Study tests ML to sort dolphin infections by bloodwork trends
Bottom line
A new American Journal of Veterinary Research study suggests machine learning may help veterinarians distinguish whether a bottlenose dolphin's infection is bacterial, fungal, or viral by tracking how blood analytes change over time, rather than relying on single time-point results alone. In this retrospective longitudinal study, researchers Ashley Barratclough and Abby M. McClain analyzed blood samples collected from professional-care dolphins between 1995 and 2025 and used random forest models to evaluate up to 63 hematology and biochemistry analytes across 33 confirmed disease episodes: 11 bacterial, 10 fungal, and 12 viral. The work builds on earlier long-term dolphin biomarker research showing that serial blood data can reveal clinically meaningful disease patterns that may be missed in isolated snapshots. (pmc.ncbi.nlm.nih.gov)
Why it matters: For veterinary teams working with dolphins and other managed marine mammals, differentiating infectious causes can be especially difficult because bacterial and fungal organisms may be isolated from both healthy and diseased animals, and definitive diagnosis often depends on culture, imaging, or repeated assessment over time. A model that uses temporal blood analyte dynamics could support earlier clinical decision-making, help refine antimicrobial or antifungal choices, and improve monitoring when case numbers are small and advanced diagnostics are limited. Prior dolphin infectious disease research has also shown distinct immune-response patterns across bacterial, fungal, and viral illness, supporting the idea that longitudinal lab trends may carry diagnostic value. (int-res.com)
What to watch: The next step will be whether these models are externally validated in additional dolphin populations and eventually adapted into practical decision-support tools for clinicians. (pmc.ncbi.nlm.nih.gov)
Key facts
- Study type
- Retrospective longitudinal observational study
- Species
- Bottlenose dolphins (Tursiops truncatus)
- Goal
- Distinguish bacterial, fungal, and viral infections from longitudinal blood analytes
- Model
- Random forest
- Samples reviewed
- Blood samples collected from 1995 through 2025
- Confirmed disease episodes
- 33 total
- Episode breakdown
- 11 bacterial, 10 fungal, and 12 viral
- Analytes evaluated
- Up to 63 hematology and biochemistry analytes
- Main limitation
- Retrospective, small sample size, and professional-care dolphins only
Machine learning is making a new push into marine mammal diagnostics, with a newly reported AJVR study finding that longitudinal bloodwork may help distinguish bacterial, fungal, and viral infections in bottlenose dolphins. The study, by Ashley Barratclough and Abby M. McClain, used random forest models trained on serial hematology and biochemistry data from confirmed disease episodes in professional-care dolphins, aiming to answer a familiar clinical question with a more data-driven tool: what kind of infection is this? (pmc.ncbi.nlm.nih.gov)
That question has been a persistent challenge in dolphin medicine. Earlier work in bottlenose dolphins has shown that potentially pathogenic bacteria and fungi can be recovered in settings where clinical significance isn't always clear, making interpretation of culture results difficult without broader clinical context. At the same time, dolphin infectious disease research has documented meaningful differences in host immune responses across bacterial, fungal, and viral infections, suggesting that the animal's blood profile over time may contain a usable diagnostic signal. (int-res.com)
According to the study abstract, the investigators reviewed blood samples collected from 1995 through 2025 and included 33 confirmed infectious disease episodes: 11 bacterial, 10 fungal, and 12 viral. They evaluated up to 63 blood analytes and applied random forest modeling in a retrospective, longitudinal observational design. That temporal framing matters. Previous work from a 25-year dolphin cohort has already shown that serial biomarker analysis can identify physiologic patterns that would be hard to capture from one-off blood draws alone. (pmc.ncbi.nlm.nih.gov)
The study also fits into a broader movement toward computational diagnostics in both veterinary and human medicine. Machine learning models based on routine blood data are already being explored in people to separate bacterial from viral infections, and dolphin researchers have recently tested machine learning in other health-assessment contexts as well. What stands out here is the attempt to apply that approach to a species where case numbers are limited, longitudinal records are unusually valuable, and treatment decisions often have to be made before definitive diagnostics are complete. (arxiv.org)
Independent expert reaction specific to this paper was limited in publicly available sources at the time of reporting. Still, the surrounding literature helps explain why the work is likely to draw attention. Reviews of cetacean fungal disease, including aspergillosis, describe frequent diagnostic complexity and overlap with other infectious processes, while clinical reports in dolphins have noted that inflammatory blood markers can persist or behave differently depending on concurrent disease and treatment response. Taken together, that background supports the study's premise that dynamic, multianalyte interpretation may outperform static thresholds in some cases. (mdpi.com)
Why it matters: For veterinary professionals, the practical value is less about replacing culture, imaging, or pathogen-specific testing, and more about improving triage and pattern recognition. If validated, a temporal blood-analyte model could help clinicians decide when a case looks more consistent with bacterial disease that may justify antimicrobial escalation, when fungal disease should stay high on the differential, or when a viral pattern may argue for a different monitoring and treatment strategy. In managed marine mammal settings, where individuals may have years of baseline data and repeated sampling is feasible, this kind of tool could be especially useful. (frontiersin.org)
There are also clear limitations. The study is retrospective, the number of confirmed episodes is small, and the population comes from professional-care dolphins rather than free-ranging animals. That means performance may not generalize across institutions, age groups, husbandry conditions, or mixed infections. As with most machine learning studies in niche species, the key question isn't whether the model worked in-development, but whether it holds up prospectively and across new datasets. (pmc.ncbi.nlm.nih.gov)
What to watch: Watch for the full paper's publication details, model performance metrics, and any follow-up validation studies in additional marine mammal populations. The most meaningful next milestone will be evidence that the approach improves real-world diagnostic confidence or treatment decisions, not just retrospective classification accuracy. (pmc.ncbi.nlm.nih.gov)