Improved Imputation Accuracy Research Articles

Missing Value Imputation Method for Multiclass Matrix Data Based on Closed Itemset

Handling missing values in matrix data is an important step in data analysis. To date, many methods to estimate missing values based on data pattern similarity have been proposed. Most previously proposed methods perform missing value imputation based on data trends over the entire feature space. However, individual missing values are likely to show similarity to data patterns in local feature space. In addition, most existing methods focus on single class data, while multiclass analysis is frequently required in various fields. Missing value imputation for multiclass data must consider the characteristics of each class. In this paper, we propose two methods based on closed itemsets, CIimpute and ICIimpute, to achieve missing value imputation using local feature space for multiclass matrix data. CIimpute estimates missing values using closed itemsets extracted from each class. ICIimpute is an improved method of CIimpute in which an attribute reduction process is introduced. Experimental results demonstrate that attribute reduction considerably reduces computational time and improves imputation accuracy. Furthermore, it is shown that, compared to existing methods, ICIimpute provides superior imputation accuracy but requires more computational time.

Genome-wide imputation using the practical haplotype graph in the heterozygous crop cassava.

Genomic applications such as genomic selection and genome-wide association have become increasingly common since the advent of genome sequencing. The cost of sequencing has decreased in the past two decades; however, genotyping costs are still prohibitive to gathering large datasets for these genomic applications, especially in nonmodel species where resources are less abundant. Genotype imputation makes it possible to infer whole-genome information from limited input data, making large sampling for genomic applications more feasible. Imputation becomes increasingly difficult in heterozygous species where haplotypes must be phased. The practical haplotype graph (PHG) is a recently developed tool that can accurately impute genotypes, using a reference panel of haplotypes. We showcase the ability of the PHG to impute genomic information in the highly heterozygous crop cassava (Manihot esculenta). Accurately phased haplotypes were sampled from runs of homozygosity across a diverse panel of individuals to populate PHG, which proved more accurate than relying on computational phasing methods. The PHG achieved high imputation accuracy, using sparse skim-sequencing input, which translated to substantial genomic prediction accuracy in cross-validation testing. The PHG showed improved imputation accuracy, compared to a standard imputation tool Beagle, especially in predicting rare alleles.

Bridging gaps in demographic analysis with phylogenetic imputation.

Phylogenetically informed imputation methods have rarely been applied to estimate missing values in demographic data but may be a powerful tool for reconstructing vital rates of survival, maturation, and fecundity for species of conservation concern. Imputed vital rates could be used to parameterize demographic models to explore how populations respond when vital rates are perturbed. We used standardized vital rate estimates for 50 bird species to assess the use of phylogenetic imputation to fill gaps in demographic data. We calculated imputation accuracy for vital rates of focal species excluded from the data set either singly or in combination and with and without phylogeny, body mass, and life-history trait data. We used imputed vital rates to calculate demographic metrics, including generation time, to validate the use of imputation in demographic analyses. Covariance among vital rates and other trait data provided a strong basis to guide imputation of missing vital rates in birds, even in the absence of phylogenetic information. Mean NRMSE for null and phylogenetic models differed by <0.01 except when no vital rates were available or for vital rates with high phylogenetic signal (Pagel's λ > 0.8). In these cases, including body mass and life-history trait data compensated for lack of phylogenetic information: mean normalized root mean square error (NRMSE) for null and phylogenetic models differed by <0.01 for adult survival and <0.04 for maturation rate. Estimates of demographic metrics were sensitive to the accuracy of imputed vital rates. For example, mean error in generation time doubled in response to inaccurate estimates of maturation time. Accurate demographic data and metrics, such as generation time, are needed to inform conservation planning processes, for example through International Union for Conservation of Nature Red List assessments and population viability analysis. Imputed vital rates could be useful in this context but, as for any estimated model parameters, awareness of the sensitivities of demographic model outputs to the imputed vital rates is essential.

Efficient phasing and imputation of low-coverage sequencing data using large reference panels.

Low-coverage whole-genome sequencing followed by imputation has been proposed as a cost-effective genotyping approach for disease and population genetics studies. However, its competitiveness against SNP arrays is undermined because current imputation methods are computationally expensive and unable to leverage large reference panels. Here, we describe a method, GLIMPSE, for phasing and imputation of low-coverage sequencing datasets from modern reference panels. We demonstrate its remarkable performance across different coverages and human populations. GLIMPSE achieves imputation of a genome for less than US$1 in computational cost, considerably outperforming other methods and improving imputation accuracy over the full allele frequency range. As a proof of concept, we show that 1× coverage enables effective gene expression association studies and outperforms dense SNP arrays in rare variant burden tests. Overall, this study illustrates the promising potential of low-coverage imputation and suggests a paradigm shift in the design of future genomic studies.

