Integrated Framework for Fusing Radiogenomics and Metabolic Modelling via Deep Learning in Ovarian Cancer

This repository contains the code and data necessary to reproduce the results presented in the paper:

N. Eftekhari, A. Saha, S. Verma, G. Zampieri, S. Sawan, A. Occhipinti, C. Angione, "Fusing imaging and metabolic modelling via multimodal deep learning in ovarian cancer."

---

Table of Contents

• Overview
• Requirements
• Installation
• Feature Reduction
• Image Segmentation
• Metabolic Modelling
• Survival Models

---

Overview

This repository includes a Jupyter Notebook, scripts, and data necessary to perform RNA feature extraction and image segmentation for ovarian cancer research. The notebook supports the integration of radiogenomics and metabolic data using deep learning, enabling personalized oncology research.

Requirements

The repository requires the following Python libraries:

• numpy
• pandas
• matplotlib
• scikit-learn
• nibabel
• SimpleITK
• tensorflow-gpu==2.1 (or tensorflow-mkl==2.1 for CPU)

Install these dependencies using the following command:

pip install -r requirements.txt

Installation

git clone https://github.com/Angione-Lab/Integrated-framework-for-fusing-radiogenomics-and-metabolic-modelling-via-deep-learning.git
cd Integrated-framework-for-fusing-radiogenomics-and-metabolic-modelling-via-deep-learning

Feature Reduction

1. Data Preparation:

Ensure that your RNA data file is in the required format as specified in the notebook. Place the data file in the appropriate directory, or update the file path in the notebook accordingly.

2. Running the Notebook: Launch the Jupyter Notebook server and for example open RNA_feature_vector.ipynb

jupyter notebook RNA_feature_vector.ipynb

Image Segmentation

Requirements

conda create -n ovarian_unet_py37 python=3.7 conda activate ovarian_unet_py37

conda install tensorflow-gpu==2.1 in case got error: conda install tensorflow-estimator=2.1.0

conda install tensorflow-mkl==2.1 (cpu version) pip install nibabel SimpleItk matplotlib tqdm

Training:

train/train.py.py

Prediction:

train/ovarian_unet_predict.py

Metabolic Modelling

This code uses a metabolic model and transcriptomic data to set reaction bounds in accordance with gene expression levels. The code loads a predefined metabolic model (human1.mat) and adjusts reaction bounds for specific reactions using gene expression profiles.

Key components:

1. Loading the Model: human1.mat model is loaded.
2. Setting Bounds: Bounds for selected reactions are modified based on user-defined gamma and threshold values.
3. Transcriptomic Data Integration: Gene expression data for each patient is mapped to the model.
4. Flux Analysis: Reaction fluxes are calculated under different threshold and gamma values.

Requirements

• MATLAB (with COBRA Toolbox installed)
• Metabolic model file human1.mat
• Gene expression data files: gene_exp.mat, gene_ids.mat, and patient_ids.mat

Ensure the COBRA Toolbox is installed and properly configured.

Survival Models

The notebook contains steps to load and preprocess data, merge multi-omics datasets, and apply several machine learning algorithms for survival analysis. Key components of the analysis include data normalization, survival model fitting, and model evaluation using concordance index.

Requirements

• Python 3.x
• NumPy
• Pandas
• Seaborn
• Matplotlib
• Scikit-learn
• scikit-survival
• Lifelines