sppPCA

sppPCA applies sequential projection pursuit principal component analysis to -omic datasets with non-random missing values to produce robust low-dimensional projections without imputing missing data.

Key Features:

• Missing-value handling: Operates directly on datasets with large numbers of non-random missing values without requiring imputation.
• Imputation-free variance estimation: Avoids variance distortions introduced by imputation methods by defining components without filling missing entries.
• Sequential projection pursuit: Sequentially projects the dataset to define principal components in the presence of missing entries.
• Robust low-dimensional projections: Produces lower-dimensional representations that are robust and informative for downstream analyses.
• Comparative performance: Demonstrated to outperform traditional imputation-based PCA approaches on complex -omic datasets.

Scientific Applications:

• Exploratory data analysis: Visualizing global structure and patterns in complex -omic datasets with missing values.
• Dimensionality reduction for label-free mass spectrometry: Reducing dimensionality of label-free mass spectrometry data while accounting for non-random missingness.
• Clustering and classification preprocessing: Providing reliable low-dimensional inputs for clustering and classification without imputation artifacts.
• Variance-sensitive analyses: Enabling analyses that require accurate variance estimates in the presence of missing data.

Methodology:

sppPCA sequentially projects the dataset to define principal components directly in the presence of missing entries, avoiding imputation and the resultant inaccurate variance estimates.