Highly accurate classification of biological spores by culture medium for forensic attribution using multiple chemical signature types and machine learning

Future proliferation of biological expertise and new technology may increasingly lower the difficulty to produce biological organisms for misuse. Rapid attribution of a biological attack is needed to quickly identify the person or lab responsible and prevent additional attacks by enabling the apprehension of suspects. Here, triplicate batches of Bacillus anthracis Sterne strain ( Ba St) spores were grown in a total of seven amateur and professional media. Multiple orthogonal analytical signatures (peptides, metabolites, lipids by fatty acid methyl ester (FAME) analysis, bulk organic profile, and trace elements) were collected from the Ba St spores. The proteomics and metabolomics analyses identified promising attribution signature compounds that are unique to each of the seven production methods. In addition, while each of the signature types showed varying degrees of value individually for attributing Ba St spores to the culture medium used to prepare them, fusing results from all five signatures types to increase sourcing robustness and using a random forest sourcing algorithm yielded 100% hold-one-batch-out cross-validation classification accuracy and an average relative source probability for the correct source 5.5× higher than the most probable incorrect source. These preliminary results provide a proof-of-concept for the development of forensic examinations that can attribute biological agents to production methods for use in future investigations.