Electrochemical label-free pathogen identification for bloodstream infections diagnosis: Towards a machine learning based smart blood culture bottle

Bloodstream infections are a growing public health concern. Current pathogen identification systems are based on complex and expensive devices, intended for use in centralized laboratories. Subsequent identification re-quires harmful chemical reagents, specialized personnel and time. Here we describe a new approach for rapid and decentralized diagnosis of positive blood cultures using electrochemical sensors. By implementing a multi -material potentiometric platform in a blood culture bottle, we have developed a portable system for pathogen identification of Gram and of genus. Bacterial growth in human blood generates a specific label-free multiplex electrochemical fingerprint according to the detected species. Analysis of these fingerprints using homemade machine learning algorithms allow for rapid identification of the pathogen after detection (14 species and 9 genus) (GRAM= 99 % accuracy 5.75 h after detection, GENUS= 85 % accuracy 7.8 h after positivity) without further handling of the contaminated sample.