DGMP

DGMP identifies cancer driver genes by integrating a directed graph convolutional network (DGCN) with a multilayer perceptron (MLP) to learn multi-omics features and the directed topology of gene regulatory networks (GRNs).

Key Features:

• Integration of Multi-Omics Pan-Cancer Data: Uses gene expression, mutation profiles, copy number variations, and DNA methylation to represent diverse molecular alterations across cancers.
• Directed Graph Convolutional Network (DGCN): Learns gene multi-omics features together with GRN topological structure while preserving the directed nature of regulatory interactions.
• Multilayer Perceptron (MLP) Integration: Augments DGCN to reduce bias toward graph topology and emphasize intrinsic gene feature signals.
• Performance and Validation: Demonstrates superior performance relative to competing methods across three different GRNs and includes ablation studies examining the contribution of the MLP component.
• Comprehensive Gene Identification: Identifies driver genes beyond highly mutated genes, including those with differential expression, aberrant DNA methylation, and genes interacting with other cancer-related genes in GRNs.

Scientific Applications:

• Precision Oncology Research: Enables molecular-level analysis of cancer initiation and progression to inform targeted therapy development and personalized treatment strategies.
• Cancer Driver Gene Discovery: Facilitates identification of both known and novel driver genes to support cancer diagnosis, prognosis, and therapeutic intervention studies.

Methodology:

Integrates a directed graph convolutional network (DGCN) with a multilayer perceptron (MLP) using pan-cancer multi-omics data (gene expression, mutation profiles, copy number variations, DNA methylation), preserves directed GRN topology during feature learning, and evaluates performance across three GRNs with ablation studies assessing the MLP contribution.