Machine learning analysis facilitates the identification of liver biomarkers in postmortem diagnosis of heat stroke.

BACKGROUND: Clarifying the cause of death of heat stroke (HS) is critical for judicial handling of related cases. Yet, postmortem diagnosis of HS remains challenging owing to the lack of specific morphology and biomarkers. Liver damage frequently acts as a direct cause in HS. This study aimed to explore the postmortem diagnosis of HS based on the protein biomarkers in liver by machine learning methods.

METHODS: Electron microscopy was used to observe the morphological features of hepatocytes in HS. Five candidate protein biomarkers were selected based on the literature, the eletron microscopy observation, and the previously proteomics results. Western blotting and Immunohistochemistry were used to detect candidate proteins expression in both HS rat models and human cases. Lastly, a machine learning algorithm (eXtreme Gradient Boosting, XGBoost) was employed to detect the most discriminative biomarkers.

RESULTS: The mitochondrial damage, autophagy could be observed in hepatocytes of HS. Five candidate protein biomarkers (LRPPRC, CPT2, GPX2, LATS1, and ULK2) were decreased in both HS rat models and HS human cases. The postmortem temporal changes of biomarkers showed that LATS1 and CPT2 were stable. XGBoost revealed LATS1 and CPT2 were the most discriminative biomarkers. The combination of LATS1 and CPT2 achieved optimal diagnostic efficiency, with a sensitivity of 90.91 % and a specificity of 100 %. A diagnostic software was constructed and was applied to five human cases, all of which were recognized.

CONCLUSION: This study provides a promising valid strategy for practical diagnosis of death from HS by integrating liver LATS1 and CPT2 protein biomarkers.