Training, Benchmark, Inference and Pre-trained models

Cell adhesion is a fundamental biological process underpinning various physiological and pathological phenomena, including tissue repair and cancer metastasis. Traditional assays for assessing cell adhesion, while straightforward, suffer from poor reproducibility and low throughput. This study introduces a novel deep learning-based approach utilizing the You Only Look Once (YOLO) convolutional neural networks to automate cell detection and counting, thereby enhancing the precision and efficiency of cell adhesion assays. Employing the YOLOv3, YOLOv5, YOLOv8 and YOLOv9 architectures, we address the challenges of variability in cell density and illumination within adhesion experiments. Our methodology involved the analysis of cell cultures using AQP2-RCCD1 cells and subsequent adhesion assays, with the data captured and processed using the latest YOLO models. These models were trained on various image resolutions to assess the trade-offs between image quality and computational efficiency, significantly optimizing the detection process. In commitment to open science principles, the source code and trained models will be shared to foster innovation and collaboration in the field. The study not only demonstrates the integration of cutting-edge AI in cell biology but also sets a precedent for the wider application of machine learning in biomedical research.