchmarks against named methods are run with each evaluator under NDA.
Structure shown is the experimental crystal (PDB 3TCT), included for reference. Not engine output. The engine scores this pair at 0.87 from the protein sequence plus a standard ligand structure.
Three 42-residue sequences with identical amino-acid counts (6 A, 12 E, 12 K, 12 L): one designed, two shuffled. A model scoring on composition could not tell them apart. TROOPIR separates them by 15.7 kcal/mol, because the score depends on sequence order.
Run the same sequence twice and the score is byte-identical, SHA-256 verified. The Monte Carlo structure step is the one stochastic stage, and we report it separately rather than hide it.
This demonstrates determinism and sequence-order sensitivity, the foundation. Head-to-head accuracy benchmarks against AlphaFold and FEP+ are defined and run with each evaluator under NDA.
4a61b8fce43aed8f on both.Output is deterministic, reproducible, and produced from sequence alone on commodity compute. Identical input returns an identical score, every run.
The method is protected as trade secret and filed in three USPTO provisional patents, with full enabling disclosure deposited at each filing.
Per-residue protein stability prediction from sequence, deterministic, on one CPU. Validated against published experimental benchmarks, answers that otherwise take months of experimental work, from sequence in seconds. Head-to-head accuracy available under NDA.
Aggregation and misfolding-hotspot prediction. Flags the position-specific regions where a protein starts to aggregate.
Variant-effect prediction. Ranks mutations by predicted impact ahead of experimental characterization.
Binding-site and druggability scoring. Protein-ligand compatibility on a single deterministic axis, no docking pose search.
Drug-resistance prediction. Flags which variants escape which drugs, before clinic, before wet lab.
Every protein-ligand pair returns one deterministic compatibility score from 0 to 1, derived from first-principles, solvent-aware physics, not a machine-learned correlation against a training set. Compatible at 0.80 and above, incompatible at 0.50 and below. The 0.50 to 0.80 band requires interpretation against use-case context.
The 0.49 and 0.87 values shown are illustrative engine outputs on the TTR system. Score-to-experiment correlation on independent benchmarks is in active validation and shared with evaluators under NDA, alongside head-to-head protocols against AlphaFold, Glide, and FEP+ defined at scoping.
TROOPIR does not replace structure prediction or free-energy calculation. It runs first, fast and cheap, to decide where those are worth running.
| Stage | TROOPIR | AlphaFold 3 | FEP+ |
|---|---|---|---|
| Role | Deterministic pre-screen | Structure prediction | Affinity gold standard |
| Input | Sequence | Sequence + MSA | 3D structure + ligand |
| Training data | None | MSA-dependent | Force-field parameters |
| Score determinism | Byte-identical | Model-dependent | Stochastic sampling |
| Compute | Minutes, 1 CPU | Minutes, GPU | 12 to 24 hrs, GPU cluster |
| Use it to | Rank what to fold or simulate | Get a 3D structure | Quantify binding affinity |
Run TROOPIR first to triage. Spend the GPU-hours on the candidates it ranks worth the cost.
Three public-domain worked examples spanning the range: FDA-approved drug-target pair, comparative ranking across a drug class, and mechanism discrimination on an intrinsically disordered target. Each is reproducible from the cited sequence and ligand pair. Experimental accuracy benchmarks are delivered to evaluators under NDA.
Read the full TTR + tafamidis worked example with head-to-head comparison against AlphaFold, Vina, and FEP+, or view a sample canonical report on the same system. Public-domain target, public-domain ligand, representative engine output.
TROOPIR was built outside the academic publication cycle by design. The methodology was filed as a sequence of three USPTO provisional patents across May and June 2026, with full enabling disclosure deposited at each filing rather than held for conference timing. The mega-omnibus filing alone carries 44 independent claims across stability prediction, binding site detection, variant classification, and therapeutic compound identification.
The output framework was validated against FDA-approved drug-target pairs before any external evaluation was offered. Initial validations include TTR plus tafamidis (the Pfizer Vyndamax amyloidosis stabilizer) and the GLP-1 family of incretin therapeutics. Evaluator engagements proceed under one-way NDA. Engine capacity is deliberately limited to maintain the five-business-day output SLA per engagement.
TROOPIR is led by a single principal who is the inventor of the methodology, the sole signatory on evaluator NDAs, and the direct point of contact for every inquiry. The single-principal LLC structure is deliberate: trade-secret-protected method and patent enforcement stay under undiluted control during the commercialization phase. All inquiries are reviewed by the principal within 48 hours.
— Read the founder essay Why Deterministic on the recipe-versus-receipt distinction and why single-principal is the right structure at this stage.
How an evaluation proceeds from first contact to delivered output.
You name the protein, peptide, or compound system you intend to evaluate. Therapeutic context optional.
One-way evaluator nondisclosure agreement executed via electronic signature. Preview the draft.
Sequences and ligand structures delivered through the secure channel established at NDA execution.
Engine runs deterministically on the supplied input.
First output bundle delivered within five business days of NDA execution.
Five-indicator output: stability, aggregation, variants, pocket fit, resistance.
Commercial terms scoped against your specific use case.
Engine access priced at $100,000 per token. One token unlocks the Tier 1 verdict on one sequence. Additional tokens on the same sequence unlock deeper tiers up to Tier 5 for the full detail. Compound co-development terms available under NDA. Workshop fees due on receipt. Wand (the compound co-development track) upfront and milestones due within seven calendar days of invoice. Royalties quarterly within thirty days of quarter close.
For evaluators with a target in hand. Enter your corporate email and a short note on what you want evaluated. Your unique NDA signing link is emailed within seconds. Sequence intake happens after NDA execution on a separate intake page under appropriate protective instruments.
For business development and senior R&D leadership exploring the platform without a specific target in hand. Tell us what you are trying to accomplish, and we schedule an initial 30-minute call.