Validation Receipt · 01 / TTR + Tafamidis
Validation Receipt · Public Domain Targets

When the engine predicts what the FDA already approved.

TROOPIR's deterministic compatibility score, run cold on transthyretin and tafamidis, separates the FDA-approved stabilizer from a glucose distractor by 0.38 points on a 0–1 axis. The prediction was made without training data, without averaging, and without prior knowledge of the clinical outcome. Reproducible from public sequence and SMILES alone.

Score (TTR · Tafamidis)
0.87 / 1.00
Distractor (TTR · Glucose)
0.49
Separation
0.38
FDA Status
Approved

The problem TTR poses to computational drug discovery.

Transthyretin is the textbook hard case for ligand stability prediction. The native protein is a homotetramer. The pathology is misfolding. The therapeutic mechanism is not catalysis or competitive inhibition; it is kinetic stabilization of the tetramer by a small molecule that binds the thyroxine pocket. A correct prediction has to score the ligand against a folded ensemble property, not a single docked pose.

Most computational methods either pose-search the ligand against a static binding site (Glide, AutoDock Vina, Schrödinger's docking suite) or estimate free-energy differences using molecular dynamics ensembles (FEP+, TI). Both classes can place tafamidis in the TTR pocket. Neither answers the actual clinical question, which is whether the ligand-bound tetramer is materially more stable than the apo tetramer under physiological conditions.

The community moved tafamidis through the clinic empirically. Pfizer's Vyndaqel and Vyndamax are now standard of care for transthyretin amyloid cardiomyopathy. The retrospective question for any deterministic computational platform is whether the platform, given only the TTR sequence and the tafamidis structure, would have flagged it.

What TROOPIR returned, run cold.

The engine was provided the human TTR amino acid sequence (UniProt P02766) and the tafamidis SMILES. No prior context. No training on TTR or any related amyloid system. The engine returned a single deterministic compatibility score on the 0–1 axis along with the four supporting output channels of the canonical bundle.

TROOPIR compatibility score · TTR + ligand
Tafamidis
0.87
Tolcapone (off-label literature)
0.62
Glucose (distractor)
0.49

Three deterministic runs against the same TTR target. The FDA-approved stabilizer ranks first by 0.25 above the second-best literature candidate and 0.38 above the negative control. The same seed, the same constants, the same inputs return the same numbers byte-for-byte on every run.

The deterministic nature of the output means the result of a TROOPIR run is auditable in the regulatory sense of the word. The same inputs, run again, return the same score. No averaging, no ensemble noise, no stochastic seed dependence.
From the TROOPIR Framework Reference

Head-to-head against the alternatives.

This comparison is not an attack on neighboring methods. AlphaFold solved a structural prediction problem of historic importance. Vina and FEP+ are workhorses of computational chemistry. They are also, by construction, doing different things than what TROOPIR does on the TTR question. The comparison below specifies which.

PropertyTROOPIRAlphaFold (incl. AF3)AutoDock VinaFEP+ (Schrödinger)
Stability-of-bound-state score, single axis Yes No (structural prediction, not stability) No (pose score, not stability) Possible with extensive setup
Deterministic (byte-identical re-runs) Yes No (stochastic sampling) No (random seed) No (MD ensemble)
No training data dependency Yes No (trained on PDB) N/A (physics-only) N/A (physics-only)
Wall-clock for a single TTR + ligand score Minutes Minutes (structure only) Seconds Hours to days
Returns a regulatory-auditable trail Yes No No Possible with custom logging
Engine method disclosed in publication No (trade-secret-protected, patent-disclosed) Yes Yes Partially

The methods are not interchangeable. TROOPIR is purpose-built to score the stability question on a single deterministic axis. The neighboring methods solve adjacent problems well; they do not solve the stability question deterministically and they do not survive a regulatory audit that asks "can you re-run this and get the same number."

What we disclose, what we don't.

TROOPIR's licensing posture is straightforward: outputs are public; method is private. This is consistent with USPTO trade-secret doctrine and with the way many pharmaceutical assays operate (the result is publishable; the recipe is not).

Disclosed for this validation receipt

  • Inputs. Human TTR amino acid sequence (UniProt P02766) and tafamidis SMILES, both public-domain identifiers.
  • Outputs. Deterministic compatibility score (0–1 axis), comparative ranking against tolcapone and glucose distractor.
  • Reproducibility. The same inputs produce the same output on every engine run; the score is not an average across stochastic samples.
  • Validation context. Tafamidis is the FDA-approved transthyretin amyloid cardiomyopathy stabilizer marketed by Pfizer as Vyndaqel and Vyndamax.
Not disclosed: the underlying scoring methodology, sampling parameters, internal nomenclature, vocabulary, architecture, or any code path. These are protected by the TROOPIR trade secret framework and disclosed under license only to executed counterparties.

What this implies for an evaluation engagement.

Three things follow if the TTR + tafamidis receipt holds up under your team's review:

First, deterministic prediction of an FDA-approved stabilizer is operationally meaningful, not just academically interesting. A computational platform that retrospectively flags tafamidis from sequence and SMILES alone, without training, is a candidate for prospective hit identification on related targets that have not yet found their stabilizer.

Second, the regulatory-audit posture is unique. Methods that return the same output every time on the same input are the only methods that survive an FDA records request that asks for re-derivation. Most current computational chemistry methods do not survive that request without elaborate version-pinning and seed-archival infrastructure.

Third, the validation cost to your team is low. A scoping call walks through the engine envelope on a target of your choosing. The first deterministic output bundle on your target is delivered within five business days of NDA execution and a token order. If the Tier 1 verdict does not satisfy your team, you spend no further tokens; if it does, additional tokens on the same sequence unlock deeper tiers up to Tier 5.

Take the next step

Run TROOPIR on a target of your choosing.

Briefing call within 48 hours. NDA executed electronically on the same domain. First output bundle delivered within five business days post-NDA.

References · Tafamidis FDA approval: 2019 (Vyndaqel, Vyndamax, Pfizer). Transthyretin: UniProt P02766. ATTR-CM clinical evidence: ATTR-ACT trial (Maurer et al., NEJM 2018). This document is a validation receipt and not a substitute for clinical or regulatory advice.