Detect influential points of feature rankings.

Area of Science:

• Bioinformatics
• Data Analysis
• Statistical Modeling

Background:

• Feature rankings are essential in bioinformatics for data interpretation.
• Influential points (IPs) can significantly distort these rankings, often going undetected.
• Overlooked IPs can lead to inaccurate biological insights.

Purpose of the Study:

• To investigate the impact of influential points on feature rankings in bioinformatics.
• To develop and evaluate a novel method for detecting influential points.
• To underscore the importance of identifying IPs for reliable downstream analyses.

Main Methods:

• A leave-one-out approach was employed to assess the influence of individual data points.
• A novel rank comparison method utilizing adaptive top-prioritized weights was developed.
• The proposed IP detection method was validated on multiple public datasets, including TCGA gene expression data.

Main Results:

• The developed method successfully identified influential points in gene expression datasets.
• Results demonstrate that IPs can substantially distort feature rankings.
• Detected rank distortions can negatively impact subsequent analyses, such as pathway enrichment.

Conclusions:

• Influential points have a significant impact on feature rankings and subsequent bioinformatics analyses.
• Routine detection of influential points is crucial but currently underutilized.
• The developed IP detection method is available as an R package named 'findIPs'.