KNNimpute

KNNimpute imputes missing values in gene expression microarray datasets using a weighted k-nearest neighbors algorithm to enable complete-data analyses such as hierarchical clustering and K-means clustering.

Key Features:

• Weighted KNN algorithm: Uses a weighted k-nearest neighbors approach to estimate missing entries.
• Gene/sample similarity: Leverages similarity between genes or samples to predict missing values from nearest neighbors.
• Microarray focus: Specifically targets missing values in gene expression microarray data matrices.
• Comparative evaluation: Was evaluated against SVDimpute (Singular Value Decomposition), row average, and zero-filling methods.
• Robustness across missingness levels: Demonstrated performance across 1% to 20% missing data and various parameter settings.
• Improves clustering analyses: Enhances applicability of downstream methods such as hierarchical clustering and K-means clustering by providing complete data matrices.

Scientific Applications:

• Clustering of gene expression data: Enables hierarchical clustering of microarray experiments by supplying imputed values for incomplete matrices.
• K-means clustering: Facilitates K-means clustering on datasets with missing microarray values.
• Robust gene expression analysis: Reduces the impact of incomplete data on sensitivity and robustness of downstream analyses.

Methodology:

Imputes missing microarray values using a weighted k-nearest neighbors algorithm that predicts entries from similar genes or samples; validated in a comparative study (PMID: 11395428) against SVDimpute (Singular Value Decomposition), row average, and zero-filling across parameter settings and 1% to 20% missingness on real datasets.