Major data analysis errors invalidate cancer microbiome findings.

Area of Science:

• Microbiome research
• Cancer diagnostics
• Bioinformatics

Background:

• A previous study reported high accuracy in classifying 33 cancer types using machine learning predictors based on microbial organism correlations.
• This study aimed to re-analyze the original data to validate these findings.

Approach:

• Re-analysis of the original dataset and computational methods.
• Identification and correction of errors in genome database and data transformation processes.
• Evaluation of machine learning classifier performance based on corrected data.

Key Points:

• Fundamental flaws identified in the original study's data and methods.
• Millions of false positive bacterial findings due to misidentification of human sequences.
• Artificial signatures created by data transformation errors, leading to inaccurate machine learning predictions.
• The original microbiome-based cancer classifiers are invalidated by these methodological errors.

Conclusions:

• The microbiome-based classifiers for cancer identification presented in the original study are entirely incorrect.
• Subsequent studies relying on the same flawed data are also likely invalid.
• This re-analysis highlights the critical importance of data integrity and rigorous methodology in microbiome and cancer research.