Pleco Kineτics

What this is

A research-and-product effort focused on a specific gap in drug discovery infrastructure: the layer where binding kinetics, target selectivity, and mechanism-grounded prediction intersect.

We do three things:

Curate kinetic and binding-affinity data to publication-grade quality, with provenance, multi-modality reproducibility weighting, and explicit data-quality flagging.

Model with formalism chosen by substrate mechanism — not GKSL by default, not deep learning by default. Empirical rate laws, two-state equilibrium, multi-state classical master equations, or genuine quantum dynamics — whichever the physics supports.

Pre-register methodology, training data, and acceptance criteria before evaluation. Document honest negative results. No post-hoc reinterpretation.

What's in development

Active research substrate: carbonic anhydrase (CA) inhibitor selectivity.

We're building a PLBD-trained ML predictor for CA II vs CA IX selectivity, using multi-modality experimental data (SPR + ITC + FTSA + SFA) combined via reproducibility weighting. To our knowledge, no such predictor exists in the literature; the methodology paper is in progress.

Datasets currently curated:

• 42 kinase targets, 32K+ rows of binding kinetic data with provenance
• ABL1 binding mode annotations (literature-verified, 121 rows)
• Full PLBD enumeration: 12 carbonic anhydrase isoforms across 4 experimental modalities
• 135 paired k_on/k_off measurements on ABL1 (rare data; preserved for future kinase work)

Methodology:

The four-phase research protocol — mechanism study, formalism mapping, textbook rate laws, ML — runs separately for each substrate. Cross-substrate parameter inheritance is not assumed. Each substrate paper is its own claim.

What this isn't

To prevent overinterpretation:

• Not a general kinetic prediction API. The current product is research-stage. Predictions are not yet available for general queries.
• Not "send us structures, we predict τ_res." That framing was retired in our 2026-04-19 methodology reassessment. We do not promise universal kinetic prediction.
• Not a competitor to kinetic measurement vendors (Biacore, Octet, etc). Pleco does not measure kinetics. Pleco curates and models data that other labs produce.
• Not validated beyond the active research substrate. Claims will scope to whatever Run 3 + Run 4 (CA selectivity) honestly demonstrate, with peer-reviewed methodology paper as the receipt.

Roadmap

Now: Run 3 + Run 4 evaluation on CA II vs CA IX selectivity. Pre-registered acceptance bars (R² ≥ 0.78 / 0.55 / 0.40 tiered). All results reported with equal prominence; no cherry-picking.

Next: CA selectivity preprint, then peer-reviewed venue. Post-publication: scoped API access for collaborating labs.

Beyond: Methodology applies to additional substrate domains (GPCRs, enzymes, ion channels) on the same mechanism-first → pre-registered → ML evaluated path. No commitments until each substrate's Phase 1 mechanism study completes.