Diagnostic and economic evaluation of matrix-assisted laser desorption/ionization time-of-flight mass spectrometry with machine learning for screening of Johne's disease from dairy cow serum

Johne's disease (JD) is an infectious bacterial disease (Mycobacterium avium ssp. paratuberculosis) that primarily affects the intestinal tract and associated lymph nodes of ruminants, especially in cattle. It is both a herd-level problem and an individual animal problem, as much of the infection is subclinical. This disease has a significant impact due to drastic production losses and early culling of animals. Here, we used MALDI-TOF MS for diagnostic screening of JD state from serum samples collected in the panhandle of Texas. We used several machine learning algorithms for observing patterns embedded in MALDI-TOF spectra using a training set of 37 serum samples. A separate set of scoring and evaluation samples (n = 25) from different animals was used to evaluate the accuracy of each machine learning model and assess the sensitivity (sens) and specificity (spc). Next, a partial budget analysis was used to calculate the net present value of using a MALDI-TOF–based test in a testing and culling strategy relative to existing tests and culling techniques. The random forest model yielded predictive results of optimal sens = 0.80, spc = 0.83, and Youden index (J) above 0.63. The partial budget analysis found that using a MALDI + PCR scenario was the most cost-effective strategy, reducing the net present value of JD losses by $13.11 per head compared with the no-control scenario. We conclude that MALDI-TOF MS of serum samples with machine learning provides a moderately effective diagnostic tool for JD diagnosis in cattle.