Detecting fabrication in large-scale molecular omics data.

Area of Science:

• Biomedical Research
• Computational Biology
• Data Science

Background:

• Scientific fraud, including data fabrication and falsification, is a significant issue within the research community.
• Traditional methods like peer review and audits are insufficient for detecting sophisticated, computationally-driven fraud in big-data omics.
• Advances in high-throughput omics technologies generate vast datasets, necessitating advanced fraud detection techniques.

Purpose of the Study:

• To develop and evaluate machine learning methods for detecting data fabrication in biomedical research.
• To adapt fraud detection strategies from the financial sector, specifically digit-frequency analysis, for scientific applications.
• To assess the efficacy of machine learning in identifying fraudulent patterns within large-scale omic datasets.

Main Methods:

• Utilized machine learning models to analyze gene copy-number data for fraud detection.
• Employed digit-frequency analysis as a feature set for machine learning models.
• Made all data and analysis scripts publicly available for transparency and reproducibility.

Main Results:

• Machine learning models achieved 58-100% accuracy in predicting fraud using gene copy-number data.
• Models utilizing digit frequency as input features demonstrated high accuracy, ranging from 82% to 100%.
• These findings indicate machine learning's potential for robust fraud detection in omics research.

Conclusions:

• Machine learning offers a powerful computational tool for identifying fabricated data in biomedical big-data.
• Digit-frequency analysis is a viable and effective feature for detecting scientific misconduct.
• Updated methods are crucial for maintaining research integrity in the era of omics and big-data.