Accelerating imputation of missing genotypes using parallel computing.

Owing to massive jump in DNA technology, large-scale genomic datasets, including valuable information, have become available. While this is a prodigious opportunity, and it can also be a big challenge because analysing these large datasets with current computers and software tools is very difficult and may take days or even weeks to complete. Novel approaches such as parallel computing have been suggested to deal with these large datasets. Here, the effect of parallel computing on the performance of random forest (RF) algorithm for imputation of missing genotypes was studied. To this end, the genotypic matrices were simulated for, respectively, 500, 1000, 2000, and 3000 single-nucleotide polymorphism (SNP) for 500, 1000 and 2000 individuals, respectively. Then, 50% of genotypic information was masked and imputed by RF. The per cent of genotypes correctly imputed was used to measure accuracy of genotype imputation. Serial and parallel computing were applied to the data. In comparison to serial computing, parallel computing did not affect the accuracy of imputation, and the accuracy was the same in both scenarios. However, regarding computational time, parallel computing accelerated the analyses significantly in a way that it reduced the running time up to 63%. This was due to the fact that in the serial computing, only 10% of the processing power of the central processing unit (CPU) of the machine was used by the RF, while in the parallel computing, 55% of the processing power of the CPU was utilized. Therefore, as parallel computing significantly reduced the computing time and does not affect the accuracy of the results, this approach should be exploited by researchers to analyse large genomic datasets.

Relative contribution of Imprinting, X chromosome and Litter effects to phenotypic variation in economic traits of sheep.

Data on Zandi sheep were analysed to quantify maternal and paternal imprinting, X chromosome and litter effects' contribution to phenotypic variation in birth weight (BW), weaning weight (WW), growth rate (GR), Kleiber ratio (KR), efficiency of growth (EF) and relative growth rate (RGR). To this end, a two-step approach was adopted. In the first step, each trait was analysed with a series of 16 animal models, which were identical for fixed and autosomal additive genetic effects but differed for combinations of maternal permanent environmental, maternal genetic, X chromosome and litter effects. For each trait, the best model was selected by the Akaike information criterion (AIC) and likelihood ratio tests (LRTs). In the second step, three additional models were fitted by adding maternal imprinting, paternal imprinting or both (models 17, 18 and 19) to the best model selected in the first step. Estimators of bias, dispersion and accuracy of breeding values estimated within 19 models with whole, and partial data were used to evaluate how well were the 19 models in estimating breeding values for the animals when their records were masked. For all traits studied, fitting the litter effect led to a better data fit. Also, it resulted in noticeable decreases in residual variance and other maternal variances. For growth traits, models containing the X-linked effects fitted the data substantially better than corresponding models without the X-linked effects. For BW, WW and GR, estimates of X-linked heritability ( h s 2 ) ranged between 0.09 (GR) and 0.14 (BW). Ignoring X-linked effects from the genetic evaluation model resulted in significant inflated autosomal additive genetic variance. For BW, WW, EF and RGR, models containing the imprinting effects provided a better fit of the data than otherwise identical models. Imprinting effects contributed significantly to the phenotypic variation of these traits in a range between 5% (RGR) and 8% (BW, WW). A sharp decline was observed in autosomal additive genetic variance following including imprinting effects in the model (27% to 40% depending on the trait). The least bias and dispersion, as well as greater accuracies for breeding values of focal animals, were for a model which included imprinting, X-linked and litter effects. It was concluded that imprinting, X-linked and litter effects need to be included in the genetic evaluation models for growth and efficiency-related traits of Zandi lambs.

The relation between the genetic architecture of quantitative traits and long-term genetic response.

The genetic architecture of a quantitative trait refers to the number of genetic variants, allele frequencies, and effect sizes of variants that affect a trait and their mode of gene action. This study was conducted to investigate the effect of four shapes of allelic frequency distributions (constant, uniform, L-shaped and U-shaped) and different number of trait-affecting loci (50, 100, 200, 500) on allelic frequency changes, long term genetic response, and maintaining genetic variance. To this end, a population of 440 individuals composed of 40 males and 400 females as well as a genome of 200 cM consisting of two chromosomes and with a mutation rate of 2.5 × 10(-5) per locus was simulated. Selection of superior animals was done using best linear unbiased prediction (BLUP) with assumption of infinitesimal model. Selection intensity was constant over 30 generations of selection. The highest genetic progress obtained when the allelic frequency had L-shaped distribution and number of trait-affecting loci was high (500). Although quantitative genetic theories predict the extinction of genetic variance due to artificial selection in long time, our results showed that under L- and U-shapped allelic frequency distributions, the additive genetic variance is persistent after 30 generations of selection. Further, presence or absence of selection limit can be an indication of low (<50) or high (>100) number of trait-affecting loci, respectively. It was concluded that the genetic architecture of complex traits is an important subject which should be considered in studies concerning long-term response to selection.

Genetic analysis of growth rate and Kleiber ratio in Zandi sheep.

