Quantitative prediction of integrase inhibitor resistance from genotype through consensus linear regression modeling

Table 1 Performance of RAL first/second order linear model

			Correlation with measured phenotype (R²)		FIRST *vs.* SECOND order (p-value)
		N	FIRST order	SECOND order	FIRST *vs.* SECOND order (p-value)
train	ALL	991	0.96	0.97	0.0019
validation	ALL	67	0.79	0.80	0.3753
population seen	ALL	144^a	0.90	0.90	0.1530
population seen	NOMIX	105	0.95	0.95	0.3558
	MIX > = 1	39	0.73	0.71	0.0373
	MIX > = 1 (no outliers)	36	0.84	0.81	0.0063
population unseen	ALL	171	0.76	0.78	0.0992
population unseen	NOMIX	153	0.79	0.81	0.1358
	MIX > = 1	18	0.59	0.58	0.7482
	MIX > = 1 (no outliers)	16	0.78	0.78	0.8819

^aFor 144 out of 153 samples a matched genotype-phenotype was available.
Performance in R² of first order and second order linear model on the clonal geno/pheno training database, on the externally PhenoSense measured IN site-directed mutants (validation) and on population data, seen and unseen. After Bonferroni correction for multiple testing no significant differences in R² performance are seen between the first and second order model on the validation dataset or population seen/unseen datasets (p-values listed are unadjusted).

ISSN: 1743-422X