Quantitative Structure–Mutation–Activity Relationship Tests (QSMART) model for protein kinase inhibitor response prediction

Table 2 Comparisons of drug response prediction by QSMART, DNN and statistical methods

Cancer type	\({\#IC}_{{50}}\)	QSMART model						Performance \(({R}^{{2}})\)
		#All	#Drug	#Cancer features		#Interactions		QSMART + (NN/RF/SVM/EN)				Compared method
		Features	Features	Residue	Others	DxM	Others	NN	RF	SVM	EN	\(\hbox {ANOVA}^{{*}}\)	\(\hbox {MCA}^{{**}}\)
AG	2971	62	31	0	9	4	18	0.879	0.588	0.581	0.293	0.672	0.656
Bone	3410	84	52	0	13	4	15	0.856	0.621	0.667	0.370	0.693	0.819
Breast	4706	129	70	5	26	12	16	0.880	0.604	0.673	0.496	0.702	0.814
CNS	4250	114	65	0	23	11	15	0.858	0.678	0.719	0.465	0.774	0.851
Cervix	1044	37	29	0	3	1	4	0.864	0.696	0.768	0.455	0.809	0.824
Endometrium	1073	33	21	0	4	4	4	0.878	0.596	0.580	0.328	0.769	0.832
Haematopoietic	4204	119	58	3	24	28	6	0.858	0.615	0.649	0.429	0.679	0.807
Kidney	2458	73	51	0	3	0	19	0.836	0.681	0.734	0.415	0.794	0.820
Large intestine	4628	141	53	10	14	50	14	0.814	0.617	0.692	0.495	0.736	0.794
Liver	1348	48	35	0	4	2	7	0.836	0.646	0.678	0.377	0.730	0.859
Lung (NSCLC)	9205	207	72	7	35	47	46	0.854	0.641	0.707	0.513	0.728	0.819
Lung (others)	7206	162	58	2	16	46	40	0.859	0.602	0.687	0.470	0.725	0.791
Lymphoid	13302	291	72	54	30	86	49	0.873	0.647	0.740	0.495	0.758	0.834
Oesophagus	3337	91	58	0	17	4	12	0.841	0.657	0.699	0.452	0.771	0.838
Ovary	3502	113	64	2	18	9	20	0.844	0.659	0.690	0.522	0.741	0.810
Pancreas	2421	84	60	0	7	0	17	0.833	0.693	0.737	0.492	0.784	0.816
Pleura	1431	36	23	0	5	0	8	0.805	0.629	0.623	0.303	0.776	0.837
Skin	5732	132	64	9	21	15	23	0.875	0.694	0.706	0.458	0.754	0.800
Soft tissue	1938	63	45	0	10	2	6	0.818	0.612	0.671	0.404	0.758	0.786
Stomach	2327	83	49	0	13	16	5	0.836	0.592	0.638	0.392	0.720	0.842
Thyroid	1352	33	25	0	5	0	3	0.830	0.644	0.680	0.398	0.798	0.853
UAT	3856	126	50	1	14	4	57	0.881	0.750	0.758	0.600	0.792	0.841
Urinary tract	1454	68	47	0	5	9	7	0.863	0.645	0.683	0.433	0.754	0.847
Overall	87155							0.863	0.655	0.710	0.460	0.755	0.823

The best performance for each cancer type is highlighted in underlined. The performance of each machine learning method is based on 10-fold cross-validation
\(\hbox {ANOVA}^{*}\), analysis of variance, which did not undergo 10-fold cross-validation. \(\hbox {MCA}^{{**}}\), multiscale convolutional attentive, a drug response prediction method [36]. The performance of MCA is based on its prediction for PKI response (Additional file 2). AG, autonomic ganglia; CNS, central nervous system; DxM, drug–mutation interaction term; EN, elastic net; NN, neural networks; NSCLC, non-small cell lung cancer; \(\hbox {R}^{2}\), coefficient of determination; RF, random forests; SVM, support vector machine; UAT, upper aerodigestive tract; \(\#\hbox {IC}_{50}\), the number of drug responses; #Nodes, the number of nodes in the first and second hidden layers of neural networks

ISSN: 1471-2105