How this project sits relative to the published ENPP1-inhibitor landscape. Every external claim below is grounded in a retrieved, DOI-verified record (PubMed + ChEMBL + ClinicalTrials.gov, queried programmatically). This document credits the work we build on and states honestly where we add value and where others are ahead.
Sources: 92 PubMed records triaged to 59 ENPP1 medicinal-chemistry / selectivity
/ structure papers; the full ChEMBL bioactivity landscape for ENPP1 (CHEMBL5925),
ENPP2/autotaxin (CHEMBL3691), and ENPP3 (CHEMBL5580); 23 ENPP1-related clinical
trials. Retrieval scripts and raw pulls are in analysis/data/ and benchmark/.
ENPP1 is a validated immuno-oncology target: it is the dominant extracellular hydrolase of cGAMP, the second messenger of the cGAS–STING innate-immunity pathway, so ENPP1 inhibition sustains STING signaling and anti-tumor immunity. The clinical and medicinal-chemistry field is well ahead of any single computational screen:
- Clinical stage. Four dedicated ENPP1-inhibitor oncology programs are in Phase 1 (NCT05978492, NCT05270213, NCT06063681 recruiting; NCT06724042 not-yet-recruiting). A separate ENPP1 axis — enzyme replacement (INZ-701) for ENPP1 deficiency / GACI — runs up to Phase 3 (the ENERGY program). ENPP1 is a two-sided target: inhibit it in cancer, restore it in calcification disease.
- Best-in-class chemistry is already excellent. The literature bar we hold ourselves against: ISM5939 (Insilico Medicine, IND-cleared, 0.63 nM, >15,000x vs ENPP2, >3,400x vs ENPP3); AVA-NP-695 (a selective ENPP1 inhibitor shown to activate STING and reduce metastasis, Baird et al. 2022, PMID 36235254); an oral ENPP1 inhibitor designed with generative AI as a next-generation STING modulator (Nature Communications 2025, PMID 40410143); and an inhibitor with ultralong target-residence time framed as an innate-immune checkpoint (Cell Reports Medicine 2025, PMID 40914167).
- Active chemotype series (2020-2026). At least a dozen distinct scaffolds are
under public medicinal-chemistry optimization: imidazo[1,2-a]pyrazines, pyrrolo-
pyrimidines/pyridines, [1,2,4]triazolo[1,5-a]pyrimidines, quinoline/quinazoline
carboxylics, arylsulfonates, dihydropyrimido-pyrimidinones, benzotriazoles, and
pyrazole derivatives (full DOI list in
benchmark/benchmark_literature_full.csv).
Where we stand: we do not claim novel chemistry or a clinical candidate. Our leads are existing ChEMBL sulfamides/sulfonamides. This project is a method and a counter-screen, not a drug announcement — and that framing is honest about a field that already has IND-cleared molecules.
This is the strongest defensible contribution, and it comes straight from the data. Querying ChEMBL for potent compounds (pChEMBL >= 7) across the three ENPP paralogs:
| Paralog | Target | Potent compounds (pChEMBL>=7) | Max pChEMBL |
|---|---|---|---|
| ENPP2 / autotaxin | CHEMBL3691 | 1,546 | 11.0 |
| ENPP1 | CHEMBL5925 | 199 | 10.05 |
| ENPP3 | CHEMBL5580 | 3 | 10.05 |
Two facts fall out of this table:
- The public bioactivity is wildly asymmetric. Autotaxin has three decades of drug-discovery history (1,546 potent compounds); ENPP3 has essentially none (3). Any ENPP1 program that wants to claim ENPP3 selectivity cannot look it up — there is almost no public ENPP3 SAR to look up.
- Head-to-head selectivity data is almost non-existent. Of every compound in ChEMBL across all three paralogs, exactly one (CHEMBL5566114) has been assayed on more than one ENPP paralog. Cross-paralog selectivity — the quantity every ENPP1 program needs — is almost never measured and published.
A structure-based computational counter-screen is a rational, cheap way to generate selectivity hypotheses where the assay data does not exist. That is precisely what this repository does, and the data gap above is why it is worth doing.
