MicroICS: predictive phenotyping of Listeria monocytogenes biofilms from three-dimensional structural features

Biofilms underpin microbial survival, yet their three-dimensional structure remains difficult to quantify systematically. We present the Microbial Image Classification Suite (MicroICS), an open-source framework for predictive phenotyping of microbial communities from 3D biofilm images. MicroICS consists of three independent but interoperable modules: feature extraction from 3D biofilm images, machine learning-based classification, and an inference module for applying trained models to new, unseen images. Each module can be run independently, and the classification module accepts externally generated features, enabling integration with existing quantitative image analysis tools. As a pilot study, we demonstrate the framework using eight epidemiologically diverse Listeria monocytogenes strains, with strain differentiation and trait-based grouping as proof-of-principle tasks. MicroICS extracted over 2,700 quantitative structural features from Syto 9-labelled biofilm images. An optimised random forest model achieved human baseline-level accuracy in strain classification on previously unseen images, including biofilms experimentally perturbed by food extracts. BiofilmQ-derived features incorporated into the classification module yielded higher accuracy with fewer features than either tool alone, confirming the framework's extensibility. Extending the framework to clonal complex-based clinical grouping demonstrates utility beyond strain identity. MicroICS is applicable to any organism or condition for which suitable 3D biofilm images can be obtained.