Accurate splice site prediction using support vector machines

BMC Bioinformatics

Table 1 Performance evaluation (auROC and auPRC scores) of six different methods on the NN269 and DGSplicer Acceptor and Donor test sets. MC denotes prediction with a Markov Chain, EBN the method proposed in [36], and MC-SVM the SVM based method described in [32] (similar to [21]).

	MC	EBN	MC-SVM	LIK	WD	WDS
NN269
Acceptor
auROC	96.78	-	96.74^†	98.19	98.16	98.65
auPRC	88.41	-	88.33^†	92.48	92.53	94.36
Donor
auROC	98.18	-	97.64^†	98.04	98.50	98.13
auPRC	92.42	-	89.57^†	92.65	92.86	92.47
DGSplicer
Acceptor
auROC	97.23	95.91*	95.35*	-	97.50	97.28
auPRC	30.59	-	-	-	32.08	28.58
Donor
auROC	98.34	96.88*	95.08*	-	97.84	97.47
auPRC	41.72	-	-	-	39.72	35.59

The remaining methods are based on SVMs using the locality improved kernel (LIK) [21], weighted degree kernel (WD) [28] and weighted degree kernel with shifts (WDS) [34]. The values marked with an asterisk were estimated from the figures provided in [32]. The values marked with † are from personal communication with the authors of [32].

ISSN: 1471-2105