A Deep Learning Framework for Quantifying Dynamic Self-Organization in Myxococcus xanthus
Under starvation, Myxococcus xanthus bacteria initiate a multicellular developmental program in which cells move to form fruiting bodies and differentiate into distinct cell types. Many genes affecting this process have been identified, and it is assumed that perturbing genes within the same pathway induces similar changes in the phenotype, although those changes may be subtle or obscured by pleiotropic effects. However, these pathways cannot be systematically mapped, as there are no systematic methods for quantifying phenotype similarity. Here, we applied deep learning techniques to quantify the phenotype patterns and self-organization dynamics of 292 genetically distinct strains. By integrating ResNet and StyleGAN2 to construct a Variational AutoEncoder (VAE) and utilizing a Siamese network for phenotypic similarity metrics, we efficiently encoded high-resolution microscopy images into 13-dimensional feature vectors, capturing phenotypic variability in aggregation patterns across time and strains. Human evaluation confirmed that our models reconstructions were visually indistinguishable from real images and closely aligned with input phenotypes. Importantly, the feature space is interpretable: individual dimensions correlate with biological features such as aggregate number and size, and extrapolation along these dimensions produces predictable morphological changes. Remarkably, our model revealed that developmental phenotypes are predictable from the earliest stages before visible aggregate formation begins. This predictability held across both genetic and environmental sources of variation, suggesting fundamental constraints on developmental trajectories and indicating that subtle phenotypic variations carry important information. These results demonstrate how machine learning can reveal hidden structure in complex multicellular dynamics and provide scalable methods for phenotypic analysis without manual annotation.