Explaining decisions of graph convolutional neural networks: patient-specific molecular subnetworks responsible for metastasis prediction in breast cancer

Table 3 Actionable genes identified by the MTB report workflow

Patient	Gene	Expression	Known Var	Predicts
615695	HSPB1	Normal	expression	Response to gemcitabine
	ABL1	High	GoF	Response to ABL TK inhibitors (imatinib, desatininb, ponatinib, regorafenib …)
	AKT1	High	GoF	Response to PI3K, AKT, MTOR inhibitors; resistance to BRAF inhibitors
	ERBB2	High	GoF	Response to ERBB2, EGFR, MTOR, AKT inhibitors
	MAPK3	High	GoF	Resistance to EGFR inhibition
519217	HSPB1	Normal	expression	Response to gemcitabine
	CTNNB1	High	GoF	Response to everolimus + letrozole; resistance to Tankyrase inhibitors
	EGFR	High	GoF	Response to EGFR, ERBB2, HSP90 and MEK inhibitors
	ERBB2	High	GoF	Response to ERBB2, EGFR, MTOR, AKT inhibitors
	JUN	High	overexpr	Response to irbesartan (angiotensin II antagonist)
	MCL1	High	GoF	Resistance to anti-tubulin agents
	PTPN11	High	GoF	Response to MEK inhibitors
615233	FOS	High	overexpr	Response to irbesartan (angiotensin II antagonist)
	PTPN11	High	GoF	Response to MEK inhibitors
150990	HSPB1	Normal	expression	Response to gemcitabine
	ESR1	High	GoF	Response to novel ER degraders, fulvestrant, tamoxifen

Genes from the PPI subnetworks were matched to known genomic alterations (Known Var) that predict either response or resistance to drugs (Predicts). High and low gene expression were matched to gain of function (GoF) and loss of function (LoF) genomic variants, respectively

ISSN: 1756-994X