Predicting ligand-dependent tumors from multi-dimensional signaling features

Area of Science:

• Oncology
• Computational Biology
• Molecular Biology

Background:

• Targeted therapies benefit a small subset of solid tumors driven by single oncogenes.
• High tumor ligand levels correlate with poor patient survival, yet unselected therapies show limited efficacy.
• Ligand addiction is proposed as a novel driver of tumor growth.

Purpose of the Study:

• To explore ligand addiction as a driver of tumor growth.
• To develop a predictive model for ligand-dependent cancer cell proliferation.
• To investigate the relationship between growth factors and tumor growth in vivo.

Main Methods:

• Utilized Bagged Decision Trees (BDT) on high-dimensional signaling features from a computational model.
• Developed a mechanistic, multi-pathway model incorporating receptor heterodimerization.
• Trained the model on seven cancer cell lines and validated its predictive capacity on two independent cell lines by adjusting receptor expression.

Main Results:

• Successfully predicted ligand-dependent proliferation across 58 cell lines using the BDT approach.
• The computational model accurately predicted signaling in independent cell lines by modifying receptor expression levels.
• Predicted tumor growth response in patient samples correlated with high growth factor expression in the tumor microenvironment.

Conclusions:

• Ligand addiction represents a potential driver of tumor growth.
• The developed computational model offers a method to predict ligand-dependent proliferation.
• Co-evolution of growth factors and tumor microenvironment is suggested, impacting in vivo tumor growth.