Title: Noncoding RNAs and deep learning neural network discriminate multi-cancer types

Abstract: Detecting cancers at early stages can dramatically reduce mortality rates. Therefore, practical cancer screening at the population level is needed. Here, we develop a comprehensive detection system to classify all common cancer types. By integrating artificial intelligence deep learning neural network and noncoding RNA biomarkers selected from massive data, our system can accurately detect cancer vs healthy object with 96.3% of AUC of ROC (Area Under Curve of a Receiver Operating Characteristic curve). Intriguinely, with no more than 6 biomarkers, our approach can easily discriminate any individual cancer type vs normal with 99% to 100% AUC. Furthermore, a comprehensive marker panel can simultaneously multi-classify all common cancers with a stable 78% of accuracy at heterological cancerous tissues and conditions. This provides a valuable framework for large scale cancer screening. The AI models and plots of results were available in this https URL