Clinical Biomarker Validation Pipeline (Sepsis/AKI)

Statistical Inference & Feature Selection for Low-N Pre-Clinical Trials

📌 Project Overview

This repository contains a robust statistical pipeline designed to validate novel biomarkers for the early detection of Sepsis-induced Acute Kidney Injury (AKI). Unlike standard "Big Data" problems, pre-clinical medical device trials are constrained by small sample sizes ($n < 50$) and high biological variance. This project implements rigorous non-parametric testing and dimensionality reduction to identify diagnostic signals that outperform the current clinical gold standard (Serum Creatinine).

🚀 Key Technical Achievements

• Dimensionality Reduction: Applied Principal Component Analysis (PCA) to segregate septic vs. non-septic clusters in high-dimensional biomarker space.
• Diagnostic Benchmarking: Conducted ROC (Receiver Operating Characteristic) analysis to compare novel biomarkers against traditional indicators (Lactate, Creatinine), establishing optimal decision thresholds for clinical intervention.
• Drift Correction: Implemented normalization routines to account for inter-subject biological variability in rodent models.
• Modular Architecture: Refactored monolithic notebook (~800 lines) into professional Python package with reusable utilities

🛠️ Tech Stack & Methodology

• Language: Python 3.9+
• Libraries:
  • pandas (Data Manipulation)
  • scikit-learn (PCA, Metrics)
  • seaborn (Statistical Visualization)
  • scipy.stats (Hypothesis Testing)
  • matplotlib (Plotting)
• Statistical Approach:
  • Mann-Whitney U Test: Utilized for hypothesis testing due to the non-normal distribution of physiological data.
  • AUC-ROC: Used as the primary metric to evaluate sensitivity/specificity trade-offs (achieved AUC > 0.85 for target biomarkers).

📊 Results Snapshot

Comparison of Diagnostic Power (Sample Data)

Note: While Serum Creatinine (traditional) shows a delayed response, the novel biomarker panel demonstrates earlier sensitivity to sepsis onset.