Addressing heterogeneous sensitivity in biomarker screening with application in NanoString nCounter data

Area of Science:

• Biomedical science
• Statistical genetics
• Bioinformatics

Background:

• Biomarkers are crucial for disease screening and prognosis, often identified using high-throughput methods.
• Challenges in biomarker detection include background noise and variable signal strength, leading to reduced sensitivity and inflated false positive rates.
• Current methods often overlook sample-specific detection heterogeneity, compromising biomarker discovery accuracy.

Purpose of the Study:

• To introduce a novel statistical method, the sensitivity adjusted likelihood-ratio test (SALT), for robust biomarker screening.
• To address the issue of overlooked heterogeneity in detection sensitivity in high-throughput molecular measurements.
• To improve the accuracy and power of biomarker discovery by properly controlling false positive rates.

Main Methods:

• Developed the sensitivity adjusted likelihood-ratio test (SALT) to account for sample- and feature-specific detection sensitivities.
• Utilized NanoString nCounter data to estimate these individual detection sensitivities.
• Compared the performance of SALT against unadjusted methods in biomarker screening simulations and analyses.

Main Results:

• Demonstrated that sample- and feature-specific detection sensitivities can be reliably estimated from NanoString nCounter data.
• Showcased that incorporating these estimated sensitivities into SALT significantly improves biomarker screening.
• SALT effectively controls false positive rates and offers increased statistical power compared to unadjusted approaches.

Conclusions:

• The sensitivity adjusted likelihood-ratio test (SALT) provides a statistically sound framework for biomarker screening in high-throughput data.
• Accurate estimation and integration of detection sensitivity are critical for reliable biomarker discovery.
• SALT represents a significant advancement in improving the reliability and power of molecular biomarker identification.