Myeloid genomics

Genomic
Co-occurrence
Analysis

Five driver mutations. Zero matches in 26,642 screened patients.

Henrik G. Røine The Educational Equality Institute Patient-researcher, N=1 study

DNMT3A R882H
IDH2 R140Q
SETBP1 G870S
PTPN11 E76Q
EZH2 V662A
del(7) Monosomy
Accepted for poster ISMB 2026 VarI COSI track

From sequence to structure to drug binding: a six-axis AI pipeline for myeloid variant interpretation

Poster
#623
Accepted
6 May 2026
Author
Henrik G. Røine
Affiliation
The Educational Equality Institute

Zero quintuple co-occurrences

Screened
26,642 patients across GENIE + HARMONY
Expected
1 in 1.9 billion under pairwise-corrected model
Detection
55,875 patients needed for 50% probability

Analysis

Mutation Profile Variant table, progressive funnel, karyotype Co-occurrence 190 gene pairs, Fisher exact tests, heatmap Cross-Database 10+ databases, GDC, cBioPortal, HARMONY SKI Domain SETBP1 hotspot map, lollipop plot

Computational

ESM-2 Scoring Masked marginal LLR, attention maps, embedding disruption Molecular Docking Boltz-2 affinity, DiffDock 2.0, Chai-1 complexes Pathway Networks STRING PPI, PrimeKG, GenePT, DISCOVER

Pathogenicity

ACMG Classification CancerVar, EVE, REVEL, CADD, gnomAD, ClinGen DMS Validation SHP2 deep mutational scanning, MaveDB, PS3 evidence Benchmark 40 SETBP1+ profiles, 284 variants, 80.9% concordance Statistical Rarity Bayesian estimation, extreme value theory

Case Studies

EZH2 V662A Novel unreported variant, SET domain LoF, PRC2 axis DDX41 + SETBP1 Enrichment analysis, O/E = 5.16 Clonal Evolution PyClone-VI clustering, CCF, fish plots

Clinical

Clinical Trials Trial matching, VICC meta-KB, drug repurposing Prognosis Expected vs actual survival analysis Methods Six-axis pipeline, tool inventory, data downloads