Computational Models for Prediction of Response to Antiretroviral Therapies

This review describes the state-of-the-art in statistical, machine learning, and expert-advised computationalmethods for the evaluation and optimization of combination antiretroviral therapy, with respect tothe virologic outcomes in HIV-1-infected patients. Currently employed methodologies are based on theparadigm for which mutations present in patient viral genotypes, selected either by treatment or alreadytransmitted to the patient as resistant mutants, are the major drivers of virologic outcomes. Genotypicinterpretation systems have been designed with the prime objective of characterizing the resistanceto individual drugs, deriving scores from the association of viral genotypes with in vitro phenotypicdrug susceptibility or in vivo response to treatment. Nevertheless, the very large range of possible drugcombinations and of viral mutational patterns leads to an extremely complex scenario, making predictionof in vivo treatment response extremely challenging. To deal with such complexity, machine learningmethods are being increasingly explored, thanks to the availability of exponentially growing HIV databases in recent years. The combination of genotypic interpretation systems with other laboratory markers,treatment history, past clinical events, and the usage of data-driven techniques has dramatically raisedthe confidence in predicting virologic outcomes. A few of these systems have been implemented asfree web-services, indicating ranks of suitable combination antiretroviral therapy regimens given apatient’s clinical background. Future perspectives in the field foresee the extension of therapy optimizationsystems to newly approved antiretroviral drug targets and the prediction of other clinical outcomes,rather than the sole virologic response.