Validation of In Vitro Trained Transcriptomic Radiosensitivity Signatures in Clinical Cohorts.

Area of Science:

• Oncology
• Genomics
• Bioinformatics

Background:

• Transcriptomic personalization of radiation therapy (RT) is a growing field.
• Previous in vitro studies showed poor performance of independent models.
• Reproducibility of radiosensitivity signatures in clinical settings needs thorough assessment.

Purpose of the Study:

• To evaluate the clinical reproducibility of published radiosensitivity signatures.
• To assess the impact of microarray normalization methods on radiosensitivity predictions.
• To benchmark the performance of signatures against control models in clinical cohorts.

Main Methods:

• Assessed agreement between radiosensitivity predictions using different normalization methods.
• Benchmarked control signatures from resampled in vitro data in clinical cohorts.
• Performed survival analysis and gene set enrichment analysis on clinical transcriptomic data.

Main Results:

• Normalization methods significantly impacted radiosensitivity index (RSI) values, with limits of agreement exceeding 20%.
• No published signature demonstrated significant improvement over resampled controls for predicting clinical outcomes.
• Overlapping biological processes, including proliferation and immune responses, were associated with cancer outcomes in both RT-treated and non-RT-treated patients.

Conclusions:

• Different normalization methods are not interchangeable and affect radiosensitivity predictions.
• The clinical utility of published radiosensitivity signatures remains uncertain due to overlap with general cancer outcome genes.
• Existing signatures may lack specificity as key biological processes influencing outcomes are shared between radiation-treated and non-treated patients.