iMISS

iMISS estimates missing values in microarray datasets by integrating information from multiple reference microarray datasets to derive neighbor genes and improve local imputation algorithms such as Local Least Squares (LLS).

Key Features:

• Integrative Approach: Integrates data from multiple reference microarray datasets to identify reliable neighbor genes for imputation.
• Neighbor-Gene Derivation: Derives a consistent list of neighboring genes for each gene with missing values by aggregating information across reference datasets.
• Submatrix Imputation Assessment: Employs submatrix imputation to evaluate whether individual reference datasets are sufficiently informative for integration.
• Order-Statistics-Based Integration: Incorporates order-statistics-based integrative imputation algorithms to combine evidence across datasets.
• Enhanced Algorithm Performance: Improves accuracy of local estimation algorithms, reporting up to 15% improvement over state-of-the-art Local Least Squares (LLS) and showing gains relative to K-Nearest Neighbors (KNN) in datasets with limited samples, high missingness, or noisy measurements.

Scientific Applications:

• Microarray preprocessing: Performs missing-value imputation during preprocessing of microarray data to mitigate the impact of missing entries on downstream analyses.
• Time-series and complex experiments: Supports imputation in time-series or other complex microarray experiments where sample size or data complexity challenges standard methods.
• Small, noisy, or high-missingness datasets: Targets datasets with limited samples, high missing-data rates, or noisy measurements to improve reliability of downstream biological interpretation.

Methodology:

Identify neighbor genes using reference microarray datasets, apply submatrix imputation to assess dataset informativeness, and integrate reference information to enhance local estimation algorithms such as LLS.