Article citationsMore>>

Xu, J.Y., Pradhan, A. and Duraisamy, K. (2021) Conditionally Parameterized, Discretization-Aware Neural Networks for Mesh-Based Modeling of Physical Systems. Advances in Neural Information Processing Systems, 34, 1634-1645.

has been cited by the following article:

• TITLE: Hybrid Mechanistic-Machine Learning Models of Neurotransmitter Dynamics: Integrating Biological Knowledge with Deep Learning for Drug Response Prediction and Disease Mechanism Discovery

  AUTHORS: Rocco de Filippis, Abdullah Al Foysal

  KEYWORDS: Neurotransmitter Dynamics, Hybrid Modelling, Physics-Informed Neural Networks, Dopamine, Serotonin, Drug Response Prediction, Neural Odes, Precision Psychiatry, Computational Neuroscience

  JOURNAL NAME: Open Access Library Journal, Vol.13 No.5, May 27, 2026

  ABSTRACT: Understanding neurotransmitter dynamics is fundamental to predicting psychiatric drug response and elucidating disease mechanisms. Traditional mechanistic models capture biological principles but struggle with parameter estimation and individual variability, while pure machine learning approaches lack biological interpretability. We present a hybrid framework integrating physics-informed neural networks with mechanistic models of dopaminergic and serotonergic systems. Our approach combines ordinary differential equation-based kinetics with neural ODE architectures, enabling data-driven parameter inference while preserving biological constraints. We evaluated the framework using simulated and experimental data from antipsychotic and antidepressant studies. The hybrid model achieved prediction accuracy of R-squared equals 0.89 for drug response, outperforming both pure mechanistic models (R-squared equals 0.72) and pure machine learning approaches (R-squared equals 0.78). Parameter sensitivity analysis identified release rate and reuptake kinetics as key determinants of drug response variability. Multi-scale integration demonstrated improved prediction accuracy across molecular, synaptic, and circuit levels. Personalized dosing optimization using the hybrid framework improved treatment response by 23% compared to standard dosing protocols. These findings establish hybrid mechanistic-machine learning models as a powerful approach for understanding neurotransmitter dynamics, with significant potential for precision psychiatry and drug development applications.