Identifying D4T Resistance

Brendan Larder from Virco reported here that he has developed a computer system model for identifying d4T resistance. Previously, it has been widely recognized that d4T resistance was very difficult to characterize and identify. Here is his abstract.

The complexity and diversity of HIV drug resistance mutation patterns makes it difficult to interpret genotypic testing results. This is especially in the case for d4T resistance, where associating specific mutations to phenotypic resistance remains a challenge. We present a systematic method to investigate the relationship between mutation patterns and corresponding phenotypic resistance using neural networks. In this study, three neural network models were developed to investigate how mutation patterns influence d4T resistance.

One model was based on the 9 RT mutations listed in the Stanford sequence database associated with d4T resistance (62V, 69D, 69N, 69SXX, 75I, 75T, 116Y and 151M). The other were based on adding either 17 or 51 extra RT mutations present at relatively high frequency in d4T resistance samples in our relational database. To train and test these neural network models, we used a total of 2286 samples, 188 of which were randomly selected as a test data set. An optimal solution for each of the models was obtained using the same training and testing data sets.

The results demonstrated that the 9-mutation model gave a low resistance prediction rate (46%) using the independent test data set and in fact it was even difficult to obtain reasonable concordance in the training set (42%). However, the 26- and 60- mutation models could be well trained and also provided a higher prediction rate (65% and 68%, respectively) for resistance (defined as > 3-fold increase relative to a sensitive control) using the test data set. In order to discover which mutations had contributed to this improved prediction, discordant samples from the 9-mutation model were identified and the corresponding genotypes were analysed.

In total, 15 additional mutations occurred in at least 30% of these samples, including 41L, 67N, 118I, 210W, 211K, 214F and 215Y. A number of these mutations had already been included in the 26- and 60-mutation models. In conclusion, these results show that at least 26 RT mutations may play a role in d4T resistance, including AZT resistance mutations. Refinement of these models should further enhance our understanding of the genetic basis of d4T phenotypic resistance.