= Imputation pipeline = [[TOC()]] There are at the moment two collaborating initiatives: * VU: Mathijs Kattenberg and Joukejan Hottenga using IMPUTE * UMCG: Lude Franke, Harm-Jan Westra, George Byelas, Morris Swertz using Beagle The objective is to bring these pipelines into the same space so they can be properly compared and optimized. TODO: describe the protocols here; == Description from Harm-Jan == The imputation pipeline has changed, in such a way that it was reduced to only a few steps. To facilitate QC and conversion steps, I've bundled our conversion tools in one single program called ImputationTool.jar. Here, I shortly describe the steps that need to be in the new pipeline, in placeholders I also describe what the commands could look like, if you would implement this in a shellscript (or java program). These examples can be the complete execution steps of the pipeline. '' Commands to run locally: '' === Step 1 === * According to Harm-Jan: if the dataset is in binary plink format, use plink —recode to convert back to ped+map * Transform .bed , .bim and .fam files into ASCII format {{{ /Users/alexandroskanterakis/Tools/plink/plink-1.07-mac-intel/plink --bfile /Users/alexandroskanterakis/Data/Finnish_cohort/CD_Finnuncorr.maf05 --ped /Users/alexandroskanterakis/Data/Finnish_cohort/CD_Finnuncorr.maf05.ped --map CD_Finnuncorr.maf05.map --recode }}} * Produces the files: {{{ -rw-r--r-- 1 alexandroskanterakis staff 11761120 Oct 5 12:11 plink.map -rw-r--r-- 1 alexandroskanterakis staff 4898208052 Oct 5 12:11 plink.ped }}} * real Execution time: 26m20.723s * creates rather big files 4.6G plink.ped === Step 2 === * Convert dataset to trityper format, if it is in ped+map format. {{{ java -Xmx4g -jar /Users/alexandroskanterakis/Tools/imputation/ImputationTool/dist/ImputationTool.jar --mode pmtt --in /Users/alexandroskanterakis/Data/Finnish_cohort/ --out /Users/alexandroskanterakis/Data/Finnish_cohort/ real 74m52.547s user 38m14.646s sys 5m25.472s }}} * Files Created: {{{ -rw-r--r-- 1 alexandroskanterakis staff 2449071024 Oct 5 15:05 GenotypeMatrix.dat -rw-r--r-- 1 alexandroskanterakis staff 46196 Oct 5 13:54 Individuals.txt -rw-r--r-- 1 alexandroskanterakis staff 98791 Oct 5 13:54 PhenotypeInformation.txt -rw-r--r-- 1 alexandroskanterakis staff 10771996 Oct 5 13:54 SNPMappings.txt -rw-r--r-- 1 alexandroskanterakis staff 5006368 Oct 5 13:54 SNPs.txt }}} === Step 3 === * compare the dataset to be imputed to the reference dataset (for example HapMap?2 release 24, also in TriTyper? format), and remove any snps for which the haplotypes are different, or do not correlate to the reference dataset. Also remove any SNP that is not in the reference. Save the output as Ped+Map {{{ java -Xmx4g -jar /Users/alexandroskanterakis/Tools/imputation/ImputationTool/dist/ImputationTool.jar --mode ttpmh --in /Users/alexandroskanterakis/Data/Finnish_cohort/ --hap /Users/alexandroskanterakis/Data/HapMap2-r24-CEU/ --out /Users/alexandroskanterakis/Data/Finnish_cohort/referenceOutput/ real 60m53.623s user 30m35.172s sys 2m34.325s }}} * Created Files (for each chromosome): {{{ -rw-r--r-- 1 alexandroskanterakis staff 221184 Oct 5 16:07 chr1.dat -rw-r--r-- 1 alexandroskanterakis staff 1004358 Oct 5 16:06 chr1.excludedsnps.txt -rw-r--r-- 1 alexandroskanterakis staff 442368 Oct 5 16:07 chr1.map -rw-r--r-- 1 alexandroskanterakis staff 441350 Oct 5 16:07 chr1.markersBeagleFormat -rw-r--r-- 1 alexandroskanterakis staff 5802372 Oct 5 16:07 chr1.ped -rw-r--r-- 1 alexandroskanterakis staff 117708 Oct 5 16:06 chr1.warningsnps.txt }}} === Steps 4-9 === 4. split the ped files in batches of 300 samples {{{ * mkdir -p ".$datasetLocation."