ecpc: an R-package for generic co-data models for high-dimensional prediction.

Area of Science:

• Statistical modeling
• Machine learning
• Bioinformatics

Background:

• High-dimensional data prediction involves more variables than samples, aiming for optimal predictor identification or variable selection.
• Co-data, or complementary variable information, can enhance prediction accuracy by weighting variable importance.
• Existing methods using the R package ecpc had limitations in efficiently modeling continuous co-data.

Purpose of the Study:

• To extend the ecpc package for generic co-data models, specifically addressing continuous co-data.
• To improve the efficiency and performance of high-dimensional prediction and variable selection.
• To provide a more flexible framework for incorporating diverse co-data types.

Main Methods:

• Developed an extension for generic co-data models, utilizing a linear regression framework to model prior variance weights based on co-data.
• Employed empirical Bayes moment estimation for co-data variable estimation.
• Extended the methodology to accommodate generalized additive and shape-constrained co-data models and integrated ridge to elastic net penalties.

Main Results:

• The extended ecpc package (version 3.1.1+) demonstrates faster computation compared to the original method.
• Achieved improved prediction and variable selection performance, particularly for non-linear relationships within continuous co-data.
• Validated the enhanced methods through simulation studies and presented genomics examples.

Conclusions:

• The updated R package ecpc now supports linear, generalized additive, and shape-constrained additive co-data models for superior high-dimensional prediction and variable selection.
• This advancement offers a more robust and efficient tool for researchers dealing with complex, high-dimensional datasets.
• The enhanced package is publicly available for use in statistical and bioinformatics applications.