Machine Learning

Microbiome Disease Prediction

Predictive model to predict diseases based on patient microbiota data using machine learning methods.

Course: Introduction to Data ScienceCo-authors: Diego Quezada

Objectives

1Predict diseases from high-dimensional human microbiome data.
2Compare machine learning classifiers for small-sample, high-dimensional scenarios.
3Identify key microbial species associated with each disease.

SVM with linear kernel achieved 85% accuracy, outperforming Random Forest and XGBoost in this high-dimensional scenario.
PCA reduced 154,000+ features to 25 components while retaining 90% of variance, enabling effective model training.
Specific bacterial species (e.g., Bacteroides, Prevotella) emerged as key biomarkers for different diseases.
The model shows disease-specific patterns: increasing one disease probability decreases others due to single-diagnosis training data.