Deep learning with evolutionary and genomic profiles for identifying cancer subtypes

Area of Science:

• Genomics
• Computational Biology
• Cancer Research

Background:

• Accurate cancer subtype identification is essential for precision oncology.
• Evolutionary conservation of genes may hold functional significance in cancer.
• The role of evolutionary conservation in distinguishing cancer subtypes is not well understood.

Purpose of the Study:

• To investigate the utility of evolutionarily conserved genes for cancer subtype identification.
• To develop novel deep learning-based strategies for cancer subtyping using conserved genes.
• To compare the performance of these novel strategies against existing methods.

Main Methods:

• Identification of evolutionarily conserved genes (core genes) involved in cancer-related pathways.
• Development of a feature-based strategy (FES) and an image-based strategy (IMS) integrating gene profiles with deep learning.
• Validation of strategies using breast cancer subtypes and multiple cancer types, comparing against random sets and PAM50 classifier.

Main Results:

• Core genes are predominantly involved in cell growth and metabolism pathways.
• The core gene set-based FES demonstrated higher accuracy for breast cancer subtype identification compared to other FES methods.
• Both IMS and FES using the core gene set outperformed other strategies in classifying breast cancer and multiple cancer types.
• IMS showed reproducibility across different gene expression data types (RNA-seq, microarray).

Conclusions:

• Evolutionary conservation-based models offer a valid and effective approach for cancer subtype identification.
• The identified core gene set provides valuable clues for distinguishing cancer subtypes.
• These findings support the integration of evolutionary information into precision diagnostic tools for cancer.