Table 3.

Quantitative summary of the significant findings of RPVT, COMBI and DeepCOMBI. For each of the three competitor methods, the numbers of replicated and unreplicated hits (i.e. the number of true and false positives) as well as precision and error rates are presented. Pairwise tests for the null hypothesis of identical distributions for DeepCOMBI and the two baseline methods are performed and the corresponding P-values are given

	Number of significant SNPs of
	RPVT	DeepCOMBI method	COMBI method
SNPs that have achieved P < 10−5 in at least one external study	33 (49% precision)	31 (79% precision)	31 (58% precision)
SNPs that have not achieved P < 10−5 in an external study	35 (51% error rate)	8 (21% error rate)	22 (42% error rate)
Overall	68	39	53
Pairwise P-value (two-sided Fisher’s exact test)	DeepCOMBI versus RPVT = 0.002		DeepCOMBI versus COMBI = 0.0435