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.
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.
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.
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.
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.
| Property | TROOPIR | AlphaFold (incl. AF3) | AutoDock Vina | FEP+ (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."
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).
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.
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.