Molecular signatures that can be transferred across different omics platforms

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics
• Proteomics

Background:

• Molecular signatures are vital for guiding cancer treatment decisions.
• Applying these signatures to data from diverse experimental protocols and platforms remains a significant challenge.

Purpose of the Study:

• To develop and validate a statistical approach for robust molecular signature application across different technical platforms.
• To assess the impact of technical variability on prediction accuracy and treatment recommendations.

Main Methods:

• Analysis of paired tumor data (Burkitt lymphoma, diffuse large B-cell lymphoma) using transcriptomics (nanoString nCounter, Affymetrix Gene Chip) and proteomics (SWATH, SRM) platforms.
• Development of a statistical model accounting for sample and feature effects to quantify technical variability.
• Evaluation of linear signatures, particularly those with feature weights summing to zero, for robustness.

Main Results:

• A statistical model successfully accounted for 69-94% of technical variability in the data.
• Linear signatures with feature weights summing to zero demonstrated superior robustness and consistency across platforms and data types.
• Stable predictions were achieved across transcriptomics and proteomics data, as well as between fresh-frozen and formalin-fixed paraffin-embedded tissues.

Conclusions:

• The developed statistical model and the 'zeroSum' R package offer a robust solution for applying molecular signatures across diverse data types and experimental conditions.
• This approach enhances the reliability of predictions, thereby improving the potential for accurate treatment recommendations in precision oncology.