Enhancing HIV Drug Resistance Prediction Using Bidirectional LSTM Neural Networks

One major obstacle that humanity faces today is the ability of viruses to develop resistance to antibiotics or other drugs used in treating the diseases. In the current study, we utilise a specifically developed Bidirectional Long Short-Term Memory (LSTM) neural network to identify HIV drug resistant strains in an effort to help clinicians identify the most appropriate treatment options so that the development new resistant strains is prevented as much as is possible. By incorporating the Bidirectional LSTM model, temporal dependency analysis is improved and deepened in both forward and backward directions, improving the model’s performance and interpretability. The accuracy of the model is measured by MAE and R² where the result from MAE is 16.151 of PI (Protease Inhibitors), 14.703 of NRTI (Nucleoside Reverse Transcriptase Inhibitors) and 18.815 of NNRTI (Non-Nucleoside Reverse Transcriptase Inhibitors) while result from R² is 0.914 of PI, 0.942 of NRTI and 0.885 of NNRTI. Consequently, these results suggest the custom model may be useful to refine drug resistance predictions to progress the accuracy of HIV treatment.