PC-DOT

PC-DOT implements a dot-product (cosine similarity) ranking method for principal component regression to select optimal principal components for genomic prediction.

Key Features:

• PC-DOT ranking: Ranks principal components by dot product (cosine similarity), prioritizing PCs that are both highly correlated with the response variable and have substantial eigenvalues.
• Comparison frameworks: Benchmarked against PC-Eigen and PC-SS and demonstrated superior prediction ability across one simulated dataset and three real genomic datasets encompassing 15 traits.
• Computational efficiency: Employs the HAT matrix to reduce computational complexity during cross-validation on reference data.
• Implementation: Provided as an R-based implementation for principal component regression workflows.

Scientific Applications:

• Genomic prediction: Selection of optimal PCs for predictive modeling of complex traits in genomic datasets.
• Dimensionality reduction and interpretation: Improves dimensionality reduction for high-dimensional genomic data to produce more interpretable regression models.

Methodology:

Ranks PCs within principal component regression using dot product (cosine similarity), compares performance to PC-Eigen and PC-SS, uses the HAT matrix to streamline cross-validation computations, and is implemented in R; evaluated on one simulated and three real genomic datasets covering 15 traits.