ODEs vs LSTM-RNNs: Modeling HIV Viral Dynamics
HIV-1 is a virus that chronically infects over 30 million individuals globally each year and is the cause of the chronic condition of AIDS. In a paper done by Perelson et al, they found that a phagocytosis based logistic clearance model (PLCM) (mediated by antibody dependent cellular phagocytosis (ADCP)) as the dominant mechanism of action best recapitulated the data provided by a publicly available clinical trial done by the NIH in Lynch et al. The PLCM captures the viral dynamics and recapitulates the data well, but one may require a good understanding of mathematics and microbiology as well as the time to isolate the desired results of the model. In a clinical setting, this may not be the most effective manner in which to use the models prediction to make medical assessments. In this paper, we aim to train and evaluate several LSTM-RNN models in the task of predicting HIV viral dynamics using a synthetic dataset generated from running simulations of the PLCM. In comparing four LSTM-RNN models using synthetic data generated from ODE simulation, we found that, to varying degrees of success, these models were able to capture the viral dynamics generated by the system of differential equations.