Spliceator: multi-species splice site prediction using convolutional neural networks

BMC Bioinformatics

Table 4 Composition of the 8 datasets

Dataset	Quality of sequences	No. of positive sequences			No. of negative sequences	Type of negative sequences	Ratio
Dataset	Quality of sequences	Donor		Acceptor	No. of negative sequences	Type of negative sequences	Ratio
AS_0	Unconfirmed and confirmed	12,000	12,000		12,000	FP only	1:1
AS_1					12,000	4000 exons, 4000 introns and 4000 FP	1:1
AS_2					24,000	FP only	1:2
AS_10					120,000	FP only	1:10
GS_0	Confirmed	10,973	11,179		11,000	FP only	1:1
GS_1					11,000	3650 exons, 3650 introns and 3700 FP	1:1
GS_2					22,000	FP only	1:2
GS_10					110,000	FP only	1:10

Composition of the 8 datasets used to study the impact of (i) the type of negative examples (only FP sequences vs. heterogeneous data with exons, introns and FP sequences), (ii) the ratio of positive to negative examples (1:1, 1:2 and 1:10), (iii) data quality (‘Confirmed’ and ‘Unconfirmed’ sequences in the AS datasets vs. only Confirmed sequences in the GS datasets
FP, False Positive; GS, Gold Standard; AS, All Sequences

ISSN: 1471-2105