Comparing Imputation Procedures for Affymetrix Gene Expression Datasets Using MAQC Datasets.

Area of Science:

• Genomics
• Bioinformatics
• Microarray Analysis

Background:

• The MicroArray Quality Control (MAQC) project provided valuable datasets for assessing microarray precision and comparability.
• Performance evaluation of missing value imputation methods on MAQC datasets has not been previously reported.

Purpose of the Study:

• To evaluate the performance of various imputation procedures on Affymetrix microarray data from the MAQC project.
• To identify the most accurate and least accurate methods for handling missing values in this context.

Main Methods:

• Utilized MAQC Affymetrix datasets to assess imputation methods.
• Simulated 5% and 10% data deletions and imputed missing values.
• Employed 1000 simulations, averaging results for robust comparison.
• Evaluated performance using multiple error measures.

Main Results:

• Local least squares imputation with k=4 exhibited the highest accuracy across considered error measures.
• K-nearest neighbor imputation with k=1 resulted in the highest error rate.
• Results were consistent for both 5% and 10% data deletion scenarios.

Conclusions:

• For Affymetrix microarrays processed with MAS 5.0, local least squares (k=4) is the optimal imputation method for missing values.
• K-nearest neighbor (k=1) is the least effective imputation method for this dataset and preprocessing scheme.