Table 1.

Performance of the FAM score, measured by the identification and prioritization of AEDs

Epi	Identification of AEDs (AUROC)			Prioritisation of AEDs (average percentile)		P
	More effective AEDs from all drugs (mean ± SD)	Less effective AEDs from all drugs (mean ± SD)	More from less effective AEDs	More effective AEDs	Less effective AEDs
HS	0.65 ± 0.13	0.36 ± 0.18	0.87	73	27	8 × 10–3
GE	0.85 ± 0.04	0.69 ± 0.09	0.71	93	70	<1 × 10–6
JME	0.88 ± 0.04	0.76 ± 0.08	0.72	96	86	<1 × 10–6
CAE	0.75 ± 0.05	0.45 ± 0.15	0.79	85	48	2.9 × 10–5

Constituents of the ‘More effective AEDs’ and ‘Less effective AEDs’ drug-sets are specific to each phenotype. ‘Less effective AEDs’ comprise the set of less effective, ineffective or aggravating AEDs for that phenotype. AUROC is calculated using drugs’ FAM scores. AUROC for identifying AEDs from all drugs is computed using the technique of random under-sampling, and presented as mean ± standard deviation (see Supplementary methods). Prioritization is calculated using drugs’ ranks, when all drugs have been ranked from highest to lowest predicted effect on the phenotype. Prioritization result shown is the average (median) rank of AEDs, expressed as a percentile; it is equivalent to the percentage of all drugs ranked below the middle-ranked AED (see Supplementary methods). AUROC, area under the receiver operating characteristics; CAE, childhood absence epilepsy; Epi, epilepsy type or syndrome; GE, generalized epilepsy; HS, focal epilepsy with hippocampal sclerosis; JME, juvenile myoclonic epilepsy; P, permutation-based P-value after Benjamini–Hochberg correction; SD, standard deviation.