/batches/ * split -a2 -l$batchSize $pedAndMapOutputLocation $batchOutputLocation }}} 5. run linkage2beagle to convert the ped and map files to beagle format {{{ for each batch do java -Xmx7g -jar linkage2beagle.jar data=$batchOutputLocation/chr$chromosome.dat pedigree=$batchOutputLocation/chr$chromosome.ped.$batch beagle=$beagleLocation/chr$chromosome.bgl.$batch done }}} '' Commands to run in server: '' 6. run the actual imputation on the batches on the cluster (needs hapmap to be recoded to beagle format as well, but I have these files for you) {{{ for each batch do java -Xmx11g -Djava.io.tmpdir=\$TMPDIR -jar beagle.jar unphased=$beagleLocation/chr$chromosome.bgl.$batch phased=$referenceLocation/HM2_Chr$chromosome-BEAGLE markers=$referenceLocation/markers_Chr$chromosome.txt missing=0 out=$outputLocation/Chr$chromosome/chr$chromosome-$batch done }}} '' Commands to run locally: '' 7. convert the beagle imputed files into trityper format {{{ java -Xmx4g -jar ImputationTool.jar bttb $outputLocation Chr/ChrCHROMOSOME-BATCH $imputedTriTyperLocation $numSamples }}} 8. correlate the imputed snps to the snps in the original dataset {{{ java -Xmx4g -jar ImputationTool.jar corr $trityperOutputLocation $datasetName $imputedTriTyperLocation $imputedDatasetName }}} 9. (if needed) convert to other formats (plink dosage / ped+map)) That's basically it. A lot simpler than the previous version, don't you think? The required tool is attached to this e-mail, but might still be a bit buggish. Any recommendations are therefore more than welcome. == IMPUTE pipeline == TODO: paste shell script descriptions of each step. == BEABLE pipeline == TODO: paste shell script descriptions of each step == Discussion == === Mixing platforms may influence imputation results === We into some troubles, resulting in our test statistic being highly inflated (which is indicative of false positive results). We thought of some possible causes which might explain this effect, although we should still test them: * '''SNPs with bad imputation quality''': we should remove SNPs with an R2 value < 0.90 prior to GWAS. These values are stored alongside the beagle imputation output. Taking a more stringent cutoff seemed to decrease the inflation, although you lose half of the SNPs. * '''batch effects caused by overrepresentation of a certain haplotype within an imputation batch''': for each batch of samples, beagle estimates a best fitting model to predict the genotypes of the missing SNPs, which is dependent upon both the input data as the reference dataset. Cases and controls should be therefore randomly distributed across the batches. Another option is to use impute, rather than beagle, since its batches are across parts of the genome, instead of samples. * '''difference in source platform''': different platforms have different SNP content. When you impute datasets coming from different platforms, the resulting model which is based on the input data is also different. When associating traits in a GWAS meta-analysis, these differences may account for a platform specific effect. We should therefore remove the SNPs which are non-overlapping between such platforms, prior to imputation, and impute the samples after combining the datasets. This would remove such a platform-bias, although would also cause a huge loss of available SNPs, when the overlap between platforms is small. However, in my opinion, this problem is similar to the batch effect problem, and can possibly be resolved by randomizing the sample content of the batches: the model will then possibly be fitted to the data that is available. In any case the datasets that are used in a meta-analysis should be imputed together.