CODEX, a neural network approach to explore signaling dynamics landscapes.

Area of Science:

• Cellular and Molecular Biology
• Computational Biology
• Systems Biology

Background:

• Live cell biosensor studies generate vast amounts of single-cell, multi-dimensional time-series data on cell signaling dynamics.
• Traditional analysis relies on predefined features, which may miss crucial information due to lack of prior biological system knowledge.

Purpose of the Study:

• To introduce CODEX, a data-driven approach using convolutional neural networks (CNNs) for unbiased pattern identification in cell signaling time-series data.
• To enable discovery of novel biological insights without requiring a priori system information.

Main Methods:

• Development and application of CODEX, a CNN-based framework for analyzing high-dimensional single-cell trajectory data.
• Utilizing explainable AI to derive biological insights from CNN predictions.
• Visualization of single-cell trajectories in a low-dimensional space, identification of prototypic trajectories, and extraction of distinctive motifs.

Main Results:

• CODEX successfully identifies complex patterns in cell signaling dynamics.
• The method provides novel insights into ERK and Akt signaling pathways in response to growth factors.
• CODEX recapitulates known findings in p53 and TGFβ-SMAD2 signaling pathways, validating its efficacy.

Conclusions:

• CODEX offers a powerful, data-driven alternative to feature-based analysis of cell signaling dynamics.
• The approach enhances the discovery of biological mechanisms by uncovering previously unrecognized patterns.
• CODEX facilitates a deeper understanding of cellular communication and responses to stimuli.