Two-pass imputation algorithm for missing value estimation in gene expression time series.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Gene expression data often contains missing values, hindering downstream analysis.
• Existing imputation methods are frequently unsuitable for time series expression profiles.
• Accurate estimation of missing values is crucial for reliable gene expression data mining and classification.

Purpose of the Study:

• To develop a novel imputation algorithm specifically designed for gene expression time series data.
• To evaluate the performance of different Dynamic Time Warping (DTW)-based imputation strategies.
• To provide an optimized C++ implementation of the most effective imputation algorithm.

Main Methods:

• Utilized Dynamic Time Warping (DTW) distance to measure similarity between time expression profiles.
• Developed and compared three DTW-based imputation algorithms: position-wise, neighborhood-wise, and two-pass.
• Evaluated algorithm accuracy on yeast expression time series data and benchmarked against weighted K-Nearest Neighbors (KNN).

Main Results:

• The two-pass DTWimpute algorithm demonstrated superior performance, especially with higher proportions of missing data.
• DTW-based imputation outperformed the widely used weighted KNN algorithm.
• The C++ implementation offers optimized performance and flexibility with initial imputation methods.

Conclusions:

• DTW techniques offer significant added value for missing value estimation in gene expression time series.
• The two-pass DTWimpute algorithm is a robust and accurate solution for handling missing data in genomic time series.
• The optimized C++ software facilitates improved analysis of gene expression time series data.