Using enhancing signals to improve specificity of ab initio splice site sensors

In this paper, we describe a new approach to improve the precision of splice site annotation in human genes. The problem is known to be extremely challenging since the human splice signals are highly indistinct and frequent cryptic sites confuse signal sensors. There is a strong evidence that exonic splicing enhancers (ESE) and exonic splicing silencers (ESS) influence commitment to splicing at early stages. We propose the use of a naive Bayesian network (BN) combined with Boltzmann machine splice sites sensor, to improve the specificity of splice site prediction. The SpIiceScan program is implemented to demonstrate feasibility of specificity enhancement based on ESE/ESS signals interactions. SpliceScan is more sensitive than GeneSplicer and NNSplice for the same specificity. The designed method is of particular value for ab initio gene annotation.