Model-based autoencoders for imputing discrete single-cell RNA-seq data.

Deep neural networks have been widely applied for missing data imputation. However, most existing studies have been focused on imputing continuous data, while discrete data imputation is under-explored. Discrete data is common in real world, especially in research areas of bioinformatics, genetics, and biochemistry. In particular, large amounts of recent genomic data are discrete count data generated from single-cell RNA sequencing (scRNA-seq) technology. Most scRNA-seq studies produce a discrete matrix with prevailing 'false' zero count observations (missing values). To make downstream analyses more effective, imputation, which recovers the missing values, is often conducted as the first step in pre-processing scRNA-seq data. In this paper, we propose a novel Zero-Inflated Negative Binomial (ZINB) model-based autoencoder for imputing discrete scRNA-seq data. The novelties of our method are twofold. First, in addition to optimizing the ZINB likelihood, we propose to explicitly model the dropout events that cause missing values by using the Gumbel-Softmax distribution. Second, the zero-inflated reconstruction is further optimized with respect to the raw count matrix. Extensive experiments on simulation datasets demonstrate that the zero-inflated reconstruction significantly improves imputation accuracy. Real data experiments show that the proposed imputation can enhance separating different cell types and improve the accuracy of differential expression analysis.

Does the leaf economic spectrum hold within plant functional types? A Bayesian multivariate trait meta-analysis.

The leaf economic spectrum is a widely studied axis of plant trait variability that defines a trade-off between leaf longevity and productivity. While this has been investigated at the global scale, where it is robust, and at local scales, where deviations from it are common, it has received less attention at the intermediate scale of plant functional types (PFTs). We investigated whether global leaf economic relationships are also present within the scale of plant functional types (PFTs) commonly used by Earth System models, and the extent to which this global-PFT hierarchy can be used to constrain trait estimates. We developed a hierarchical multivariate Bayesian model that assumes separate means and covariance structures within and across PFTs and fit this model to seven leaf traits from the TRY database related to leaf longevity, morphology, biochemistry, and photosynthetic metabolism. Although patterns of trait covariation were generally consistent with the leaf economic spectrum, we found three approximate tiers to this consistency. Relationships among morphological and biochemical traits (specific leaf area [SLA], N, P) were the most robust within and across PFTs, suggesting that covariation in these traits is driven by universal leaf construction trade-offs and stoichiometry. Relationships among metabolic traits (dark respiration [Rd ], maximum RuBisCo carboxylation rate [Vc,max ], maximum electron transport rate [Jmax ]) were slightly less consistent, reflecting in part their much sparser sampling (especially for high-latitude PFTs), but also pointing to more flexible plasticity in plant metabolistm. Finally, relationships involving leaf lifespan were the least consistent, indicating that leaf economic relationships related to leaf lifespan are dominated by across-PFT differences and that within-PFT variation in leaf lifespan is more complex and idiosyncratic. Across all traits, this covariance was an important source of information, as evidenced by the improved imputation accuracy and reduced predictive uncertainty in multivariate models compared to univariate models. Ultimately, our study reaffirms the value of studying not just individual traits but the multivariate trait space and the utility of hierarchical modeling for studying the scale dependence of trait relationships.

A bird\u2019s-eye view of Italian genomic variation through whole-genome sequencing

The genomic variation of the Italian peninsula populations is currently under characterised: the only Italian whole-genome reference is represented by the Tuscans from the 1000 Genome Project. To address this issue, we sequenced a total of 947 Italian samples from three different geographical areas. First, we defined a new Italian Genome Reference Panel (IGRP1.0) for imputation, which improved imputation accuracy, especially for rare variants, and we tested it by GWAS analysis on red blood traits. Furthermore, we extended the catalogue of genetic variation investigating the level of population structure, the pattern of natural selection, the distribution of deleterious variants and occurrence of human knockouts (HKOs). Overall the results demonstrate a high level of genomic differentiation between cohorts, different signatures of natural selection and a distinctive distribution of deleterious variants and HKOs, confirming the necessity of distinct genome references for the Italian population.

Genotype imputation and reference panel: a systematic evaluation on haplotype size and diversity.

