Table 3.

Codon context information improves the prediction accuracy of the central codon.

Secondary structure	Random codon positions	Accuracy	Accuracy degradation using random codons	% Accuracy degradation using random codons	p value
$\beta$ mode	[] - All true codons	0.46	−	−	−
	[$-2$, $-1$, $+1$, $+2$]	0.43	0.02	5.4	0.01*
	[$-1$, $+1$]	0.45	0.01	2.6	0.1
	[$-2$, $+2$]	0.45	0.005	1.1	0.1
	[$-2$, $-1$]	0.46	0.003	0.6	0.4
	[$+1$, $+2$]	0.44	0.02	4.0	0.02*
	[$-2$ ,$+1$, $+2$]	0.43	0.03	5.9	0.001*
	[$-2$, $-1$, $+2$]	0.46	0.0	− 0.2	0.6
	[$+1$]	0.44	0.02	3.5	0.02*
	[$-1$]	0.47	− 0.01	− 2.4	0.9
	[$-2$]	0.46	0.0	− 0.4	0.7
	[$+2$]	0.45	0.01	1.5	0.1
$\alpha$ mode	[] - All true codons	0.47	−	−	−
	[$-2$, $-1$, $+1$, $+2$]	0.42	0.05	10.7	< 0.001*
	[$-1$, $+1$]	0.43	0.04	8.7	< 0.001*
	[$-2$, $+2$]	0.46	0.01	2.3	0.01*
	[$-2$, $-1$]	0.45	0.02	4.9	< 0.001*
	[$+1$, $+2$]	0.44	0.02	5.1	< 0.001*
	[$-2$, $+1$, $+2$]	0.44	0.03	6.4	< 0.001*
	[$-2$, $-1$, $+2$]	0.45	0.02	4.9	< 0.001*
	[$+1$]	0.45	0.02	4.2	< 0.001*
	[$-1$]	0.45	0.02	4.7	< 0.001*
	[$-2$]	0.46	0.01	1.3	0.07
	[$+2$]	0.46	0.01	2.3	0.002*

The accuracy with which a central codon is predicted was tested when different combinations of context positions were provided with random synonymous codon information. Random codon position indices are arranged according to the central codon which is being predicted. An empty set of random positions corresponds to using true codons in all the positions and this serves as the baseline for calculating accuracy degradation. Significant results (p value $<0.05$) are indicated with *.

Significant values are in bold.