The mutations shown are those that occurred at least 10 times in the genotype-phenotype dataset. For each mutation, the y axis indicates the magnitude of the mean coefficient of 50 LSR runs (10*5-fold cross-validation), and the error bar indicates the standard deviation from the mean. Positive coefficients indicate mutations that decrease drug susceptibility. Negative coefficients indicate mutations that increase drug susceptibility. Bars representing coefficients whose cross-validated means are >=3 standard deviations from zero are blue; other coefficient bars are white, indicating a lack of statistical significance after cross-validation.

The genotype-phenotype correlation dataset contained 1,848 isolates on which in vitro susceptibility tests were performed using the PhenoSense assay (Monogram, South San Francisco, USA): 3TV (lamivudine; n=1,813), ABC (abacavir; n=1,687), AZT (zidovudine; n=1,819), TDF (tenofovir; n=1,510)

Viruses with sequences containing mixtures at the major drug resistance mutation positions had been excluded because the presence of mixtures at these positions may confound genotype-phenotype correlations (n=489). Additionally redundant viruses obtained from the same individual that contained the same pattern of major drug resistance mutations had been excluded to minimize bias that would result from over-representing highly similar viruses (n=180).