Deep learning enables quantitative subcellular analysis of plant-microbe interfaces

Abstract

Specialized host-microbe interfaces are central to cellular interactions in plants. Intracellular structures such as haustoria formed by filamentous pathogens mediate nutrient exchange and effector delivery to host cells. Despite their biological importance, the lack of quantitative frameworks has largely confined the study of these interfaces to qualitative observations, limiting our ability to compare infection strategies, cellular responses, and spatial organization across cells and tissues. Here, we present HFinder , a deep learning-based framework for automated detection, segmentation, and quantitative analysis of plant-microbe interfaces in confocal images. Using an object-centric deep learning approach, HFinder enables robust identification of haustoria, microbial hyphae, and host organelles across diverse imaging conditions and pathosystems. We demonstrate that this framework supports quantitative analyses of subcellular processes at host-microbe interfaces, including effector secretion, perturbation of host cellular processes, and immune receptor accumulation at haustoria. HFinder provides a practical and scalable solution for the systematic digitalization of plant infection imaging data and establishes a general framework for quantitative studies of cellular dynamics at host-microbe contact zones.

Code and data availability

The paper's HFinder pre-trained models (trained phenotyping models/checkpoints) are explicitly deposited on Zenodo with a public DOI matching an allowed URL. The training image dataset is also stated to be publicly available on Zenodo, but no separate authors' URL for it is given in the supplied blocks, so only the pre