Multi-Method ML Diagnostic and Prognostic Classifier Dashboard

An interactive React dashboard for visualizing results from multi-method machine learning classification pipelines. Designed for biomedical research, clinical diagnostics, and prognostic prediction workflows.

🎯 Features

Machine Learning Analysis

• Multiple ML Methods: Random Forest, SVM, XGBoost, KNN, MLP with soft/hard voting ensembles
• Feature Selection: Forward, backward, and stepwise selection methods
• Permutation Testing: Statistical validation against random chance
• Cross-Validation: Repeated k-fold CV with comprehensive metrics
• Class-Specific Risk Scores: Per-sample risk scores for each class (0-100 scale)

Visualization Dashboard

• Model Performance Comparison: Side-by-side comparison of all ML methods
• ROC Curves: Interactive multi-model ROC curves with threshold analysis
• Confusion Matrices: Visual matrices with derived metrics (Accuracy, Precision, Sensitivity, Specificity, F1, NPV)
• Feature Importance: Ranked feature visualization with stability analysis
• Expression Box Plots: True box plots showing median, IQR, whiskers, and mean for top features across classes
• Dimensionality Reduction: PCA, t-SNE, and UMAP clustering visualizations
• Permutation Distributions: Actual vs. permuted performance comparison
• Calibration Curves: Model probability calibration assessment
• Profile Rankings: Sample-level prediction confidence rankings with risk scores
• Feature Selection Visualization: Per-method feature selection comparison

Export & Reporting

• HTML/PDF Reports: Comprehensive export of analysis results
• Model Export: R code snippets for external predictions
• Batch Processing: Compare multiple datasets

🚀 Quick Start

Dashboard Setup

# Clone the repository
git clone <YOUR_GIT_URL>

# Navigate to the project directory
cd <YOUR_PROJECT_NAME>

# Install dependencies
npm install

# Start the development server
npm run dev

R Script Setup

Two R scripts are available:

1. Cross-Validation Mode (scripts/multi_ML_classifier.R): Full pipeline with k-fold CV
2. Full Training Mode (scripts/multi_ML_classifier_full_training.R): Train on 100% of data

Running the Analysis

# Create conda environment
conda env create -f public/ml_classifier_env.yml
conda activate ml_classifier

# Run CV-based analysis
Rscript scripts/multi_ML_classifier.R \
  --expr expression_matrix.txt \
  --annot sample_annotation.txt \
  --target diagnosis \
  --outdir results

# Or run full training (no CV)
Rscript scripts/multi_ML_classifier_full_training.R \
  --expr expression_matrix.txt \
  --annot sample_annotation.txt \
  --target diagnosis \
  --outdir results

Load the generated ml_results.json into the dashboard.

📊 Input Data Format

Expression Matrix (tab-delimited .txt or .tsv)

• Rows: Features (genes, proteins, metabolites, etc.)
• Columns: Samples (column names are sample IDs)

Gene    Sample1  Sample2  Sample3
BRCA1   5.2      4.8      6.1
TP53    3.1      3.5      2.9
...

Sample Annotation (tab-delimited .txt or .tsv)

• Must contain sample IDs matching expression matrix columns
• Must contain target variable column for classification

SampleID  diagnosis  age  stage
Sample1   1          45   II
Sample2   0          52   I
Sample3   1          38   III
...

🔬 ML Method Guidance

Method	Best For	Dataset Size
Random Forest	Robust baseline, feature importance	Any
SVM	High-dimensional data, clear margins	Small-Medium
XGBoost	Large datasets, complex patterns	Large (>100)
KNN	Simple patterns, interpretable	Small-Medium
MLP	Complex non-linear relationships	Large (>100)

Note: XGBoost and MLP require larger datasets for optimal performance. Random Forest, SVM, and KNN are more robust on small biological datasets (<50 samples).

📈 Risk Scoring

The pipeline calculates class-specific risk scores for each sample:

• Risk Score Class 0 (Negative): Probability × 100 for the negative class
• Risk Score Class 1 (Positive): Probability × 100 for the positive class

These scores range from 0-100 and indicate the model's confidence that a sample belongs to each class. Higher scores indicate stronger classification confidence.

🛠 Technology Stack

• Frontend: React 18, TypeScript, Vite
• UI Components: shadcn/ui, Tailwind CSS
• Charts: Recharts
• R Backend: caret, randomForest, e1071, xgboost, nnet, pROC

📁 Project Structure

├── scripts/
│   ├── multi_ML_classifier.R              # CV-based analysis script
│   └── multi_ML_classifier_full_training.R # Full training script
├── public/
│   ├── ml_classifier_env.yml              # Conda environment
│   └── demo_data/                         # Sample input files
├── src/
│   ├── components/                        # React components
│   ├── types/                             # TypeScript interfaces
│   └── data/                              # Demo data
└── README.md

📖 References

• IntelliGenes
• Li et al. 2022 - Permutation Strategy

📝 License

MIT License

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.