Glioblastoma integrated preclinical report: five curated targets, 34 registry candidates through a three-stage sensor funnel, a PIK3CA docking pilot at sub-angstrom top-pose RMSD and a measured N-methyl-alpelisib design line
A preclinical research package for glioblastoma assembled by the HELIX Discovery engine on 2026-11-21. Every figure below is read back from the deposited artifacts, not retold from prose. Target selection (stage 1, 112 HTTP attempts, offline model_calls: 0). From the disease record MONDO_0018177 / MeSH D005909 the engine carried forward EGFR, PIK3CA, IDH1, MGMT and KDR as the working panel; FGFR1, PDGFRA and ITGAV passed the DepMap lineage-selectivity rule, and PIK3CA and EGFR were added on association grounds. Ten drug-target pairs were fixed in the run configuration (temozolomide-MGMT, lomustine-MGMT, bevacizumab-VEGFA, regorafenib-KDR, vorasidenib-IDH1, alpelisib-PIK3CA plus the EGFR, PTEN, TP53 and TERT reference arms) over cell lines U-87MG and T98G. Compound intake (stage 2, 197 HTTP attempts). The sensor funnel across ChEMBL, DGIdb, BindingDB, PharmacoDB, FDA and DrugComb routes produced 34 unique registry records; 18 structures were resolved to clean SMILES for chemistry. Chemistry (stage 3, 45 HTTP attempts). pKa and cLogD were computed with an RDKit worker for the full CNS MPO panel; no external model calls and no paid inference were used anywhere in the three stages. Docking pilot (PIK3CA, 4JPS). AutoDock-Vina via meeko, three seeds (1941 / 3347 / 5689): top-pose RMSD 0.87-0.88 A. Design line A (N-methyl-alpelisib) measured TPSA -8.79 A^2, H-bond donors 2 to 1, median amide N to Val851 backbone O distance 3.18 to 6.35 A, and median Vina score -8.31 against -9.91 for the reference. Live dossier run. A separate LIVE pass produced 10 target-drug dossiers with 370 HTTP attempts across 18 hosts; mode LIVE, model_calls: 0, cost_usd: 0.0. The "354 HTTP attempts" quoted in the PDF covers stages 1-3 only (112 + 197 + 45); the 370 in run_summary.json belongs to the dossier run. Both numbers are measured and refer to different passes. Boundaries. The novelty and selectivity scores in this package are NOT calibrated (calibrated: false in the run summary); nothing here is a clinical, diagnostic or therapeutic claim, and no compound in it has been tested experimentally by the author. This is a documented, reproducible research output with per-host request receipts, offered as a working example of an $0 instrumented discovery chain.
Authors
- Dzianis Vashkevich
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-21
- DOI
- https://doi.org/10.5281/zenodo.24424485
- Primary Topic
- Computational Drug Discovery Methods
- Type
- article
- Field-Weighted Citation Impact
- 0.00