Build classification models for biomarker discovery and diagnostics using RandomForest, XGBoost, and logistic regression with sklearn-compatible APIs.
pip install scikit-learn xgboost pandas matplotlibTell your AI agent what you want to do:
- "Build a random forest classifier on my expression data"
- "Train an XGBoost model to predict disease status from my omics data"
- "Use logistic regression with L1 penalty to find sparse biomarkers"
- "Evaluate my classifier with ROC curves and AUC"
"Train a random forest classifier on my gene expression matrix (expression.csv) with labels in metadata.csv. Show me the ROC curve and top 20 important features."
"Build an XGBoost classifier to predict tumor vs normal from my RNA-seq counts."
"Use L1-regularized logistic regression to find a sparse set of genes that classify my samples. Which genes have non-zero coefficients?"
"I have expression data with 4 cancer subtypes. Train a classifier and show the confusion matrix."
"Compare RandomForest, XGBoost, and logistic regression on my dataset. Which has the best AUC?"
- Load expression matrix and sample labels
- Split data into train/test sets (stratified)
- Build preprocessing pipeline with scaling
- Train classifier and generate predictions
- Report metrics (accuracy, AUC, classification report)
- Visualize ROC curve and feature importances
- Always use stratified splits for imbalanced classes
- StandardScaler is essential for logistic regression; optional for tree models
- Use class_weight='balanced' when classes are imbalanced
- XGBoost parameters: use random_state not seed, n_jobs not nthread
- L1 penalty gives sparse solutions; L2 handles correlated features better
- For small datasets, consider nested cross-validation instead of train/test split