AI-QuIC: Machine Learning for Automated Detection of Misfolded Proteins in Seed Amplification Assays

Abstract

Advancements in AI, particularly deep learning, have revolutionized protein folding modeling, offering insights into biological processes and accelerating drug discovery for protein misfolding diseases. However, detecting misfolded proteins associated with neurodegenerative disorders, such as Alzheimers, Parkinsons, ALS, and prion diseases, relies on Seed Amplification Assays (SAAs) analyzed through manual, time-consuming, and potentially inconsistent methods. We introduce AI-QuIC, an AI-driven platform that automates the analysis of Real-Time Quaking- Induced Conversion (RT-QuIC) assay data, a type of SAA crucial for detecting misfolded proteins. Utilizing a well-labeled RT-QuIC dataset of over 8,000 wells--the largest curated dataset for chronic wasting disease prion detection--we applied various AI models to classify true positive, false positive, and negative reactions. Notably, our deep-learning-based model achieved over 98% sensitivity and 97% specificity. By learning directly from raw fluorescence data, deep learning simplifies the SAA-analysis workflow. Automating and standardizing SAA data interpretation with AI-QuIC provides robust, scalable, and consistent diagnostic solutions.