Here, 622 imputations were conducted with 394 customized reference panels for Han Chinese and European populations. Besides validating the fact that imputation accuracy could always benefit from the increased panel size when the reference panel was population specific, the results brought two new thoughts. First, when the haplotype size of the reference panel was fixed, the imputation accuracy of common and low-frequency variants (Minor Allele Frequency (MAF) > 0.5%) decreased while the population diversity of the reference panel increased, but for rare variants (MAF < 0.5%), a small fraction of diversity in panel could improve imputation accuracy. Second, when the haplotype size of the reference panel was increased with extra population-diverse samples, the imputation accuracy of common variants (MAF > 5%) for the European population could always benefit from the expanding sample size. However, for the Han Chinese population, the accuracy of all imputed variants reached the highest when reference panel contained a fraction of an extra diverse sample (8-21%). In addition, we evaluated the imputation performances in the existing reference panels, such as the Haplotype Reference Consortium (HRC), 1000 Genomes Project Phase 3 and the China, Oxford and Virginia Commonwealth University Experimental Research on Genetic Epidemiology (CONVERGE). For the European population, the HRC panel showed the best performance in our analysis. For the Han Chinese population, we proposed an optimum imputation reference panel constituent ratio if researchers would like to customize their own sequenced reference panel, but a high-quality and large-scale Chinese reference panel was still needed. Our findings could be generalized to the other populations with conservative genome; a tool was provided to investigate other populations of interest (https://github.com/Abyss-bai/reference-panel-reconstruction).

NARD: whole-genome reference panel of 1779 Northeast Asians improves imputation accuracy of rare and low-frequency variants

Here, we present the Northeast Asian Reference Database (NARD), including whole-genome sequencing data of 1779 individuals from Korea, Mongolia, Japan, China, and Hong Kong. NARD provides the genetic diversity of Korean (n = 850) and Mongolian (n = 384) ancestries that were not present in the 1000 Genomes Project Phase 3 (1KGP3). We combined and re-phased the genotypes from NARD and 1KGP3 to construct a union set of haplotypes. This approach established a robust imputation reference panel for Northeast Asians, which yields the greatest imputation accuracy of rare and low-frequency variants compared with the existing panels. NARD imputation panel is available at https://nard.macrogen.com/.

Improving Imputation Accuracy by Inferring Causal Variants in Genetic Studies.

Genotype imputation has been widely utilized for two reasons in the analysis of genome-wide association studies (GWAS). One reason is to increase the power for association studies when causal single nucleotide polymorphisms are not collected in the GWAS. The second reason is to aid the interpretation of a GWAS result by predicting the association statistics at untyped variants. In this article, we show that prediction of association statistics at untyped variants that have an influence on the trait produces is overly conservative. Current imputation methods assume that none of the variants in a region (locus consists of multiple variants) affect the trait, which is often inconsistent with the observed data. In this article, we propose a new method, CAUSAL-Imp, which can impute the association statistics at untyped variants while taking into account variants in the region that may affect the trait. Our method builds on recent methods that impute the marginal statistics for GWAS by utilizing the fact that marginal statistics follow a multivariate normal distribution. We utilize both simulated and real data sets to assess the performance of our method. We show that traditional imputation approaches underestimate the association statistics for variants involved in the trait, and our results demonstrate that our approach provides less biased estimates of these association statistics.

Imputation-Aware Tag SNP Selection To Improve Power for Large-Scale, Multi-ethnic Association Studies.

The emergence of very large cohorts in genomic research has facilitated a focus on genotype-imputation strategies to power rare variant association. These strategies have benefited from improvements in imputation methods and association tests, however little attention has been paid to ways in which array design can increase rare variant association power. Therefore, we developed a novel framework to select tag SNPs using the reference panel of 26 populations from Phase 3 of the 1000 Genomes Project. We evaluate tag SNP performance via mean imputed r2 at untyped sites using leave-one-out internal validation and standard imputation methods, rather than pairwise linkage disequilibrium. Moving beyond pairwise metrics allows us to account for haplotype diversity across the genome for improve imputation accuracy and demonstrates population-specific biases from pairwise estimates. We also examine array design strategies that contrast multi-ethnic cohorts vs. single populations, and show a boost in performance for the former can be obtained by prioritizing tag SNPs that contribute information across multiple populations simultaneously. Using our framework, we demonstrate increased imputation accuracy for rare variants (frequency < 1%) by 0.5–3.1% for an array of one million sites and 0.7–7.1% for an array of 500,000 sites, depending on the population. Finally, we show how recent explosive growth in non-African populations means tag SNPs capture on average 30% fewer other variants than in African populations. The unified framework presented here will enable investigators to make informed decisions for the design of new arrays, and help empower the next phase of rare variant association for global health.

