Learning collective multicellular dynamics with an interacting mean field neural SDE model.

Area of Science:

• Computational Biology
• Systems Biology
• Genomics

Background:

• Temporal single-cell RNA sequencing (scRNA-seq) enables studying dynamic biological processes.
• Integrating cell-cell interactions (CCIs) into computational models of cellular dynamics is challenging due to data complexity.

Purpose of the Study:

• To present scIMF, a novel deep-generative model for learning collective multicellular dynamics from scRNA-seq data.
• To address the challenge of integrating complex CCIs into models of cellular systems.

Main Methods:

• scIMF utilizes a deep-generative Interacting Mean Field model based on the McKean-Vlasov stochastic differential equation framework.
• Incorporates a cell-wise attention mechanism to capture nonlocal and asymmetric CCIs in high-dimensional gene expression data.

Main Results:

• scIMF accurately reconstructs gene expression at unobserved time points and infers cellular velocities across diverse scRNA-seq datasets.
• The model outperforms existing state-of-the-art methods in temporal data analysis.
• scIMF reveals biologically interpretable, non-reciprocal cell interaction patterns.

Conclusions:

• scIMF provides a robust framework for modeling collective multicellular dynamics and complex intercellular relationships.
• The model offers new insights into non-equilibrium biological systems by uncovering non-reciprocal cell communication.