Differentiation of Bacterial Isolates Using Mass Spectrometry and Machine Learning Analysis of Volatile Organic Compounds.

Identification of bacteria is a critical task in addressing bacterial contamination of the indoor environment. Standard culture-independent (e.g., PCR and DNA sequencing) or culture-dependent techniques capable of species-level identification are often costly, time-consuming, and require specialized equipment and personnel. This work presents a proof-of-concept study demonstrating the feasibility of detecting and identifying bacteria through analysis of the volatile organic compounds (VOCs) emitted from their cultures using ambient pressure mass spectrometry and machine learning. An E. coli and three Pseudomonas spp. isolates were grown in pure cultures, and VOCs were collected from culture headspaces using sorbent tubes. The VOCs thermally desorbed from the sorbent tubes were detected using a mass spectrometer typically used for residual gas analysis. The resulting data was used to train a novel machine learning (ML) model able to identify the four organisms through the VOCs emitted by independently replicated cultures. The ML model correctly correlated spectra of the collected VOCs with their species with 84% accuracy, contrasting with ∼56% accuracy based on the commonly used Bray-Curtis similarity of the resulting mass spectra or principal component analysis (PCA). Additional cultures were sampled using solid-phase microextraction (SPME), and captured VOCs were analyzed with gas chromatography-mass spectrometry (GC-MS) for compound identification. The GC-MS data revealed a unique VOC profile for each species, confirming that these organisms can be differentiated solely by VOC analysis. Specific VOCs distinguishing the isolates and contributing to the success of the ML model, such as butanoic acid and 1-propanol detected from only the E. coli cultures and 2-ethyl-1-hexanol detected from only the P. aeruginosa cultures, were also identified. By combining ambient mass spectrometry with machine learning, the approach discussed in this work achieves bacterial differentiation comparable to that of a standard benchtop approach. This work provides initial data showing that bacterial identification through VOC analysis may be possible, with potential applications in both laboratory and field settings.