Computational protocol: HIV-1 Coreceptor Usage Assessment by Ultra-Deep Pyrosequencing and Response to Maraviroc

Similar protocols

Protocol publication

[…] First, the quality of the experiments was analyzed and demultiplexing was performed with our in-house software Pyroclass []. Four methods analyzing and reporting UDPS data were then compared. Method 1: consensus sequences were generated from UDPS data at D0 and used as a surrogate of population sequencing; they were interpreted by means of the geno2pheno[coreceptor] algorithm (Max-Planck-Institute), based on recommendations from the European Consensus Group on Clinical Management of HIV-1 Tropism Testing (false-positive rate: 5%) []. Method 2: the consensus sequences were interpreted by means of the 11/25/charge rule using a module included into PyroTrop, an in-house software developed for this application. Method 3: the full set of UDPS sequences was interpreted with geno2pheno[454], an adaptation of the geno2pheno algorithm to multiple sequence files, such as those obtained with UDPS []. This algorithm uses a cutoff value of 2% for the risk of maraviroc failure []. Method 4: the full set of UDPS sequences was interpreted with our in-house software PyroTrop. Briefly, the previously described PyroMute software [] was modified to assess each validated V3 sequence by means of the 11/25/charge algorithm. PyroTrop uses three quality filters to eliminate unreliable sequences, including sequences with a PHRED quality score <20 [], sequences not spanning the full-length HIV V3 loop, and sequences that are too divergent from the HXB2 strain reference HIV sequence (GenBank accession number K03455). The latter filter was used to eliminate eventual contaminations by non-HIV sequences. Then, PyroTrop ascribes each valid V3 sequence to either CCR5 or CXCR4 by means of the 11/25/charge algorithm [] and provides the respective percentages of these two populations within the patient’s quasispecies (2% of CXCR4 variants was set as the detection threshold) [].The results are provided as the respective proportions of CCR5 and CXCR4 populations in the patient’s viral quasispecies, with a predefined lower limit of detection of 2% []. Additional modules were included in PyroTrop to assess the frequency of each amino acid at each position, allowing for “machine learning” on the full data set. [...] Sensitivity, specificity, positive predictive value (PPV, defined as the capacity of the test to correctly predict a significant viral level decrease), negative predictive value (NPV, defined as the capacity of the test to predict the lack of significant viral level decrease) and Receiver Operating Characteristic (ROC) curves were compared for the 4 methods by means of McNemar test and area under the curves (AUROC) comparisons. In addition, supervised classification using random forest, regression using lasso, and decision tree methods were used to assess the relationship between amino acid polymorphisms and the response to maraviroc. For each selected candidate polymorphism, ROC curves were plotted and the AUROCs were compared by means of bootstrap methods using an AUROC of 0.5 as non-predictive []. Analyses were made with packages randomForest, rpart, MASS, ROCR, cluster, pvclust, glmnet and pROC in R language software by means of RGui (64-bit) (v2.14.1) [,]. […]

Pipeline specifications

Software tools Geno2pheno, randomforest, Pvclust
Applications Immune system analysis, Miscellaneous
Organisms Human immunodeficiency virus 1, Human immunodeficiency virus 2, Homo sapiens
Diseases HIV Infections