Research state

engine online canine hemangiosarcoma × human angiosarcoma phase 0 locked · free track live · paid track running

Compute lanes

docking · cofolding · MD · omics

Validated results

docking, cofold, MD-checkpoint, cross-species crux

Hypotheses falsified

then sharpened into the endothelial/PI3Kα axis

Tests passing

76.5% coverage · CI green

COMPUTE LANES

Lane	Compute	Status	Last result
gnina docking CNN docking · gnina v1.3.1	A100	verified	alpelisib→PI3Kα redock RMSD 1.80 Å
Boltz-2 cofolding structure + affinity	A100	verified	alpelisib–PI3Kα iptm 0.99 · ~11 nM
OpenMM MD checkpoint / resume	T4·GPU	verified	cross-invocation resume on durable Volume
Omics TME review deconvolution · purity-adjusted	CPU	verified	PIK3CA crux + purity confound caught

EVIDENCE — PROOF CAPSULES

Claim tested	Method	Readout	Signal
PIK3CA-mut → immunosuppressed TME	omics	washes out on purity adj.	refutes
PIK3CA-mut → endothelial/vascular axis (human AS)	omics	AUC 0.96 · p=0.001	supports
alpelisib engages PI3Kα (pocket)	docking	RMSD 1.80 Å · −8.9 kcal/mol	supports
alpelisib–PI3Kα binding (co-fold + affinity)	cofolding	iptm 0.99 · ~11 nM (vs 4.6 nM)	supports

Every capsule carries who ran it, the target, the limitations, and a directional signal — and promotes onto a claim only through the operator write-gate.

H1 — CROSS-SPECIES DRUGGABILITY · SUPPORTED

Hypothesis H1: alpelisib engages canine PI3Kα comparably to human (conserved druggable pocket) → the PI3Kα therapeutic rationale plausibly translates to canine HSA.

POSITIVE CONTROL · human PI3Kα

supports · iptm 0.99 · ~11 nM

NEGATIVE CONTROL · VEGFR2 (off-target)

refutes · P(binder) 0.11 · ~4,300 nM

H1 PROPER · canine PI3Kα · kinase domain 100% identical to human

supports · iptm 0.99 · P(binder) 0.92 · ~11 nM — matches human

MD CONFIRMATION · 250 ps GPU dynamics

stable · ligand RMSD 1.25 Å · alpelisib stays in the pocket

Verdict: supported. Alpelisib's PI3Kα binding pocket is 100% conserved between human and dog (99.8% overall); the cofold predicts identical engagement (~11 nM). The rig discriminates — the VEGFR2 off-target drops to P(binder) 0.11 / ~4,300 nM — so the on-target "supports" is trustworthy. The human PI3Kα rationale translates to canine HSA at the target level. Caveat: target conservation ≠ tumor efficacy.

FLYWHEEL — WHAT IT GIVES BACK

dataset

Cross-species cohort

Harmonized canine HSA ↔ human AS, genotype × expression, reproducible pipeline. License-clean (referenced, not re-hosted).

registry

Canine TME signatures

16 panels · 85 markers · both canine assemblies. The dog-specific gene-set gap, filled.

Coverage gap, stated plainly: no canine HSA single-cell atlas exists, and HSA tumor cells are themselves endothelial — so v0 deconvolution is immune-only; a tumor-purity model needs a new HSA reference.