We will develop a set of statistical techniques that improve the prediction of the response of mutated Human Immunodeficiency Virus Type 1 (HIV-1) to anti-retroviral therapy. These techniques will have applicability to a wide array of clinical decisions beyond HIV where genotypic and phenotypic data may be used to predict patient outcomes. Current approaches to predicting clinical outcomes of anti-retroviral therapy, for the purpose of drug regimen selection, do not demonstrate strong concordance1,2. Two key reasons for these shortcomings are the lack of suitable statistical models that accurately characterize the affect of the many different combinations of mutations, and the lack of statistically significant samples of HIV/AIDS patients whose data includes baseline clinical status, treatment history, scans of the viral genomes, and clinical outcomes. It is often the case that the potential genetic and phenotypic predictors and their interactions result in a number of independent variables (IVs) that is large relative to the number of measured outcomes. The problem of limited data is compounded by the many different combinations of ART and viral mutations that are encountered in practice. To address this problem in part, Gene Security Network has already developed a system that facilitates the aggregation of genetic and clinical data sets into standardized computable format, using software cartridges tailored to each local source of data. This proposal focuses on the other aspect of the solution to limited data, namely the development of novel statistical methods that improve outcome prediction when the number of potential predictors is large compared to the number of measured samples. The Gene Security Network team has developed and published3,4 algorithms that create sparse models for predicting in vitro drug response based on viral genetic sequences. These approaches performed better than all previously published algorithms.1,5-7 Our aims for the phase I project are i) to extend the theoretical technique that underlies the superior performance of our models using in vitro data and ii) to implement and obtain FDA approval for a system at the Stanford Virology Lab that will produce enhanced in vitro drug susceptibility reports for treating physicians. In the potential phase II project, we will extend our techniques for modeling the in vitro response to modeling the more complex in vivo responses measured in terms of CD4+ and viral load counts. We would again seek FDA approval for the enhanced reporting system for phase II, which will rank regimens for the treating physician based on genetic and clinical data. A clinical trial would be hosted by the Stanford Virology Lab to demonstrate efficacy of the phase II enhanced reports in terms of improved outcomes and/or reduced cost of treatment. This project will improve the statistical methods used in predicting HIV-1 drug response, and will facilitate the use of these enhanced models for better treatment decisions. The statistical methods developed, and the software system for enhanced reporting based on laboratory cartridges, will have application to many diseases beyond HIV/AIDS where complex geno-pheno models can be used to enhance treatment decisions. Given the rollout of ART drugs around the world,25 the emergence of resistant strains of the virus is inevitable, both due to the low genetic barrier to resistance27-33 and to poor drug adherence.34 The rapidly decreasing cost of HIV genetic sequencing35 makes the selection of drugs based on viral genetic sequence an attractive option, rather than the more costly and involved in vitro phenotype measurement.36,37 However, current models for predicting response to anti-retroviral therapy do not demonstrate strong concordance, and physicians interpretation of resistance reports for drug regimen selection vary considerably.1,2,5 This project will improve the statistical methods for predicting HIV-1 drug response, and will facilitate the use of these enhanced models for better treatment decisions. The statistical methods developed, and the software system for enhanced reporting based on laboratory cartridges, will have application to many diseases beyond HIV/AIDS where complex genotype-phenotype models can be used to enhance treatment decisions
