Improving Cancer Gene Expression Data Quality through a TCGA Data-Driven Evaluation of Identifier Filtering.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Data quality is a significant challenge in high-throughput genomics, leading to numerous filtering methods.
• Existing filtering methods yield inconsistent results, lacking clear guidance for researchers.
• Computational tools to aid analysts in selecting optimal filtering strategies are scarce.

Purpose of the Study:

• To evaluate and compare the performance of different probeset filtering methods for high-throughput genomics data.
• To provide computational support for analysts in choosing optimal filtering strategies based on research needs.
• To assess the utility of different filtering methods using paired transcriptomic and proteomic datasets.

Main Methods:

• Utilized paired transcriptomic and proteomic expression datasets from cancer studies as a testbed.
• Developed an evaluation framework using identifier mapping and correlation analysis of feature pairs.
• Estimated posterior probabilities to assess the correctness of filtered probesets and incorporated analyst-defined utilities.

Main Results:

• Tested nine published probeset filtering methods and combination strategies across two distinct testbeds.
• The Jetset filtering method demonstrated optimal performance for probeset filtering on both transcriptomic and proteomic data.
• The necessity of combining Jetset with a second filtering method varied depending on the specific testbed.

Conclusions:

• The Jetset method is a robust and effective tool for probeset filtering in high-throughput genomics, applicable across different data types.
• The study provides a framework for evaluating filtering methods, enabling informed decisions for data quality control.
• Researchers can leverage these findings to improve the reliability and interpretability of genomic data analysis.