Learning important features from multi-view data to predict drug side effects

Table 1 Performance comparison of different algorithms for side effect prediction

	Sample-AUC	Macro-AUC	Micro-AUC	LRAP	Coverage error	Ranking loss
L1SVM	0.8524 ± 0.0010	0.6196 ± 0.0056	0.8328 ± 0.0010	0.1941 ± 0.0013	2671 ± 11	0.1483 ± 0.0010
L1LOG	0.8612 ± 0.0010	0.6191 ± 0.0059	0.8418 ± 0.0010	0.2018 ± 0.0015	2562 ± 15	0.1394 ± 0.0010
PCR	0.8824 ± 0.0004	0.5034 ± 0.0033	0.8670 ± 0.0006	0.1890 ± 0.0010	2666 ± 16	0.1233 ± 0.0004
SCCA-chem	0.8500 ± 0.0019	0.5731 ± 0.0045	0.8181 ± 0.0019	0.4008 ± 0.0032	2960 ± 23	0.1507 ± 0.0020
SCCA-domain	0.9144 ± 0.0007	0.6260 ± 0.0055	0.8922 ± 0.0007	0.4757 ± 0.0011	2547 ± 13	0.0863 ± 0.0008
SCCA-GO	0.8911 ± 0.0015	0.6160 ± 0.0057	0.8579 ± 0.0015	0.4509 ± 0.0017	2789 ± 25	0.1097 ± 0.0015
SCCA-expression	0.9076 ± 0.0007	0.5159 ± 0.0031	0.8878 ± 0.0007	0.4488 ± 0.0005	2607 ± 9	0.0941 ± 0.0007
SCCA-target	0.9174 ± 0.0005	0.6159 ± 0.0051	0.8968 ± 0.0005	0.4692 ± 0.0007	2490 ± 11	0.0834 ± 0.0005
Kernel regression	0.9185 ± 0.0005	0.6134 ± 0.0053	0.8992 ± 0.0005	0.4766 ± 0.0007	2448 ± 8	0.0821 ± 0.0005
LRSL-chem	0.9179 ± 0.0003	0.5595 ± 0.0034	0.8976 ± 0.0004	0.4614 ± 0.0005	2583 ± 10	0.0867 ± 0.0005
LRSL-domain	0.9285 ± 0.0005	0.6470 ± 0.0050	0.9104 ± 0.0007	0.4821 ± 0.0005	2174 ± 14	0.0719 ± 0.0005
LRSL-GO	0.9290 ± 0.0007	0.6441 ± 0.0043	0.9068 ± 0.0010	0.4924 ± 0.000	2255 ± 15	0.0714 ± 0.0007
LRSL-expression	0.9203 ± 0.0004	0.5131 ± 0.0013	0.9008 ± 0.0005	0.4565 ± 0.0006	2198 ± 11	0.0801 ± 0.0004
Multi-LRSL	0.9295 ± 0.000	0.6568 ± 0.0057	0.9118 ± 0.0009	0.4845 ± 0.0006	2160 ± 13	0.0709 ±0.0006

The metrics are denoted as \(\hbox {mean}\pm \hbox {standard deviation}\) . The method taking different types of features as input is indicated in the form of ’method name-feature type’
The best metric values are in italics, and the difference between the best value and the second best value of each metric is significant (student t-test, \(\hbox {p-value}<0.05\))

ISSN: 1758-2946