# Integrating Bayesian Inference and Machine Learning to Evaluate TAP and Trypsin-2 as Early Biomarkers of Systemic Inflammation in Acute Pancreatitis

## TL;DR

This study combines Bayesian and machine learning methods to assess TAP and trypsin-2 as early indicators of inflammation in acute pancreatitis.

## Contribution

The novel integration of Bayesian inference and machine learning to evaluate TAP and trypsin-2 as early biomarkers in acute pancreatitis.

## Key findings

- Urinary trypsin-2 is a better predictor of systemic inflammation than serum markers.
- Three distinct patient subgroups were identified based on biomarker levels and inflammation severity.
- Random Forest models showed the best performance in predicting inflammation using urinary trypsin-2 and age.

## Abstract

Background and Objectives: Acute pancreatitis (AP) has a wide range of clinical severity, and early prediction of disease progression is still challenging. Trypsinogen-activating peptide (TAP) and trypsin-2 serve as direct biomarkers for intrapancreatic proteolytic activation and may provide earlier pathophysiological information compared with traditional markers. Materials and Methods: In this retrospective cohort analysis involving 54 AP patients, we evaluated 24 h serum and urinary TAP and trypsin-2 concentrations by Bayesian correlation, mediation analysis, unsupervised K-means clustering, and supervised machine learning (Elastic Net and Random Forest). The analyses investigated the relationships of biomarkers with inflammation (CRP), enzymatic activities (amylase, lipase), and clinical factors, as well as inflammation severity (CRP levels). Results: Bayesian correlations indicated moderate evidence for a relationship between serum TAP and CRP (BF10 = 8.42), as well as strong evidence linking age to serum TAP (BF10 = 12.75). Serum trypsin-2 showed no correlation with CRP, while urinary trypsin-2 had a correlation with amylase (BF10 = 6.89). Mediation analysis indicated that TAP and trypsin-2 accounted for 42–44% of the impact of CRP on pancreatic enzyme elevation. Clustering revealed three phenotypic subgroups (“Mild Activation”, “Moderate System”, and “Severe Pancreatic-Renal”), the latter showing the highest levels of CRP and biomarkers. Machine learning models highlighted urinary trypsin-2 and age as the most significant predictors of inflammation, with Random Forest achieving the highest performance (R2 = 0.53). Conclusions: Early urinary trypsin-2 outperforms serum markers as a predictor of systemic inflammatory intensity, indicating total proteolytic impairment and renal clearance. This integrative analysis reveals unique biological phenotypes and highlights the potential of these biomarkers for early assessment of the inflammatory burden. Their role in predicting clinical disease progression requires prospective validation. Integrative biomarker analysis reveals unique biological phenotypes and improves assessment of inflammatory burden in PA. Larger cohorts are required for prospective validation to incorporate these biomarkers into precision-based diagnostic frameworks.