Research Use Only. KIRhub outputs are computational research artifacts. They are not validated for clinical decision-making, diagnosis, or treatment.
KIRhub 2.0· v2Sign in
← Dashboard

Mesenchymal Reversal Workbench

A self-contained research workbench around one cancer-state axis — epithelial→mesenchymal transition — built on Table S9 (92 inhibitors × 47 mesenchymal-specific kinases) and combined with the pathway-reversal model, the Oncoscape atlas, and the mutation/variant matrices.

Combination designer — minimal set-cover of the mesenchymal program

MeasuredDerived

The greedy weighted set-cover finds the smallest drug combination whose members cover complementary mesenchymal kinases to strong inhibition, prioritising the most mesenchymal-enriched kinases (Table S8 fold-change). No single inhibitor covers the whole program — this is where a combination beats monotherapy. Adjust the threshold below.

37/47 kinases covered (79% by count, 80% expression-weighted) with 5 drugs.
#Add drugNew kinasesNewly coveredCumulative coverage
1Ponatinib+18ABL2, DDR2, EGFR, EPHA2, EPHA4, EPHA5, EPHB2, FGFR1, FYN, MAP4K4, JAK1, LATS2, MAP3K3, PDGFRA, PDGFRB, PRKACB, TGFBR2, TEK18/47 (38%)
2Sunitinib+9NUAK1, AXL, CAMK2D, STK17A, MYLK, PEAK1, PRKD1, PRKD3, RPS6KA227/47 (57%)
3Pacritinib+5ACVR1, HIPK2, LIMK1, PRKCA, DAPK332/47 (68%)
4Defactinib+3BMPR2, CDK15, DYRK335/47 (74%)
5Ceritinib+2DCLK2, STK32B37/47 (79%)

Uncovered (10): MET (best 1%), AKT3 (best 1%), TGFBR1 (best 15%), MAP3K12 (best 37%), CAMK1 (best 37%), SGK1 (best 42%), PLK2 (best 45%), CAMK4 (best 76%), NEK7 (best 84%), NEK6 (best 89%) — no clinical inhibitor reaches strong inhibition on these at this threshold.

Best single agent for comparison: Gilteritinib (16/47 strong hits).

Per-sample EMT reversal across the Oncoscape atlas

MeasuredModeledDerived

Real tumors ranked by their mesenchymal (EMT hallmark) GSVA activation, each paired with the single drug the pathway-reversal model predicts will best reverse that tumor’s pathway state — the patient-level analog of the catalog-level coverage above.

Resistance-aware targeting

MeasuredReference

Which mesenchymal-program kinases are also clinical mutation drivers, and for the drugs that strongly inhibit the wild-type, whether the patient’s variant retains sensitivity or escapes (mutant residual activity from the variant assay matrix). Surfaces EMT drivers that are druggable and not escaped.

KinaseS8 fold-changeVariantsCancersWT-active drugs (escape count)
AXL5.651carcinoma_large_intestine, carcinoma_skin19 · 0 escapes
DDR23.682carcinoma_lung34 · 12 escapes
PDGFRA2.744AML-US, adnexal_tumour_skin, carcinoma_breast, carcinoma_large_intestine23 · 14 escapes
FGFR12.732carcinoma_breast, carcinoma_lung, glioma_central_nervous_system22 · 14 escapes
AKT32.592KIRC, KIRC-US, LUSC, LUSC-US1 · 0 escapes
FYN1.851AML-US, glioma_central_nervous_system15 · 0 escapes
ACVR11.6623 · 1 escape
EGFR1.5064BRCA, BRCA-EU, BRCA-US, COREAD24 · 23 escapes
MET1.202910 · 8 escapes
TEK1.1697 · 3 escapes

Confidence shortlist

MeasuredDerived

Each of the 47 mesenchymal kinases scored on the signals we can measure: S8 mesenchymal enrichment, the best clinical inhibition achievable in S9, whether it is a mutation driver, and whether any variant escapes. Note: DepMap genetic essentiality is not yet ingested (Table S7 holds only per-lineage counts), so this is not an essentiality verdict.

KinaseS8 fold-changeBest residual %Strongly druggableMutation driverResistance escape
AXL5.651no
MYLK4.2514
PDGFRB3.930
DDR23.680⚠ yes
PDGFRA2.740⚠ yes
FGFR12.730⚠ yes
AKT32.591no
TGFBR22.399
PRKCA2.092
CDK152.0530
PRKD12.027
STK17A1.9829
NUAK11.972
FYN1.850no
STK32B1.7211
ABL21.660
ACVR11.663⚠ yes
MAP4K41.660
RPS6KA21.661
PEAK11.590
DYRK31.570
DCLK21.5222
EPHA51.500
EGFR1.500⚠ yes
LATS21.374
PRKACB1.252
EPHA21.220
MET1.201⚠ yes
CAMK2D1.190
JAK11.171
DAPK31.162
TEK1.161⚠ yes
EPHB21.150
HIPK21.122
TGFBR11.1215
EPHA41.100
LIMK11.101
BMPR21.1016
PRKD31.048
MAP3K31.036
SGK12.8842
PLK22.4245
CAMK11.8637
NEK61.6789
NEK71.5484
MAP3K121.2937
CAMK41.0676