AI-QuIC machine learning for automated detection of misfolded proteins in seed amplification assays

Seed Amplification Assays (SAAs) detect misfolded proteins associated with neurodegenerative diseases, such as Alzheimer’s disease, Parkinson’s disease, ALS, and prion diseases. However, current data analysis methods rely on manual, time-consuming, and potentially inconsistent processes. We introduce AI-QuIC, an artificial intelligence platform that automates analyzing data from Real-Time Quaking-Induced Conversion (RT-QuIC) assays. Using a well-labeled RT-QuIC dataset comprising over 8000 wells, the largest curated dataset of its kind for chronic wasting disease prion seeding activity detection, we applied various AI models to distinguish true positive, false positive, and negative reactions. Notably, the deep learning-based1 Multilayer Perceptrons (MLP) model achieved a classification sensitivity of over 98% and specificity of over 97%. By learning directly from raw fluorescence data, the MLP approach simplifies the data analytic workflow for SAAs. By automating and standardizing the interpretation of SAA data, AI-QuIC holds the potential to offer robust, scalable, and consistent diagnostic solutions for neurodegenerative diseases.