Poster presentation at:
10th European AIDS Conference/EACS
17th November 2005 - 20th November 2005
Dublin, Ireland
Larder BA1, Revell AD1, Wang D1, Harrigan R2, Montaner J2, Wegner S3, and Lane C4.
1: HIV Resistance Response Database Initiative, London, UK; 2: BC Centre for Excellence in HIV/AIDS Vancouver, Canada; 3: US Military HIV Research Program, Rockville, MD, USA; 4: National Institute of Allergy & Infectious Diseases, Bethesda, MD, USA.
Background
Genotyping with rules-based interpretation is commonly used to help predict the virological responses available from alternative antiretroviral combinations. Our previous studies demonstrated that artificial neural networks (ANN) can predict virological response to combination therapy from genotype. Here we compare directly the accuracy and utility of these approaches by comparing ANN predictions with genotypic sensitivity scores (GSS) from rules-based systems in terms of correlations with actual virological responses ( ΔVL).
Method
ANN models were trained to predict ΔVL with data from 2,983 treatment change episodes (TCEs), using the following input variables: 55 resistance mutations, drugs in new regimen, baseline viral load and time to follow-up. Accuracy was tested by providing the models with input variables from 27 randomly-selected independent TCEs and correlating their predictions with actual ΔVL.
Total and normalized GSS were derived for the 27 test TCEs using Stanford (HIVdb 4.1.2), ANRS (v2004.09) and Rega (v6.2) rules-based interpretation systems. Stanford awards points for each mutation according to the impact on susceptibility to each drug and these scores were also used as a resistance measure for each test regimen. The GSS and aEUR~Stanford mutation' scores were correlated with actual ΔVL for the test TCEs.
Results
Correlations with actual ΔVL data for the test TCEs gave values of 0.75 for ANN predictions, 0.22 for Stanford mutation scores and between 0.13 (Rega) and 0.29 (ANRS) for GSS scores. Correlations between the GSS from the different rules systems gave r@ values between 0.83 and 0.91.
Conclusions
ANN were considerably more accurate predictors of virological response to combination therapy than rules-based GSS, explaining 75% of the variance in ΔVL vs 13 to 29%. Scatterplots indicated that ANN were able to differentiate between regimens that GSS scores indicated would be equally effective. These data suggest that ANN may have significant utility as treatment decision-making tools.
