Biophysically interpretable inference of cell types from multimodal sequencing data.

Area of Science:

• Single-cell genomics
• Computational biology
• Systems biology

Background:

• Multimodal single-cell genomics enables simultaneous measurement of multiple cellular processes.
• Current cell clustering methods for multimodal data often use ad hoc approaches and ignore data properties.
• Accurate cell type determination is crucial for understanding cellular heterogeneity and biological mechanisms.

Purpose of the Study:

• To develop an interpretable and consistent method for cell cluster determination in multimodal single-cell genomics data.
• To integrate different molecular measurements by modeling underlying biophysical states.
• To provide a mechanistic framework for defining cell clusters based on shared cellular processes.

Main Methods:

• Developed mechanistic K-means (meK-means), a novel clustering algorithm for multimodal data.
• Integrated nascent and mature mRNA measurements using a unifying model of transcription.
• Leveraged causal, physical relationships between molecular modalities to define clusters.

Main Results:

• meK-means effectively clusters cells by integrating multiple data modalities.
• The method identifies shared transcription dynamics that govern observed molecular counts.
• Clusters are defined by the parameters of underlying cellular processes, offering mechanistic interpretability.

Conclusions:

• meK-means provides a principled approach to multimodal single-cell data integration and clustering.
• This method enables deeper mechanistic studies of cellular heterogeneity and dynamics.
• It offers a new paradigm for defining cell clusters based on biophysical states rather than just molecular profiles.