Method for Imputing Missing Data using Online Calibration for Urban Freeway Control

Real-time traffic control systems are widely implemented on roadways around the world as a measure to improve freeway mobility. However, the systems, which rely on data from road-side and on-road sensors and other electronic equipment, continue to suffer from issues related to missing and erroneous data. While many data imputation methods are documented in the related literature, traffic control systems still lack an imputation method that is applicable in practice, accurate in imputation, and simple in computation. In response, this paper puts forth a linear imputation model that considers both temporal traffic trend and spatial detector correlations. To adapt the model to dynamic traffic variations, the imputation method was equipped with an online calibration module. The proposed imputation method was evaluated with field data from two stations on the Whitemud Drive, a busy urban freeway in Edmonton, Alberta, Canada. The proposed model benefited from its time-of-day temporal trend and outperforms the previous model that considers only spatial correlations. Moreover, the online calibration module was effective in improving imputation accuracy. Finally, the sensitivity of imputation performance was analyzed. The results show that the imputation with online calibration is more sensitive to missing data ratios than that with offline calibration. The sensitivity analysis revealed that imputation with online calibration is more suitable for online imputation in traffic control implementations.

Improving imputation in disease-relevant regions: lessons from cystic fibrosis

Does genotype imputation with public reference panels identify variants contributing to disease? Genotype imputation using the 1000 Genomes Project (1KG; 2504 individuals) displayed poor coverage at the causal cystic fibrosis (CF) transmembrane conductance regulator (CFTR) locus for the International CF Gene Modifier Consortium. Imputation with the larger Haplotype Reference Consortium (HRC; 32,470 individuals) displayed improved coverage but low sensitivity of variants clinically relevant for CF. A hybrid reference that combined whole genome sequencing (WGS) from 101 CF individuals with the 1KG imputed a greater number of single-nucleotide variants (SNVs) that would be analyzed in a genetic association study (r2 ≥ 0.3 and MAF ≥ 0.5%) than imputation with the HRC, while the HRC excelled in the lower frequency spectrum. Using the 1KG or HRC as reference panels missed the most common CF-causing variants or displayed low imputation accuracy. Designs that incorporate population-specific WGS can improve imputation accuracy at disease-specific loci, while imputation using public data sets can omit disease-relevant genotypes.

Collaborative Phenotype Inference from Comorbid Substance Use Disorders and Genotypes.

Data in large-scale genetic studies of complex human diseases, such as substance use disorders, are often incomplete. Despite great progress in genotype imputation, e.g., the IMPUTE2 method, considerably less progress has been made in inferring phenotypes. We designed a novel approach to integrate individuals' comorbid conditions with their genotype data to infer missing (unreported) diagnostic criteria of a disorder. The premise of our approach derives from correlations among symptoms and the shared biological bases of concurrent disorders such as co-dependence on cocaine and opioids. We describe a matrix completion method to construct a bi-linear model based on the interactions of genotypes and known symptoms of related disorders to infer unknown values of another set of symptoms or phenotypes. An efficient stochastic and parallel algorithm based on the linearized alternating direction method of multipliers was developed to solve the proposed optimization problem. Empirical evaluation of the approach in comparison with other advanced data matrix completion methods via a case study shows that it both significantly improves imputation accuracy and provides greater computational efficiency.

Inclusion of Population-specific Reference Panel from India to the 1000 Genomes Phase 3 Panel Improves Imputation Accuracy

Imputation is a computational method based on the principle of haplotype sharing allowing enrichment of genome-wide association study datasets. It depends on the haplotype structure of the population and density of the genotype data. The 1000 Genomes Project led to the generation of imputation reference panels which have been used globally. However, recent studies have shown that population-specific panels provide better enrichment of genome-wide variants. We compared the imputation accuracy using 1000 Genomes phase 3 reference panel and a panel generated from genome-wide data on 407 individuals from Western India (WIP). The concordance of imputed variants was cross-checked with next-generation re-sequencing data on a subset of genomic regions. Further, using the genome-wide data from 1880 individuals, we demonstrate that WIP works better than the 1000 Genomes phase 3 panel and when merged with it, significantly improves the imputation accuracy throughout the minor allele frequency range. We also show that imputation using only South Asian component of the 1000 Genomes phase 3 panel works as good as the merged panel, making it computationally less intensive job. Thus, our study stresses that imputation accuracy using 1000 Genomes phase 3 panel can be further improved by including population-specific reference panels from South Asia.

