Table 2.

Summary of the variant dataset

Outcome	Donor		Acceptor		Total
	Non-canonical	Canonical	Non-canonical	Canonical
Cryptic site creation	143	7	191	13	354
Canonical site disrupted	1,125	3,576	360	2,882	7,943
Other	7	0	10	0	17
Totals	1,275	3,583	561	2,895	8,314

We created a collection of splice variants by curating literature. During curation, we recorded metadata regarding the variant pathomechanism and the observed outcome. Based on the outcome, we categorized the variants into two major groups: (1) variants disrupting canonical splice sites and leading to activation of a cryptic splice site, or to exon skipping, and (2) variants that activate cryptic splice. 73,203 neutral variants were used as negative training examples. There were 4,858 donor variants and 3,456 acceptor variants. Of these, 1,836 were non-canonical and 6,478 were canonical (i.e., located at the ±1 or ±2 positions).