Table 1.

Comparison of eight-letter vs. four-letter predictions of acceptor and donor sites for different datasets. Where application of eight-letter alphabet enhances the prediction compared to the conventional four-letter alphabet, the data pair is shown in bold.

			Markov Models:
			Zero-order			First-order			Second-order
Dataset	Alphabet	Site	CC	Sp	Sn	CC	Sp	Sn	CC	Sp	Sn
AraClean:	4-	3'	0.8540*	0.8923	0.9039	0.8821	0.8906	0.9504	0.9011	0.9138	0.9511
	8-	3'	0.8536*	0.8911	0.9048	0.9002	0.9043	0.9611	0.9414	0.9456	0.9744
	4-	5'	0.8832	0.8871	0.9565*	0.8917	0.8989	0.9550	0.9066	0.9019	0.9737
	8-	5'	0.8817	0.8853	0.9565*	0.9088	0.9010	0.9781	0.9422	0.9351	0.9874
AtGS:	4-	3'	0.7025	0.8338	0.8763	0.7450	0.8539	0.8976	0.7787	0.8769	0.9053
	8-	3'	0.7002	0.8359	0.8705	0.7455	0.8495	0.9041	0.7943	0.8776	0.9219
	4-	5'	0.8029	0.8846	0.9224	0.8224	0.8948	0.9311	0.8455	0.9048	0.9440
	8-	5'	0.7986	0.8787	0.9252	0.8222	0.8997	0.9250	0.8576	0.9138	0.9464
HsGS:	4-	3'	0.7848	0.8649	0.9287	0.8167	0.8831	0.9403	0.8450	0.8979	0.9524
	8-	3'	0.7843	0.8635	0.9299	0.8207	0.8888	0.9375	0.8574	0.9085	0.9530
	4-	5'	0.8117	0.8779	0.9409	0.8449	0.8988	0.9510	0.8740	0.9139	0.9639
	8-	5'	0.8070	0.8788	0.9348	0.8446	0.8935	0.9570	0.8842	0.9249	0.9619
BG570:	4-	3'	0.9002	0.9046	0.9597	0.9163	0.9276	0.9570	0.9346	0.9347	0.9761
	8-	3'	0.8981	0.8996	0.9626	0.9174	0.9137	0.9748	0.9427	0.9512	0.9696
	4-	5'	0.8845	0.9056	0.9369	0.9105	0.9153	0.9644	0.9353	0.9351	0.9780
	8-	5'	0.8841	0.9070	0.9346	0.9136	0.9218	0.9615	0.9437	0.9406	0.9838
HMR195:	4-	3'	0.8812	0.8850	0.9596	0.9054	0.9169	0.9567	0.9406	0.9432	0.9774
	8-	3'	0.8832	0.8868	0.9603	0.9120	0.9172	0.9661	0.9556	0.9525	0.9885
	4-	5'	0.8775	0.8980	0.9378	0.9204	0.9290	0.9645	0.9532	0.9526	0.9850
	8-	5'	0.8783	0.9224	0.9095	0.9275	0.9458	0.9558	0.9633	0.9618	0.9892

* Data pairs with insignificant differences (Mann-Whitney test).