A generalizable 29-mRNA neural-network classifier for acute bacterial and viral infections

Area of Science:

• Infectious disease diagnostics
• Computational biology
• Host-response genomics

Background:

• Accurate differentiation between bacterial and viral infections is critical for effective patient management, particularly in sepsis cases.
• Current diagnostic methods can be slow or lack specificity, leading to delayed treatment and increased morbidity.
• Reducing antibiotic overuse is a major public health goal, requiring better tools to identify bacterial infections.

Purpose of the Study:

• To develop a generalizable host-gene-expression-based classifier for acute bacterial and viral infections.
• To validate the classifier's performance on independent patient cohorts.
• To assess the potential of the classifier as a diagnostic tool for suspected infections and sepsis at hospital admission.

Main Methods:

• A neural network classifier was trained using host mRNA expression data from 1069 patients across 18 transcriptomic studies.
• The classifier was trained using a preselected panel of 29 host mRNAs.
• The trained classifier, inflammatix-bacterial-viral-noninfected-version 1 (IMX-BVN-1), was applied to an independent cohort of 163 patients without retraining.

Main Results:

• The classifier achieved high performance in training data, with bacterial-vs-other and viral-vs-other area under the receiver-operating characteristic curves (AUROCs) of 0.92.
• In an independent cohort, IMX-BVN-1 demonstrated AUROCs of 0.86 for bacterial vs. other and 0.85 for viral vs. other infections.
• In patients admitted within 36 hours, the classifier showed improved performance with AUROCs of 0.92 (bacterial) and 0.91 (viral).

Conclusions:

• A host-gene-expression classifier (IMX-BVN-1) based on 29 mRNAs can generalize to distinguish bacterial and viral infections.
• The classifier shows promising performance in independent cohorts, including early in hospital admission.
• Further validation could establish IMX-BVN-1 as a valuable tool for early sepsis and infection assessment, potentially reducing morbidity and antibiotic misuse.