|
|
Abstract
COMPARING MATHEMATICAL METHODS USING GENOTYPE TO DETECT HIV RESISTANCE TO ANTIRETROVIRAL THERAPY
A.R. Macalalad1, R. Shafer2, V. DeGruttola3, K. Mayer4, J. Griffith1, C. Schmid1, G.A. Carpenter5, T. McKenna5, M. Woods5, M. Gonzales2, H. Selker1
1Tufts-New England Medical Center, Boston, MA, USA, 2Stanford University Medical Center, Stanford, CA, USA, 3Harvard School of Public Health, Boston, MA, USA, 4Brown University, Providence, RI, 5Boston University, Boston, MA, USA
Objective: Genotyping detects HIV mutations, but discerning patterns of mutations that confer antiretroviral resistance is difficult. This study compares logistic regression with stepwise selection (LR), recursive partitioning (RP), and neural networks (NN) in detecting HIV resistance from genotype.
Methods: Genotype/phenotype pairs were taken from Stanford HIV Sequence Database. Pairs were randomly assigned into 3 sets. Models were trained on 2 sets and tested on the third. Methods included LR, RP, and two types of NN (back propagation and ARTMAP). Model performance was compared using c statistic, which measures probability that given a choice between a susceptible sequence and a resistant one chosen at random, the model will choose the correct sequence as resistant.
Results: Preliminary results detecting resistance to PI’s show that the c statistics from the different methods are comparable for each PI: APV (0.816-0.868), IDV (0.875-0.910), NFV (0.828-0.927), RTV (0.887-0.946), SQV (0.913-0.933). LR did best with IDV and NFV, but worst with RTV. Back propagation NN did best with APV, RTV, and SQV. RP did worst with APV, IDV, NFV, and SQV. Most mutations selected by LR and RP were previously associated with resistance.
Conclusions: Preliminary results show that the performance of LR, RP, and NN was similar in detecting PI resistance using genotype. Back propagation NN tended to have better performance, but mutation patterns were easier to discern from LR and RP. Further refinement of the models will better define strengths of each method, and further analysis will reveal if performance is similar with NRTI’s and NNRTI’s.
The 2nd IAS Conference on HIV Pathogenesis and Treatment
Abstract no.
803
Suggested Citation
" A.R. Macalalad , et al.
COMPARING MATHEMATICAL METHODS USING GENOTYPE TO DETECT HIV RESISTANCE TO ANTIRETROVIRAL THERAPY.
Poster:
The 2nd IAS Conference on HIV Pathogenesis and Treatment:
Abstract no.
803"
|
|
|