Limits...
Adapt-Mix: learning local genetic correlation structure improves summary statistics-based analyses.

Park DS, Brown B, Eng C, Huntsman S, Hu D, Torgerson DG, Burchard EG, Zaitlen N - Bioinformatics (2015)

Bottom Line: This approach, especially in admixed populations, has the potential to produce misleading results, ignores variation in local structure and is not feasible when appropriate reference panels are missing or small.We applied Adapt-Mix to estimate the genetic correlation structure of both admixed and non-admixed individuals using simulated and real data.When using our method as opposed to the current standard of 'best guess' reference panels, we observed a 28% decrease in mean-squared error for imputation and a 73.7% decrease in mean-squared error for joint-testing.

View Article: PubMed Central - PubMed

Affiliation: Department of Bioengineering and Therapeutic Sciences, University of California San Francisco, San Francisco, Department of Computer Science, University of California Berkeley, Berkeley and Department of Medicine, University of California San Francisco, San Francisco, CA, USA.

Show MeSH
Histogram of the deviations from the true joint statistic when using a ‘best guess’ panel and Adapt-Mix to estimate Σ for joint-testing. (a) Joint testing for GALA II Mexicans. MXL deviations are shown in red and 1KG-Chrom is shown in blue. (b) Joint testing for GALA II Puerto Ricans. PUR deviations are shown in orange and 1KG-Chrom is shown in blue
© Copyright Policy - creative-commons
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4553832&req=5

btv230-F3: Histogram of the deviations from the true joint statistic when using a ‘best guess’ panel and Adapt-Mix to estimate Σ for joint-testing. (a) Joint testing for GALA II Mexicans. MXL deviations are shown in red and 1KG-Chrom is shown in blue. (b) Joint testing for GALA II Puerto Ricans. PUR deviations are shown in orange and 1KG-Chrom is shown in blue

Mentions: To show that the joint statistics produced by using our method for estimating correlations are unbiased (i.e. E[Jij − ] = 0), we looked at the mean difference between the true statistics and estimated statistics. Tables 3 and 4 show that the mean difference is closer to 0 when our approach is used in both the Mexicans and Puerto Ricans. The 1KG-Chrom-based correlation estimates generated differences in true versus estimated that were the closest to zero amongst all approaches. We can see from Tables 3 and 4 that 1KG-Chrom has the smallest variance for the differences in true versus estimated joint statistics. The ‘best guess’ panels had the highest variance of all approaches except for 1KG-Genome in the Puerto Ricans. Additionally, we examined all estimated joint statistics that were more than 2 chi-squared units from the truth. In Mexicans, we saw 122 such statistics for the MXL and 22 for 1KG-Chrom (Fig. 3a). A similar trend is seen in Puerto Ricans as well, with 53 large deviations for the PUR and 3 for 1KG-Chrom (Fig. 3b). The decrease in frequency and magnitude of large differences demonstrates that using Adapt-Mix can help reduce the number of false positives in a joint analysis using reference panels. However, high deviations seen in both methods indicate that regardless of approach there is potential to misestimate the pairwise correlation coefficients of SNPs.Fig. 3.


Adapt-Mix: learning local genetic correlation structure improves summary statistics-based analyses.

Park DS, Brown B, Eng C, Huntsman S, Hu D, Torgerson DG, Burchard EG, Zaitlen N - Bioinformatics (2015)

Histogram of the deviations from the true joint statistic when using a ‘best guess’ panel and Adapt-Mix to estimate Σ for joint-testing. (a) Joint testing for GALA II Mexicans. MXL deviations are shown in red and 1KG-Chrom is shown in blue. (b) Joint testing for GALA II Puerto Ricans. PUR deviations are shown in orange and 1KG-Chrom is shown in blue
© Copyright Policy - creative-commons
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4553832&req=5

btv230-F3: Histogram of the deviations from the true joint statistic when using a ‘best guess’ panel and Adapt-Mix to estimate Σ for joint-testing. (a) Joint testing for GALA II Mexicans. MXL deviations are shown in red and 1KG-Chrom is shown in blue. (b) Joint testing for GALA II Puerto Ricans. PUR deviations are shown in orange and 1KG-Chrom is shown in blue
Mentions: To show that the joint statistics produced by using our method for estimating correlations are unbiased (i.e. E[Jij − ] = 0), we looked at the mean difference between the true statistics and estimated statistics. Tables 3 and 4 show that the mean difference is closer to 0 when our approach is used in both the Mexicans and Puerto Ricans. The 1KG-Chrom-based correlation estimates generated differences in true versus estimated that were the closest to zero amongst all approaches. We can see from Tables 3 and 4 that 1KG-Chrom has the smallest variance for the differences in true versus estimated joint statistics. The ‘best guess’ panels had the highest variance of all approaches except for 1KG-Genome in the Puerto Ricans. Additionally, we examined all estimated joint statistics that were more than 2 chi-squared units from the truth. In Mexicans, we saw 122 such statistics for the MXL and 22 for 1KG-Chrom (Fig. 3a). A similar trend is seen in Puerto Ricans as well, with 53 large deviations for the PUR and 3 for 1KG-Chrom (Fig. 3b). The decrease in frequency and magnitude of large differences demonstrates that using Adapt-Mix can help reduce the number of false positives in a joint analysis using reference panels. However, high deviations seen in both methods indicate that regardless of approach there is potential to misestimate the pairwise correlation coefficients of SNPs.Fig. 3.

Bottom Line: This approach, especially in admixed populations, has the potential to produce misleading results, ignores variation in local structure and is not feasible when appropriate reference panels are missing or small.We applied Adapt-Mix to estimate the genetic correlation structure of both admixed and non-admixed individuals using simulated and real data.When using our method as opposed to the current standard of 'best guess' reference panels, we observed a 28% decrease in mean-squared error for imputation and a 73.7% decrease in mean-squared error for joint-testing.

View Article: PubMed Central - PubMed

Affiliation: Department of Bioengineering and Therapeutic Sciences, University of California San Francisco, San Francisco, Department of Computer Science, University of California Berkeley, Berkeley and Department of Medicine, University of California San Francisco, San Francisco, CA, USA.

Show MeSH