653 
Neural Networks Are More Accurate Predictors of Virologic Response to ART yhan Rules-based Genotype Interpretation Systems
Brendan Larder*1, A Geretti2, L Monno3, C Torti4, A Revell1, D Wang1, R Harrigan5, J Montaner5, and H Lane6
1HIV Resistance Response Database Initiative, London, UK; 2Royal Free Hosp Cohort, London, UK; 3Univ of Bari, Italy; 4Univ of Brescia, Italy; 5BC Ctr for Excellence in HIV/AIDS, Vancouver, Canada; and 6NIAID, NIH, DHHS, Bethesda, MD, US
Background:
HIV genotyping with
rules-based interpretation is used to help predict the virologic responses from
HAART options. The RDI has shown that artificial neural networks (ANN) can
predict virologic responses to HAART, based upon genotypes. Here we assess
directly the accuracy of these approaches by comparing ANN predictions with
genotypic sensitivity scores (GSS) from rules-based interpretation systems in
terms of correlations with actual virologic responses (DVL).
Method:
A committee of 5 ANN
models were trained to predict DVL from
genotype (55 resistance mutations), the drugs in the regimen and baseline viral
load using data from 2983
clinical treatment change episodes (TCE) from the RDI database. We used 3 independent test sets (each of 25 randomly selected TCE): 2 of clinical cohort data (RDI database and
Royal Free Hospital) and 1 from the PhenGen trial of
genotyping vs phenotyping. Accuracy was
tested by providing the ANN with baseline data from the test TCEs and
correlating the models’ averaged predictions of DVL with the
actual DVL values. Total and
normalized GSS were derived for the test TCE using Stanford (HIVdb 4.1.2), ANRS (v2004.09), and Rega
(v6.2) rules-based interpretation systems. GSS and Stanford mutation scores
(SMS), based on assumed impact on susceptibility to each drug, were also
correlated with actual DVL for the
test TCEs.
Results:
Correlations between ANN
predictions and actual DVL values for
the RDI test set gave an averaged r2
value of 0.80. SMS correlated with an r2 of 0.22 and GSS 0.19 to 0.31.
The PhenGen test data yielded r2 values of 0.32 for the ANN, compared with 0.16 for
SMS and 0.10 to 0.24 for GSS. Royal Free data gave r2 values of 0.39 for ANN compared with 0.00 for SMS and
0.00 to 0.04 for GSS. Correlations between the rules systems were substantial (r2 = 0.41 to 0.93),
highlighting their similarity. Correlations between the rules systems and the
ANN were mostly fairly weak and inconsistent in direction (r2 = 0.00 to 0.39).
Conclusions:
ANN were considerably
more accurate predictors of virologic response than rules-based systems with
all 3 test sets, explaining up to 80% of the variance in DVL vs
31%. The ANN were more accurate predicting virologic responses for patients
from clinics that have provided data used in training the ANN than for those
from ‘unseen’ clinics. The results suggest that ANN may have considerable
utility as treatment decision-making tools.
|
