High-throughput screening system for immunogenic cell death inducers using artificial intelligence-based real-time image analysis.

Immunogenic cell death (ICD) can enhance the immunogenicity of cold tumors, convert them into immune-responsive hot tumors, and improve the efficacy of cancer immunotherapy. Because ICD inducers cause cell swelling, membrane rupture, and the release of damage-associated molecular patterns (DAMPs), effective screening of ICD inducers requires a system capable of rapidly and accurately assessing morphological changes and DAMP dynamics on a large scale, thus highlighting the need for advanced image-processing capabilities. In the present study, we developed an artificial intelligence (AI)-based detector to screen for ICD inducers by identifying the typical morphologies of dying cells undergoing ICD. To enhance the performance, we applied transfer learning from fluorescent markers and fine-tuned the model using differential interference contrast (DIC) images. In addition, model-assisted labeling (MAL) improved annotation efficiency by reducing the need for manual labeling in ICD screening. In a blind test, the AI successfully identified three ICD-inducing agents from eight candidates, which were validated through analyses of cell death type, DAMP release, and immune activation. Our AI-based high-throughput screening (HTS) system efficiently identified ICD candidates using only real-time optical images, thereby significantly reducing the time and resources required for screening. In addition, the system demonstrated the ability to detect subtle morphological differences that are difficult to discern through manual analysis, indicating its potential for ICD prediction as well as for foundational research and broader screening applications.