Multiview representation learning for identification of novel cancer genes and their causative biological mechanisms

Jianye Yang1, Haitao Fu1,2, Feiyang Xue1

  • 1College of Informatics, Huazhong Agricultural University, Wuhan 430070, China.

PubMed

Insights

This study introduces IMVRL-GCN, an interpretable framework for cancer gene identification using multiview data. It successfully identifies novel cancer genes and offers insights for personalized cancer treatments.

Area of Science:

  • Genomics
  • Computational Biology
  • Precision Oncology

Background:

  • Tumorigenesis results from cancer gene dysfunction, driving uncontrolled cell proliferation.
  • A comprehensive cancer gene catalog is crucial for advancing precision oncology.
  • Existing graph neural network (GNN) methods struggle with integrating multiview data and interpretability in cancer gene identification.

Purpose of the Study:

  • To develop an interpretable representation learning framework, IMVRL-GCN, for enhanced cancer gene identification.
  • To effectively integrate and analyze shared and specific representations from multiview data.
  • To provide insights into the mechanisms of cancer gene discrimination and potential therapeutic strategies.

Main Methods:

  • Proposed an interpretable representation learning framework named IMVRL-GCN.
  • Utilized multiview data to capture shared and specific gene representations.
  • Compared IMVRL-GCN performance against state-of-the-art methods and baselines.

Main Results:

  • IMVRL-GCN demonstrated superior performance compared to existing cancer gene identification methods.
  • Identified 74 high-confidence novel cancer genes.
  • Multiview data analysis revealed the importance of shared, mutation-specific, and structure-specific representations in cancer gene discrimination.

Conclusions:

  • Shared representations correlate with gene function, while mutation-specific and structure-specific representations relate to mutagenic propensity and functional synergy.
  • Identified potential therapeutic strategies, including using afatinib for mutation-driven risks and targeting SRC interactions for specific genes.
  • The findings offer valuable insights for developing individualized cancer treatments.