Multivariate piecewise linear regression model to predict radiosensitivity using the association with the genome-wide...

Area of Science:

• Genomics and Bioinformatics
• Radiation Oncology
• Cancer Biomarkers

Background:

• Predicting patient response to radiotherapy is crucial for personalized cancer treatment and radiation risk assessment.
• Genome-wide copy number variations (CNVs) are being investigated as potential biomarkers for radiosensitivity.

Purpose of the Study:

• To develop a machine-learning method for stratifying radiosensitivity using CNVs.
• To identify specific CNVs associated with different levels of sensitivity to ionizing radiation.

Main Methods:

• Utilized Affymetrix CytoScan HD microarrays to analyze CNVs in 129 fibroblast cell strains.
• Measured radiosensitivity via the surviving fraction at 2 Gy (SF2).
• Applied a dynamic programming (DP) algorithm for piecewise multivariate linear regression to predict SF2 and identify related CNVs.

Main Results:

• The DP algorithm segmented cell strains into radio-sensitive (RS), normally-sensitive (NS), and radio-resistant (RR) groups.
• A 5-segment model identified C-3SFBP (MCC gene region) as a marker for RS cells and C-7IUVU (SLC1A6 gene region) for RR cells.
• Found that decreased copy number generally correlated with increased radiosensitivity.

Conclusions:

• The DP-based method effectively narrows down CNV markers for radiosensitivity prediction.
• SF2 partitioning improves estimation and identifies distinct markers, aiding in dose adjustment for radiotherapy.
• Identified CNV markers can potentially guide personalized radiotherapy by identifying patients who benefit from dose reduction or escalation.