A fair experimental comparison of neural network architectures for latent representations of multi-omics for drug response prediction

Table 1 Mean AUROC on the test sets from cross-validation

Drug	Omics stacking	MOLI	Super.FELT	Early integration	OmiEmbed	MOMA	PCA
Gemcitabine TCGA	0.646 ± 0.045	0.628 ± 0.118	0.588 ± 0.070	0.611 ± 0.029	0.628 ± 0.057	0.650 ± 0.029	0.614 ± 0.075
Gemcitabine PDX	0.651 ± 0.071	0.622 ± 0.098	0.646 ± 0.063	0.586 ± 0.092	0.539 ± 0.078	0.625 ± 0.034	0.607 ± 0.057
Cisplatin	0.722 ± 0.066	0.764 ± 0.039	0.753 ± 0.047	0.660 ± 0.105	0.640 ± 0.071	0.714 ± 0.075	0.710 ± 0.072
Docetaxel	0.772 ± 0.077	0.792 ± 0.097	0.813 ± 0.051	0.731 ± 0.080	0.803 ± 0.035	0.783 ± 0.060	0.723 ± 0.043
Erlotinib	0.754 ± 0.114	0.705 ± 0.062	0.744 ± 0.125	0.671 ± 0.052	0.664 ± 0.130	0.739 ± 0.105	0.660 ± 0.069
Cetuximab	0.731 ± 0.090	0.731 ± 0.033	0.768 ± 0.045	0.677 ± 0.075	0.754 ± 0.044	0.751 ± 0.043	0.655 ± 0.062
Paclitaxel	0.667 ± 0.138	0.596 ± 0.117	0.726 ± 0.121	0.607 ± 0.060	0.740 ± 0.098	0.692 ± 0.081	0.588 ± 0.124
Mean Rank	2.86	3.86	2.43	5.71	4.00	3.00	6.14

Best results are shown in bold and second best are italics. The values represent the means and standard deviations over five iterations

ISSN: 1471-2105