Improvement Technologies for Data Imputation in Bioinformatics

Abstract

Using existing software technologies for imputing missing genetic data (GD), such as Beagle, HPImpute, Impute, MACH, AlphaPlantImpute, MissForest, and LinkImputeR, has its advantages and disadvantages. The wide range of input parameters and their nonlinear dependence on the target results require a lot of time and effort to find optimal values in each specific case. Thus, optimizing resources for GD imputation and improving its quality is an important current issue for the quality analysis of digitized deoxyribonucleic acid (DNA) samples. This work provides a critical analysis of existing methods and approaches for obtaining high-quality imputed GD. We observed that most of them do not investigate the problem of time and resource costs, which play a significant role in a mass approach. It is also worth noting that the considered articles are often characterized by high development complexity and, at times, unclear (or missing) descriptions of the input parameters for the methods, algorithms, or models under consideration. As a result, two algorithms were developed in this work. The first one aims to optimize the imputation time, allowing for real-time solutions, while the second one aims to improve imputation accuracy by selecting the best results at each iteration. The success of the first algorithm in improving imputation speed ranges from 47% (for small files) to 87% of the time (for medium and larger files), depending on the available resources. For the second algorithm, the accuracy has been improved by about 0.1%. This, in turn, encourages continued research on the latest version of Beagle software, particularly in the selection of optimal input parameters and possibly other models with similar or higher imputation accuracy.