Correlation between rules-based interpretation and virtual phenotype interpretation of HIV-1 genotypes for predicting drug resistance in HIV-infected individuals.

Drug resistance testing provides useful information for managing HIV-infected patients. Phenotyping could add complementary information to genotyping and occasionally be more useful, although is less available to clinicians. Large paired geno-pheno databases have allowed the prediction of phenotypes from genotypes. However, the accuracy of these virtual phenotypes (vPT) in a clinical setting has not been well assessed yet. We analyzed the concordance between vPT and interpreted genotype (GT) in 105 samples belonging to treatment-experienced HIV-infected patients. A high concordance was seen when examining both non-nucleoside reverse transcriptase inhibitors (NNRTI) and protease inhibitors (PI) (r = 0.95 either), while it was lower for nucleoside analogs (r = 0.79). The drugs with lower concordance were abacavir (71.1%), tenofovir (71.5%) and didanosine (71.9%). In 20% of specimens (21/105), the vPT did not provide results for all approved drugs. These were mainly samples with a high number of drug resistance mutations or rare genotypes, which seem to be underepresented in the VircoNET database. Overall, there is good correlation between vPT/GT, especially for PI and NNRTI. The inclusion of additional sequences in the VircoNET database, mainly those derived from heavily treatment-experienced patients and/or from patients failing the most recently approved drugs might improve its performance.