Table 2.

Effect of each selected feature on the prediction.

Feature	Estimate	Std. Error	t value	p-value	specificity
CodingAmniote, α = 0	0.0030645	0.0005186	5.910	3.47e-09	0.473
AssociationBreaks, β = 2	-0.0150509	0.0022701	-6.630	3.42e-11	0.572
CodingMetatheria, α = 2	1.3915142	0.2395547	5.809	6.36e-09	0.611
NonCodingMetatheria, α = 3	-0.0338532	0.1649918	-0.205	0.83743	0.625
CodingMetatheria, α = 0	-0.0079558	0.0014829	-5.365	8.15e-08	0.629
NonCodingMetatheria, α = 2	-0.0154798	0.0132948	-1.164	0.24429	0.630
CodingAmniote, α = 3	-4.6993169	0.9579974	-4.905	9.38e-07	0.634
AssociationBreaks, β = 3	0.0069784	0.0024624	2.834	0.00460	0.645
NonCodingMetatheria, α = 4	0.0302276	0.1104565	0.274	0.78435	0.645

Features are listed in the order in which they were selected by the forward feature selection procedure. The coefficient estimate, standard error, t-value, and p-value reported are those obtained for the linear regression with all 9 features. The specificity reported is that of the predictor built using the features starting from the 1st row down to the current row. The specificity is calculated for a sensitivity of 0.75. For example a specificity of 0.645 means that 75% of breakpoints are comprised within 35.5% of the total length of inter-marker regions used for the analysis.