Related Experiment Video
Updated: Jun 2, 2026
03:37
Generating the Transcriptional Regulation View of Transcriptomic Features for Prediction Task and Dark Biomarker Detection on Small Datasets
Published on: March 1, 2024
Associations of Normalization and Regularization with Machine Learning Overfitting in Cross-dataset Classification of
Fei Deng1, Lanjing Zhang1,2,3,4
1Department of Chemical Biology, Ernest Mario School of Pharmacy, Rutgers University, Piscataway, NJ, USA.
Journal of Clinical and Translational Pathology
|June 1, 2026
Summary
Normalization improves machine learning (ML) performance within datasets but increases overfitting. Regularized models like LASSO offer better cross-dataset performance, suggesting cautious use of normalization in omics data analysis.
Area of Science:
- Bioinformatics
- Machine Learning in Omics Data Analysis
- Cancer Genomics
Background:
- Normalization is crucial for standardizing omics data to enhance machine learning (ML) model performance.
- The association between normalization techniques and overfitting (i.e., poor performance on unseen data) in ML models for omics data remains unclear.
- Investigating the interplay of normalization and regularization is essential for robust ML applications in omics.
Purpose of the Study:
- To examine the association between normalization methods and overfitting in machine learning models applied to omics data.
- To evaluate the impact of regularization techniques on mitigating overfitting in cross-dataset omics analyses.
- To compare the performance of ML models with and without normalization across different datasets.
Main Methods:
- Utilized three paired transcriptomic and clinical cancer datasets (lung adenocarcinoma, melanoma, glioblastoma).
- Applied ANOVA-based gene selection, six normalization methods, and six ML models for patient death classification.
- Compared intra-dataset and cross-dataset performance using balanced accuracy (BA) and area under the curve (AUC).
Main Results:
- Normalization consistently improved intra-dataset performance across all datasets.
- Normalization showed minimal or slightly decreased cross-dataset performance, indicating increased overfitting.
- The Least Absolute Shrinkage and Selection Operator (LASSO) model, without normalization, demonstrated superior cross-dataset performance.
Conclusions:
- Normalization enhances intra-dataset ML performance but exacerbates overfitting in cancer omics data.
- Regularized models, such as LASSO, effectively mitigate overfitting and yield robust cross-dataset results.
- Cross-dataset evaluation and the use of regularized models are recommended for reliable ML in omics; normalization should be applied judiciously.