Machine Learning and Data Fusion platform for Phenotype-based Pathogen Identification

Conventional methods for detecting pathogens, which are based on culturing the microorganism, are time-consuming and laborious. Machine learning provides an alternative path to identify pathogens using supervised learning algorithms. Most current computational tools utilize genomic or protein data to identify bacteria. These methods look for features in the whole genome that correlate to pathogenicity. If whole genome or closely related data is available, models can accurately predict pathogenicity. However, obtaining genetic data is time-consuming process and fails to accurately predict or identify novel species. Genetically engineered bacteria are particularly difficult to identify due to lack of sequence data. In addition, identifying new bacterial pathogens that have emerged from parallel evolution is a challenging task. An alternative approach is using phenotypic data. Phenotypic data, such as niche response, self-preservation, and the ability to harm a host, model how an organism behaves and is not susceptible to same kind of limitations a genetic approach has. Triton Systems, Inc. proposes to develop a machine learning algorithm and data fusion technique to predict pathogenicity based on phenotypic data. If successful, the software will provide an effective threat identification and countermeasure tool.