Related Experiment Video
Updated: Aug 28, 2025
09:43
Author Spotlight: Unveiling the Polyfunctionality and Heterogeneity in Immune Responses
Published on: March 8, 2024
1.8K
Envelope-based partial partial least squares with application to cytokine-based biomarker analysis for COVID-19
Yeonhee Park1, Zhihua Su2, Dongjun Chung3
1Department of Biostatistics and Medical Informatics, University of Wisconsin-Madison, Madison, Wisconsin, USA.
Statistics in Medicine
|September 16, 2022
Summary
This study introduces an envelope-based partial least squares (PLS) estimator to improve prediction accuracy when using categorical variables in regression analysis. The new method offers more efficient estimation and better predictions, as demonstrated in a COVID-19 biomarker identification.
Area of Science:
- Statistics
- Biostatistics
- Machine Learning
Background:
- Partial least squares (PLS) regression is widely used for its predictive power, but treating categorical variables as continuous can lead to biased results.
- Existing methods in PLS regression often struggle with mixed continuous and categorical predictors, potentially compromising estimation accuracy and inference validity.
Purpose of the Study:
- To develop a novel envelope-based partial PLS estimator that properly accounts for categorical predictors.
- To establish the theoretical properties (root-n consistency and asymptotic normality) of the proposed estimator.
- To demonstrate the practical utility and efficiency gains of the new method in real-world applications, such as biomarker identification.
Main Methods:
- Developed an envelope-based partial PLS estimator leveraging the connection between envelope models and PLS.
- The estimator performs PLS regression on conditional distributions, considering both continuous and categorical predictors.
- Established theoretical guarantees including root-n consistency and asymptotic normality for the new estimator.
Main Results:
- The proposed envelope-based partial PLS estimator achieves significant efficiency gains in estimation compared to traditional methods.
- Numerical studies confirm superior prediction performance and more accurate inferences when handling mixed predictor types.
- Application to COVID-19 patient data identified significant cytokine-based biomarkers and their association with clinical characteristics.
Conclusions:
- The envelope-based partial PLS approach offers a statistically robust and computationally efficient alternative for regression with mixed predictor types.
- This method enhances the reliability of statistical inference and predictive modeling in complex datasets.
- The successful application in identifying COVID-19 biomarkers highlights its potential for advancing precision medicine and biological research.
