Doppelgänger spotting in biomedical gene expression data

Area of Science:

• Bioinformatics
• Machine Learning
• Genomics

Background:

• Doppelgänger effects (DEs) in machine learning (ML) involve chance sample similarities inflating model performance.
• This leads to overconfidence in ML model deployability, with no current tools for DE identification.
• DEs pose a significant challenge in biomedical data analysis.

Purpose of the Study:

• To introduce doppelgangerIdentifier, a novel software suite for identifying and verifying DEs.
• To demonstrate the prevalence of DEs in diverse biomedical gene expression datasets.
• To provide guidelines for managing DEs and improving ML model evaluation.

Main Methods:

• Development and application of the doppelgangerIdentifier software suite.
• Analysis of multiple disease and data types to assess DE prevalence.
• Exploration of batch effect influences on DE identification sensitivity.

Main Results:

• Doppelgänger effects are pervasive across various biomedical gene expression datasets.
• The doppelgangerIdentifier software effectively identifies and verifies DEs.
• Batch effects can impact the sensitivity of DE identification algorithms.

Conclusions:

• Doppelgänger effects are a widespread issue in biomedical ML, requiring specific identification tools.
• DoppelgangerIdentifier provides a solution for detecting and managing DEs.
• Doppelgänger verification is crucial for establishing reliable ML model evaluation baselines and ensuring meaningful insights from data.