Table 2:

Matching of novel phenotypes on a GeneMatcher validation set.

Gene	PMID	Total families (Subjects)	Connected families (subjects)a
			Top-10	Top-30
BPTF 32	28942966	6 (6)	0 (0)	2 (2)
CCDC47 33	30401460	4 (4)	0 (0)	0 (0)
CHAMP1 34	27148580	4 (4)	2 (2)	4 (4)
CHD4 35	27616479	3 (3)	0 (0)	0 (0)
DDX6 36	31422817	4 (4)	4 (4)	4 (4)
EBF3 37	28017373	6 (7)	0 (0)	0 (0)
FBXO11 38	30679813	17 (17)	5 (5)	9 (9)
HNRNPK 39	26173930	3 (3)	3 (3)	3 (3)
KDM3B 40	30929739	9 (9)	0 (0)	2 (3)
LEMD2 4	30905398	2 (2)	2 (2)	2 (2)
OTUD6B 41	28343629	4 (9)	3 (4)	3 (6)
PACS2 42	29656858	6 (6)	0 (0)	2 (2)
TMEM94 22	30526868	6 (10)	5 (8)	6 (10)
WDR37 43	31327508	4 (4)	2 (2)	3 (3)
ZNF148 44	27964749	3 (3)	0 (0)	0 (0)
Total	-	79 (91)	26 (30)	40 (48)
Average	-	-	32.91% (32.97%)	50.63% (52.75%)

^aNumber of families (subjects) matched by a photo from another family in the top-10 or top-30 rank.

In the discovery mode for novel phenotypes (second use case), all cases in the gallery are without diagnosis. For the performance readout, only the correct disease gene of a match is revealed. As an example, for individuals of the TMEM94 study (shown in bold in the table), eight out of ten subjects had an image from another family within the top-10 rank, and five of the six families had at least one subject from another family in their top-10 rank. All subjects and families matched within the top 30. This table is based on the ranks from the similarity matrices in Figure 4 and Supplementary Figure 8. The accuracy of connected subjects corresponds to the accuracy of using Enc-F2G on the F2G-rare test set (shown in Table 1), but in discovery mode in a gallery of almost the same size as F2G rare gallery set.