Abstract 1304: The contribution of rare and low-frequency variants to colorectal cancer heritability

Abstract Studies that estimate complex disease heritability based on genome-wide common SNP array data have shown that a large fraction of heritability is contributed from variants that do not reach genome-wide significance at current genome-wide association study (GWAS) sample sizes. The contribution of rare variants to heritability has yet to be explored for many complex diseases. Despite the decreasing cost of sequencing, it still remains prohibitively expensive to sequence sufficient samples for well-powered genetic association studies of rare variants. However, with increasingly large and denser imputation reference panels it has become feasible to accurately impute variants with minor allele frequencies (MAFs) as low as 0.1%, enabling study of a subset of rare risk variants. In previous work, we used restricted maximum likelihood (REML) to estimate the total additive heritability of colorectal cancer (CRC) based on common SNP array genotypes. Here, we expand this work using imputed genotype data and a larger sample size. We performed whole-genome sequencing of 1,961 CRC cases and 981 controls, and subsequently imputed these haplotypes into 11,895 unrelated CRC cases and 14,659 unrelated controls that are part of the Colorectal Cancer Family Registry (CCFR) and the Genetics and Epidemiology of Colorectal Cancer Consortium (GECCO). We estimated heritability from individual-level imputed genotype data using LD- and MAF-stratified GREML, as implemented in GCTA. In total, we analyzed 17,649,167 imputed genetic variants with minor allele count &amp;gt;3. Published heritability estimates for CRC from family-based studies vary from 12% to 35%. Based on common genotyped SNPs, we previously estimated heritability to be 7.42% (95% CI: 4.71-10.12%) on the underlying liability scale, assuming a population prevalence of 0.004. For imputed genotypes, we estimate the total heritability to be 12.0% (95% CI: 9.65-14.35). Using a likelihood ratio test, we demonstrate a significant contribution of variants with MAF ≤1% to CRC genetic risk (P=0.003). Because of the imperfect imputation accuracy for very rare variants, their contribution is likely higher. These results suggest that with additional sequencing, improved imputation accuracy, and larger GWAS, we should expect to start discovering rare variant associations for CRC risk. Citation Format: Jeroen R. Huyghe, Sai Chen, Hyun M. Kang, Tabitha Harrison, Sonja I. Berndt, Stephane Bézieau, Hermann Brenner, Graham Casey, Andrew T. Chan, Jenny Chang-Claude, Steven J. Gallinger, Stephen B. Gruber, Andrea Gsur, Michael Hoffmeister, Thomas Hudson, Mark A. Jenkins, Loic Le Marchand, Polly A. Newcomb, John D. Potter, Conghui Qu, Martha L. Slattery, Joshua D. Smith, Emily White, Goncalo R. Abecasis, Li Hsu, Deborah A. Nickerson, Ulrike Peters, on behalf of CCFR and GECCO. The contribution of rare and low-frequency variants to colorectal cancer heritability [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2017; 2017 Apr 1-5; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2017;77(13 Suppl):Abstract nr 1304. doi:10.1158/1538-7445.AM2017-1304

Improved imputation accuracy of rare and low-frequency variants using population-specific high-coverage WGS-based imputation reference panel

Genetic imputation is a cost-efficient way to improve the power and resolution of genome-wide association (GWA) studies. Current publicly accessible imputation reference panels accurately predict genotypes for common variants with minor allele frequency (MAF)≥5% and low-frequency variants (0.5≤MAF<5%) across diverse populations, but the imputation of rare variation (MAF<0.5%) is still rather limited. In the current study, we evaluate imputation accuracy achieved with reference panels from diverse populations with a population-specific high-coverage (30 ×) whole-genome sequencing (WGS) based reference panel, comprising of 2244 Estonian individuals (0.25% of adult Estonians). Although the Estonian-specific panel contains fewer haplotypes and variants, the imputation confidence and accuracy of imputed low-frequency and rare variants was significantly higher. The results indicate the utility of population-specific reference panels for human genetic studies.

Imputing rare variants in families using a two-stage approach.