Our selectivity argument is not just a docking artifact — it was independently motivated in 2024. Li et al. (Cell Reports 2024, PMID 38749434) identified ENPP3 as the second and likely only other metazoan cGAMP hydrolase, accounting for all cGAMP-hydrolysis activity in ENPP1-deficient mice, and showed that selectively abolishing ENPP3's cGAMP hydrolysis restrains tumor growth. Implication: an ENPP1 inhibitor that also hits ENPP3 may be pharmacologically fine (both are immune checkpoints), but one that spares ENPP3 isolates the ENPP1 mechanism. Either way, knowing the ENPP1/ENPP3 margin is decisionrelevant — and our screen provides it.
This also sharpens a limitation we already state: our own conservation analysis shows the ENPP1 and ENPP3 catalytic pockets are ~identical in core residues, so ENPP1/ENPP3 selectivity can only come from pocket shape, not residue identity. The 2024 biology explains why that is a hard but worthwhile target.
We are not the first to dock small molecules into ENPP1, and we credit the prior work:
- Myricetin-to-natural-products virtual screen (Front. era, PMC9573336): a virtual-screening -> docking -> molecular-dynamics -> MM/GBSA pipeline for natural ENPP1 inhibitors. This is the closest open computational analog to our workflow — but it is single-target (ENPP1 only), not a paralog counter-screen, and does not report cross-paralog selectivity margins with replicate statistics.
- Benzotriazole docking-SAR (Eur. J. Med. Chem. 2026, DOI 10.1016/j.ejmech.2026.118666): reports that the sulfamide warhead coordinates the ENPP1 Zn ion and hydrogen-bonds to Asp. This is an important external check on our mechanism claim — our own Zn-coordination analysis found sulfamides do not tightly coordinate Zn (2.7A in ENPP1). The disagreement is worth flagging and re-examining; it may reflect docking protocol, pose selection, or the specific sulfamide series. We record it rather than hide it.
- Open-source docking tooling. Our stack (smina/AutoDock Vina lineage + Open Babel
- RDKit) is the standard open molecular-modeling toolchain catalogued by the Open Source Molecular Modeling community; recent reviews (PMC13073925) position this exact class of open, reproducible docking workflow as the accessible tier of structure-based design. We add: replicate statistics with bootstrap CIs, a 4-method consensus, interaction-fingerprint and Zn-coordination mechanism analysis, and a rational-analog round — layered on top of that shared foundation.
Building on the shoulders above, the incremental contributions of this repo are:
- A three-paralog counter-screen with replicate statistics, where the public record has head-to-head data for one compound. (novelty: the comparison, done with uncertainty quantification.)
- A statistical noise floor. 42-seed replicates + 10k-bootstrap CIs turn "docking score X" into "margin +/- CI", separating resolved directional signal from noise.
- Orthogonal mechanism evidence (Zn-coordination distances, interaction fingerprints, Vinardo consensus) that agrees on the robust claim (hydroxamate anti-selectivity) and flags where the promotable claim is scoring-function dependent (sulfamide direction).
- A rational-analog round (AN10_piperidinol) that improves both predicted potency and selectivity margin over the parent, with the margin CI excluding zero — a concrete, synthesizable, assay-ready hypothesis.
- Full open reproducibility — every score, CI, figure, and receptor regenerable from committed scripts.
| Dimension | Field best | This project | Verdict |
|---|---|---|---|
| Potency (ENPP1) | ISM5939 0.63 nM (assay) | leads pChEMBL 9.3-9.7 (public assay); docking only | field far ahead; we screen, don't measure |
| Selectivity data | ISM5939 >3,400x ENPP3 (assay) | directional margins +/- CI (docking) | we generate hypotheses where public assay data is absent |
| Clinical maturity | Phase 1 (x4) + Phase 3 (INZ-701) | none | field far ahead; not our aim |
| Cross-paralog head-to-head | ~none in ChEMBL (n=1) | all 6 leads x 3 paralogs, replicated | we lead on open, systematic comparison |
| Statistical rigor of docking | rarely reported | 42-seed + bootstrap CI + consensus | above the typical open docking paper |
| Reproducibility | variable | fully scripted, open | strong |
Bottom line: the field owns potency, selectivity measurement, and the clinic. This project owns the open, statistically-honest, three-paralog counter-screen and the mechanism analysis behind it — a genuine, if narrow, contribution built explicitly on the tools and biology cited above.
Retrieval provenance: PubMed E-utilities, ChEMBL API, ClinicalTrials.gov API, all
queried programmatically July 2026. DOIs cross-verified via Crossref. See
benchmark/ for the raw pulls and benchmark/benchmark_literature_full.csv for the
full DOI-linked reference list.