CMINNs: Compartment model informed neural networks - Unlocking drug dynamics.

Area of Science:

• Pharmacokinetics and Pharmacodynamics (PKPD)
• Computational Biology
• Mathematical Modeling

Background:

• Traditional PKPD models struggle with complex drug dynamics like anomalous diffusion and drug trapping in heterogeneous tissues.
• Multi-compartment models, while useful, can be overly complex for drug development.
• There is a need for simplified yet comprehensive modeling approaches to predict drug behavior.

Purpose of the Study:

• To develop an enhanced PK and integrated PK-PD modeling approach using fractional calculus and time-varying parameters.
• To effectively model anomalous diffusion, drug trapping, and escape rates in heterogeneous tissues.
• To provide insights into drug dynamics in cancer, particularly with multi-dose administrations.

Main Methods:

• Integration of fractional calculus or time-varying parameters with constant/piecewise constant parameters.
• Application of Physics-Informed Neural Networks (PINNs) and fractional PINNs (fPINNs).
• Combining ordinary differential equations (ODEs) with integer/fractional derivatives and neural networks for parameter estimation.

Main Results:

• The methodology successfully models anomalous diffusion and captures drug trapping/escape dynamics.
• Enhanced prediction of drug absorption rates and distributed delayed responses.
• Unlocking new insights into drug resistance, persistence, and pharmacokinetic tolerance.

Conclusions:

• The proposed fPINN framework offers a robust and simplified (two fractional ODEs) approach to PKPD modeling.
• This method significantly improves the depiction of complex drug dynamics and drug effects.
• Findings can streamline drug development, enhance cancer therapy predictions, and inform therapeutic strategies.