Super.FELT: supervised feature extraction learning using triplet loss for drug response prediction with multi-omics data

Table 7 Results of external validation on PDX and TCGA

Drug	Super. FELT	MOLIF	AE	ANNF	MOLI	Super.FELT E	Super. FELT M&C	MOLI*	Auto BorutaRF	SVM
5-Fluorouracil(PDX)	0.865	0.77	0.872	0.919	0.582	0.496	0.782	0.529	0.449	0.413
5-Fluorouracil(TCGA)	0.493	0.390	0.502	0.486	0.481	0.506	0.423	0.519	0.591	0.528
Cetuximab(PDX)	0.636	0.496	0.42	0.426	0.360	0.529	0.694	0.392	0.171	0.407
Cetuximab(TCGA)	0.353	0.440	0.324	0.433	0.333	0.460	0.471	0.258	0.312	0.216
Cisplatin(TCGA)	0.759	0.796	0.459	0.678	0.595	0.747	0.364	0.590	0.449	0.471
Docetaxel(TCGA)	0.611	0.538	0.498	0.637	0.421	0.452	0.570	0.493	0.422	0.528
Doxorubicin(TCGA)	0.510	0.512	0.352	0.545	0.426	0.453	0.590	0.486	0.441	0.505
Erlotinib(PDX)	0.742	0.527	0.334	0.652	0.609	0.588	0.442	0.513	0.166	0.669
Erlotinib(TCGA)	0.85	0.7	0.51	0.9	0.790	0.420	0.700	0.660	0.555	0.670
Etoposide(TCGA)	0.020	0.050	0.000	0.050	0.240	0.020	0.430	0.210	0.470	0.040
Gemcitabine(PDX)	0.682	0.694	0.468	0.691	0.663	0.553	0.686	0.772	0.447	0.640
Gemcitabine(TCGA)	0.634	0.575	0.501	0.568	0.611	0.582	0.613	0.616	0.526	0.470
Mitomycin C(TCGA)	0.640	0.080	0.000	0.067	0.333	0.320	0.573	0.147	0.247	0.187
Paclitaxel(PDX)	0.72	0.609	0.436	0.642	0.539	0.631	0.525	0.369	0.457	0.494
Paclitaxel(TCGA)	0.529	0.41	0.383	0.49	0.524	0.472	0.627	0.497	0.498	0.461
Sorafenib(TCGA)	0.818	0.775	0.889	0.772	0.871	0.557	0.698	0.938	0.645	0.914
Tamoxifen(TCGA)	0.758	0.679	0.368	0.6	0.584	0.461	0.400	0.674	0.446	0.566
Temozolomide(TCGA)	0.634	0.676	0.578	0.649	0.666	0.643	0.649	0.633	0.562	0.662
Trametinib(PDX)	0.555	0.479	0.732	0.506	0.5	0.625	0.527	0.404	0.357	0.688
AVG	0.622	0.537	0.454	0.564	0.533	0.501	0.567	0.511	0.432	0.501

Bold values indicate methods with best performance for each drug
MOLI* was the test AUC of randomly selecting hyperparameters having best validation.
MOLI was tuned using hyperparameters from Additional file 1: Table S3 of Sharifi-Noghabi et al. [25]

ISSN: 1471-2105