DABSP: Dual-directional Attention Based Multimodal Data Fusion Framework for Pan-Cancer Survival Outcome Prediction

This repository contains the implementation of DABSP, a framework for integrating histologic and transcriptomic data for pan-cancer survival outcome prediction through a dual-directional attention based multimodal data fusion framework.

Citation

If you use this code in your research, please cite:

DABSP: Integrating histologic and transcriptomic data for pan-cancer survival outcome prediction through a dual-directional attention based multimodal data fusion framework

Overview

DABSP is a multimodal deep learning framework that combines:

• Histologic data: Whole Slide Images (WSI) from histopathology
• Transcriptomic data: RNA-seq gene expression data organized by biological pathways

The framework uses a dual-directional attention mechanism with LoRA (Low-Rank Adaptation) to effectively fuse these modalities for improved survival prediction.

Features

• Multimodal fusion of WSI and omics data
• Dual-directional attention mechanism with LoRA
• Support for multiple cancer types (pan-cancer)
• 5-fold cross-validation for robust evaluation
• Multiple pathway types: xena, hallmarks, combine
• Chebyshev KAN (Kolmogorov-Arnold Network) for pathway processing

Requirements

• Python 3.x
• PyTorch
• CUDA (for GPU acceleration)

Quick Start

Example Usage

Run the example script for the COAD dataset:

bash scripts/run_coad.sh

Supported Cancer Types

The framework supports multiple TCGA cancer types. Update the STUDIES variable in the script to run on different datasets:

• COAD (Colon Adenocarcinoma)
• BRCA (Breast Invasive Carcinoma)
• BLCA (Bladder Urothelial Carcinoma)
• HNSC (Head and Neck Squamous Cell Carcinoma)
• STAD (Stomach Adenocarcinoma)
• And more...