Deep models of integrated multiscale molecular data decipher the endothelial cell response to ionizing radiation

Area of Science:

• Radiation oncology
• Vascular biology
• Genomics and proteomics

Background:

• The vascular endothelium plays a critical role in radiation therapy response for both tumors and normal tissues.
• Developing effective radiation therapy requires understanding endothelial cell behavior and identifying targets to mitigate side effects and improve efficacy.

Purpose of the Study:

• To investigate the temporal dynamics of multi-omics data in irradiated endothelial cells.
• To develop a deep learning model for extracting key features from omics data to understand radiation-induced endothelial dysfunction.
• To identify endothelial targets for improving radiation therapy outcomes.

Main Methods:

• Cultured primary human endothelial cells and exposed them to ionizing radiation.
• Collected multi-omics data, including miRNome, transcriptome, proteome, and metabolome, over time.
• Applied deep learning, specifically convolutional graph networks, for unsupervised feature extraction from the omics data.

Main Results:

• The study captured time-dependent multi-omics changes in irradiated endothelial cells.
• A deep learning model successfully extracted high-level features indicative of radiation-induced endothelial dysfunction.
• Experimental validation confirmed that identified features are involved in radiation-induced alterations of angiogenesis.

Conclusions:

• The developed deep learning strategy effectively analyzes multi-omics data to understand radiation effects on endothelial cells.
• The findings provide insights into endothelial cell responses to radiation, potentially leading to new therapeutic targets.
• This approach can help radiobiologists and radiation oncologists optimize radiation therapy by targeting endothelial pathways.