Machine Learning in Autoimmune Diseases

Overview

This project explores machine learning applications in autoimmune disease classification and genetic analysis, focusing on Pediatric Inflammatory Bowel Disease (PIBD),Rheumatoid arthritis and Narcolepsy. By integrating supervised learning, genetic analysis, and feature selection techniques, the study enhances disease prediction and biomarker identification. Work done under the guidance and support of Dr. Meenal Kowshik, Professor Department of Biological Sciences at BITS Pilani, K K Birla Goa Campus

Classification of Pediatric Inflammatory Bowel Disease (PIBD)

Objective

• Classify PIBD subtypes: Crohn’s Disease (CD), Ulcerative Colitis (UC), and Inflammatory Bowel Disease Unclassified (IBDU).
• Improve diagnostic precision using histological and endoscopic data.

Methodology

• Dataset: 287 patients with gastrointestinal abnormalities at 10 specific locations.
• Machine Learning Approach:
  • Implemented Random Forest for supervised learning.
  • Numerical encoding of gastrointestinal features to enhance model interpretability.
  • Applied Principal Component Analysis (PCA) for dimensionality reduction.

Results & Optimizations

• Achieved 65% accuracy in PIBD subtype classification.
• Applied feature engineering and hyperparameter tuning for optimization.
• Investigating Support Vector Machines (SVMs) and ensemble learning to improve classification performance.

Genetic Correlation Approach for Disease Diagnosis

Objective

• Identify key genes associated with autoimmune diseases.
• Develop a robust diagnosis model using genetic data.

Methodology

• Data Collection: Genetic profiles of healthy and affected individuals.
• Statistical Gene Analysis:
  • Computed corrected p-values to identify significant genetic markers.
  • Key genes identified: STAT1, IFI44, HLA-DRB1, and others.

Results & Insights

• Random Forest model achieved 73% prediction accuracy for disease classification.
• Identified critical genetic pathways influencing autoimmune conditions.
• Explored environmental triggers that contribute to disease progression.

Machine Learning-Based Identification of Narcolepsy Predictive Genes

Objective

• Develop an ML framework to predict narcolepsy susceptibility genes.
• Enhance diagnostic strategies using genomic data.

Methodology

• Data Sources:
  • Gene Expression Omnibus (GEO) database:
    • Training Dataset: GSE69371
    • Validation Dataset: GSE69370
• Data Preprocessing:
  • Filtered and normalized raw gene expression data.
• Gene Analysis:
  • Identified 21,154 differentially expressed genes (DEGs).
  • Extracted 309 autoimmune-related genes (ARGs) from the Gene and Autoimmune Disease Association Database (GAAD).
  • Merged DEGs and ARGs to refine 309 differentially expressed autoimmune-related genes (DEARGs).

Gene Prioritization

• Ranked genes based on statistical significance and tissue-specific expression.
• Top identified genes:
  • HLA-DQB1, MOG, CTSH, CHKB, TAC1, PENK, HLA-DQA1, DNMT1, MAP3K7, NRXN1.

Machine Learning Integration

• Utilized SVM-RFE (Support Vector Machine-Recursive Feature Elimination), Lasso Logistic Regression, and Random Forest for gene selection and classification.

Results

• Identified a refined set of genes with high predictive power for narcolepsy.
• Demonstrated associations between gene expression and disease pathology.

Tools and Techniques

• Machine Learning Algorithms: Random Forest, SVM, Lasso Regression.
• Statistical Analysis: p-value computation, feature importance analysis.
• Data Processing: Principal Component Analysis (PCA) for dimensionality reduction.
• Performance Metrics: Accuracy, Precision, Recall, F1-score, Confusion Matrix.
• Bioinformatics Databases:
  • Gene Expression Omnibus (GEO)
  • Gene and Autoimmune Disease Association Database (GAAD)

Repository & Additional Resources

GitHub Repository: PIBD
GitHub Repository: Narcolepsy
Project Report