Leveraging TME features and multi-omics data with an advanced deep learning framework for improved Cancer survival...

Area of Science:

• Neuro-oncology
• Genomics
• Computational Biology

Background:

• Glioma is an aggressive brain tumor with poor survival rates due to its invasiveness and heterogeneity.
• Accurate prognostic prediction and identification of therapeutic targets are crucial for improving patient outcomes.

Purpose of the Study:

• To integrate multi-omics data for enhanced glioma prognosis and identification of novel therapeutic targets.
• To develop and validate a robust predictive model for glioma patient survival.

Main Methods:

• Single-cell RNA sequencing (scRNA-seq) was used to identify distinct cell states (55 identified via EcoTyper framework).
• Multi-omics datasets (transcriptomic, CNV, mutation, microbe data) from 620 samples were integrated.
• A Self-Normalizing Network (SNN) model incorporating scRNA-seq data was developed for prognosis prediction.

Main Results:

• The scRNA-seq enhanced SNN model demonstrated strong predictive performance (C-index 0.822 training, 0.817 test) with high AUC values at 1, 3, and 5 years.
• Gradient attribution analysis identified key molecular markers and enhanced model interpretability.
• High- and low-risk patient groups were validated as independent prognostic factors.

Conclusions:

• Integrating scRNA-seq and multi-omics data provides a robust approach for glioma prognosis.
• The developed model supports clinical decision-making and aids in identifying potential therapeutic strategies, such as HDAC inhibitors.