A Deep Learning Framework for Studying Single-Cell Bacterial Dynamics

Document Type

Poster

Publication Date

4-17-2026

Abstract

Fluorescence microscopy combined with microfluidic platforms enables high-resolution studies of bacterial communities in engineered environments. However, challenges such as data volume, image drift, and fluorescence variability limit reproducibility and automation. Existing tools for bacterial image analysis either rely on traditional segmentation methods that require manual parameter tuning or offer standalone deep learning algorithms without integrated analysis capabilities, forcing users to assemble separate workflows for fluorescence quantification, morphological classification, and multi-file data management. Here, we developed an open-source Python-based graphical framework that addresses these gaps by providing six selectable deep learning segmentation models within a single interface, native multi-file ND2 support for long-term microfluidic experiments, and integrated single-cell fluorescence and morphological analysis. It extracts morphological descriptors from segmented masks and classifies cells into geometric categories using a hierarchical decision tree. We validated the framework using a two-population sender-receiver bacterial community grown in monolayer chambers under controlled flow conditions. Single-cell fluorescence analysis revealed temporal synchronization between sender and receiver populations and demonstrated that segmentation-based quantification preserves biological heterogeneity obscured by bulk pixel-averaged measurements. Expression changes occurred through shifts in population distributions rather than uniform intensity increases, enabling quantification of activation thresholds and induced cell fractions over time.

This document is currently not available here.

Share

COinS