Genetic parameters for average daily gain from birth to weaning (ADGa), birth to 6 months (ADGb), weaning to 6 months (ADGc), weaning to yearling age (ADGd), and corresponding Kleiber ratios (KRa, KRb, KRc, and KRd) were estimated by using records of 3,533 Zandi lambs, descendent of 163 sires and 1265 dams, born between 1991 and 2005 at the Zandi Sheep Breeding Station at Khojir National Park, Tehran, Iran. A derivative-free algorithm combined with a series of six single-trait linear animal models was used to estimate phenotypic variance and its direct, maternal, and residual components. In addition, bivariate analyses were done to estimate (co)variance components between traits. Estimates of direct heritability (h(2)) were 0.11, 0.15, 0.09, 0.10, 0.10, 0.10, 0.06, and 0.07 for ADGa, ADGb, ADGc, ADGd, KRa, KRb, KRc, and KRd, respectively, thereby indicating the presence of low additive genetic variation for growth rate and Kleiber ratio in this population of Zandi sheep. Maternal genetic component was found to be significant on ADGa and KRa and contributed 3% and 5%, respectively, in total phenotypic variance of ADGa and KRa. A widespread range of genetic correlations among traits studied was observed. Except for negative genetic correlations between ADGa and KRc, ADGa and KRd, and between KRa and KRc, in other cases, genetic correlations were positive and moderate to very high. Phenotypic correlations ranged from -0.49 (ADGa/KRd) to 0.94 (ADGc/KRc). These results indicate that selecting for improved growth rate or Kleiber ratio in Zandi sheep would generate a relatively slow genetic progress.

Change in genetic size of small-closed populations: Lessons from a domestic mammal population.

The aim of this study was to monitor changes in genetic size of a small-closed population of Iranian Zandi sheep, by using pedigree information from animals born between 1991 and 2005. The genetic size was assessed by using measures based on the probability of identity-by-descend of genes (coancestry, f, and effective population size, N(e) ), as well as measures based on probability of gene origin (effective number of founders, f(e) , effective number of founder genomes, f(g) , and effective number of non-founder genomes, f(ne) ). Average coancestry, or the degree of genetic similarity of individuals, increased from 0.81% to 1.44% during the period 1993 to 2005, at the same time that N(e) decreased from 263 to 93. The observed trend for f(e) was irregular throughout the experiment in a way that f(e) was 68, 87, 77, 92, and 80 in 1993, 1996, 1999, 2002, and 2005, respectively. Simultaneously, f(g) , the most informative effective number, decreased from 61 to 35. The index of genetic diversity (GD) which was obtained from estimates of f(g) , decreased about 2% throughout the period studied. In addition, a noticeable reduction was observed in the estimates of f(ne) from 595 in 1993 to 61 in 2005. The higher than 1 ratio of f(e) to f(g) indicated the presence of bottlenecks and genetic drift in the development of this population of Zandi sheep. From 1993 to 1999, f(ne) was much higher than f(e) , thereby indicating that with respect to loss of genetic diversity, the unequal contribution of founders was more important than the random genetic drift in non-founder generations. Subsequently, random genetic drift in non-founder generations was the major reason for f(e) > f(ne) . The minimization of average coancestry in new reproductive individuals was recommended as a means of preserving the population against a further loss in genetic diversity.

Change in genetic size of small-closed populations: lessons from a domestic mammal population

The aim of this study was to monitor changes in genetic size of a small-closed population of Iranian Zandi sheep, by using pedigree information from animals born between 1991 and 2005. The genetic size was assessed by using measures based on the probability of identity-by-descend of genes (coancestry, f, and effective population size, Ne), as well as measures based on probability of gene origin (effective number of founders, f e, effective number of founder genomes, f g, and effective number of non-founder genomes, f ne). Average coancestry, or the degree of genetic similarity of individuals, increased from 0.81 percent to 1.44 percent during the period 1993 to 2005, at the same time that Ne decreased from 263 to 93. The observed trend for f e was irregular throughout the experiment in a way that f e was 68, 87, 77, 92, and 80 in 1993, 1996, 1999, 2002, and 2005, respectively. Simultaneously, f g, the most informative effective number, decreased from 61 to 35. The index of genetic diversity (GD) which was obtained from estimates of f g,decreased about 2 percent throughout the period studied. In addition, a noticeable reduction was observed in the estimates of f ne from 595 in 1993 to 61 in 2005. The higher than 1 ratio of f e to f g indicated the presence of bottlenecks and genetic drift in the development of this population of Zandi sheep. From 1993 to 1999, f ne was much higher than f e, thereby indicating that with respect to loss of genetic diversity, the unequal contribution of founders was more important than the random genetic drift in non-founder generations. Subsequently, random genetic drift in non-founder generations was the major reason for f e> f ne. The minimization of average coancestry in new reproductive individuals was recommended as a means of preserving the population against a further loss in genetic diversity.