Identification of models describing gene expression data leveraging machine learning methods.

Area of Science:

• Systems Biology
• Computational Biology
• Machine Learning

Background:

• Mechanistic ordinary differential equation (ODE) models are crucial for understanding cellular processes and formulating biological hypotheses.
• Data-driven inference of these models is increasing, but integrating machine learning (ML) without losing interpretability remains a challenge.
• Gene regulatory networks (GRNs) govern complex intracellular dynamics, including cell differentiation.

Purpose of the Study:

• To present a framework leveraging neural networks for identifying interpretable, data-driven mechanistic ODE models of GRNs.
• To utilize ML to suggest novel connections within GRNs, improving model accuracy and biological insight.
• To generate testable hypotheses regarding the dynamics of intracellular processes.

Main Methods:

• Development of a framework integrating neural networks with mechanistic ODE modeling.
• Application of a graph autoencoder model to infer and suggest connections in GRNs.
• Validation of the approach on time-dependent intracellular processes, such as cell differentiation.

Main Results:

• Demonstrated successful application of the graph autoencoder to suggest novel GRN connections.
• Showcased how improved graph structures enhance the identification of dynamical systems.
• Generated new hypotheses regarding the dynamics of the identified cellular processes.

Conclusions:

• The proposed framework effectively uses neural networks to identify interpretable mechanistic models of GRNs.
• This approach facilitates the generation of novel, data-driven hypotheses for complex biological systems.
• The integration of ML offers a powerful tool for advancing systems biology and understanding cellular mechanisms.