fusion-ensemble-calling

RNA Fusion Detection Pipeline

This repository contains scripts and data for detecting gene fusions from RNA sequencing results. The workflow is modular and consists of the following key components:

1. Data Preparation

• Merge raw Excel tables (Table S4 and Table S9) containing fusion predictions and cancer cell line information.
• Integrate a truth set to label correct fusions.
• Restructure the dataset for machine learning.

2. Machine Learning Model

• Only selected tools are considered: ARRIBA, FUSIONCATCHER_v1.10_June192019, STAR_FUSION_v1.5.
• Samples from breast cancer are excluded.
• Features include detection flags, J and S counts, and cancer type.
• Models are trained and compared using PyCaret, with the best model saved in models/ and predictions saved in data/.

3. Repository Structure

• data/ – raw and processed data, including merged and restructured datasets.
• models/ – trained machine learning models.
• source/ – Python scripts for dataset preparation and model training.

4. Usage

• Run dataset preparation scripts to generate a final structured dataset.
• Train the classifier on the structured dataset, save predictions, and store the trained model.
• All paths are relative to the repository root for reproducibility.

5. Reproducibility and Modularity

• The pipeline allows easy updates to input data, features, or models without changing the overall structure.

6. References

• Original dataset and supplementary tables: Genome Biology
• PyCaret documentation: https://pycaret.