Inferring structural and dynamical properties of gene networks from data with deep learning

Area of Science:

• Systems Biology
• Computational Biology
• Genomics

Background:

• Reconstructing gene regulatory networks (GRNs) is crucial for understanding biological systems.
• Existing methods face challenges in accurately inferring interaction direction, type, and causality, especially with complex networks and single-cell data.

Purpose of the Study:

• To develop and validate a deep neural network (DNN) approach for robust GRN reconstruction.
• To address limitations in inferring causality, network complexity, and single-cell data analysis.
• To apply DNNs for uncovering cell fate dynamics and quantifying system energy landscapes.

Main Methods:

• Utilized deep neural networks (DNNs) for inferring causality and reconstructing GRNs from various data types.
• Compared DNN performance against established methods like Boolean networks, Random Forest, and partial cross mapping.
• Integrated DNNs with the partial self-consistent mean field approximation (PSCA) to quantify energy landscapes.

Main Results:

• DNNs demonstrated superior performance in reconstructing regulatory networks compared to existing approaches.
• Successfully inferred complex cell fate dynamics during preimplantation development using ensemble DNN models on single-cell data.
• Developed a novel data-driven method to quantify energy landscapes in gene regulatory systems.

Conclusions:

• Deep neural networks offer a powerful and versatile tool for gene regulatory network inference and systems biology.
• The proposed DNN-based framework can decipher complex dynamical mechanisms and energy landscapes from diverse datasets.
• This methodology holds promise for applications beyond gene regulation, including ecology and disease modeling.