BackgroundRecent focus on studying rare variants makes imputation accuracy of rare variants an important issue. Many approaches have been proposed to increase imputation accuracy among rare variants, from reference panel selection to combinations of existing methods to multistage analyses. We aimed to bring the strengths of these new approaches together with our proposed two-stage imputation for family data.MethodsOur imputation methods were tested on the region from 46.75Mb to 49.25Mb on chromosome 3. We did quality control based on the proportion of missing genotypes per variant and individual, leaving 495 individuals with 761 genome-wide association studies (GWAS) variants only, 45 with 14,077 sequence variants only, and 419 with both GWAS and sequencing data. All data were prephased using SHAPEIT2 with a duo hidden Markov model algorithm prior to performing imputation. Imputations were performed 100 times, each time masking the sequence data for 1 individual and imputing it from the GWAS data. We used well-imputed genotypes, defined as a probability of greater than 0.9, above 2 different minor allele frequency cutoffs—0.01 and 0.05—from Impute2 as input for Merlin, and compared these results to Impute2 and Merlin separately. The imputed results were evaluated using correlation measurement and the imputation quality score.ResultsOur method improved imputation accuracy, measured by imputation quality score, for variants with minor allele frequency between 0.01 and 0.40, but failed to improve accuracy for variants with minor allele frequency less than 0.01 when we used a minor allele frequency cutoff of 0.01 for the Impute2 results. In contrast, our 2-stage approach with a minor allele frequency cutoff of 0.05 performed the worst of all methods for variants with minor allele frequency between 0.01 and 0.40.ConclusionsThis method gave promising results, but may be further improved by changing the inclusion criteria of Impute2 variants. More analyses are needed on a larger region with different inclusion thresholds to assess the accuracy of this approach.

Improved imputation accuracy in Hispanic/Latino populations with larger and more diverse reference panels: applications in the Hispanic Community Health Study/Study of Latinos (HCHS/SOL).

Imputation is commonly used in genome-wide association studies to expand the set of genetic variants available for analysis. Larger and more diverse reference panels, such as the final Phase 3 of the 1000 Genomes Project, hold promise for improving imputation accuracy in genetically diverse populations such as Hispanics/Latinos in the USA. Here, we sought to empirically evaluate imputation accuracy when imputing to a 1000 Genomes Phase 3 versus a Phase 1 reference, using participants from the Hispanic Community Health Study/Study of Latinos. Our assessments included calculating the correlation between imputed and observed allelic dosage in a subset of samples genotyped on a supplemental array. We observed that the Phase 3 reference yielded higher accuracy at rare variants, but that the two reference panels were comparable at common variants. At a sample level, the Phase 3 reference improved imputation accuracy in Hispanic/Latino samples from the Caribbean more than for Mainland samples, which we attribute primarily to the additional reference panel samples available in Phase 3. We conclude that a 1000 Genomes Project Phase 3 reference panel can yield improved imputation accuracy compared with Phase 1, particularly for rare variants and for samples of certain genetic ancestry compositions. Our findings can inform imputation design for other genome-wide association studies of participants with diverse ancestries, especially as larger and more diverse reference panels continue to become available.

Missing value imputation for microRNA expression data by using a GO-based similarity measure.

BackgroundMissing values are commonly present in microarray data profiles. Instead of discarding genes or samples with incomplete expression level, missing values need to be properly imputed for accurate data analysis. The imputation methods can be roughly categorized as expression level-based and domain knowledge-based. The first type of methods only rely on expression data without the help of external data sources, while the second type incorporates available domain knowledge into expression data to improve imputation accuracy.In recent years, microRNA (miRNA) microarray has been largely developed and used for identifying miRNA biomarkers in complex human disease studies. Similar to mRNA profiles, miRNA expression profiles with missing values can be treated with the existing imputation methods. However, the domain knowledge-based methods are hard to be applied due to the lack of direct functional annotation for miRNAs. With the rapid accumulation of miRNA microarray data, it is increasingly needed to develop domain knowledge-based imputation algorithms specific to miRNA expression profiles to improve the quality of miRNA data analysis.ResultsWe connect miRNAs with domain knowledge of Gene Ontology (GO) via their target genes, and define miRNA functional similarity based on the semantic similarity of GO terms in GO graphs. A new measure combining miRNA functional similarity and expression similarity is used in the imputation of missing values. The new measure is tested on two miRNA microarray datasets from breast cancer research and achieves improved performance compared with the expression-based method on both datasets.ConclusionsThe experimental results demonstrate that the biological domain knowledge can benefit the estimation of missing values in miRNA profiles as well as mRNA profiles. Especially, functional similarity defined by GO terms annotated for the target genes of miRNAs can be useful complementary information for the expression-based method to improve the imputation accuracy of miRNA array data. Our method and data are available to the public upon request.