markabney / Kinpute

Kinpute

Introduction

Kinpute is a software package to improve genotype imputation by using identity by descent (IBD) information. In addition it includes a tool for selecting a subset of individuals from the study on whom to perform sequencing. These sequenced individuals then serve as an internal reference panel during the imputation.

Kinpute has the following features:

• Select an optimal set of study individuals to sequence as an internal reference panel.
• Works in conjunction with LD based imputation methods to deliver a final, highly accurate set of imputed genotype probabilities.
• Uses a novel method to extract more information from IBD than can be done with standard LD based approaches.
• Individuals can be closely or distantly (and cryptically) related.
• Does not require a pedigree for related individuals.
• Does not require phase information (and is thus robust to phasing errors).

Note that Kinpute is primarily intended to work in conjunction with standard LD-based imputation methods. That is, you would first perform imputation with a standard tool and reference panel (possibly including the internal reference panel) and use the resultant genotype probabilities as “prior” genotype probabilities to Kinpute. If, for some reason, it is not possible to run a standard LD-based imputation method, Kinpute can still be run, in which case allele frequencies are used to generate prior genotype probabilities. The efficacy of Kinpute is also closely tied to the quantity of IBD that exists in the study population. Areas of the genome in an individual that shares little or no IBD with members of the internal reference panel will show little change in the prior genotype probabilities. In the presence of substantial IBD, however, Kinpute can dramatically improve the quality of the imputed genotypes.

Installation

Kinpute is a self contained C++ software package. There are no dependencies other than having a C++11 compliant compiler. There are two source code files which compile to separate executables. To install simply compile using the commands shown below and place the resulting executables in a directory that is in your PATH.

Kinselect

Kinselect will evaluate a study sample and select a group of a specified size to sequence and serve as a reference panel. It is contained in the source file kinselect.cpp and can be compiled, for instance, with g++:

g++ -std=c++11 -O3 kinselect.cpp -o kinselect

Kinpute

Kinpute will do the actual genotype imputation and is contained in the source code file kinpute.cpp. To compile using g++, for instance:

g++ -std=c++11 -O3 kinpute.cpp -o kinpute

Usage

If you have a sample of individuals with either known or unknown relatedness (possibly distant), and need to select a subset of the sample to sequence, kinselect can help you choose the subset which should be maximally informative for imputation of sequence data to the rest of the sample. If you already have a portion of you study subjects that are sequenced, kinpute can be used to improve (or do) the imputation into the rest of the study sample.

Kinselect

Kinselect needs as input a list of IDs of the study subjects, the kinship coefficients between those subjects, and the number of individuals that will be sequenced. Note that the kinship coefficients could be obtained either from a pedigree or estimated from genotype data (as is done, for instance, when computing a genetic relatedness matrix). As output, kinselect provides two lists, one of the individuals that were selected to be sequenced, and the other is the remainder of the study subjects.

Command line arguments

-s samplefile

samplefile is the file that contains a list of the study sample. A subset of the listed study sample will be sequenced while the remainder will have their sequence data imputed. The file must have one ID per line. ID’s must not have any space characters.

-i kinshipfile

kishipfile is formatted so that each line has three, white-space separated fields. The first two fields are subject IDs and the third field is the kinship coefficient of that pair.

-n N

N is the number of subjects from the samplefile which should be selected to be sequenced.

-or outfile1

Output file outfile1 contains the list of IDs that will comprise the subjects who should be sequenced. That is, the internal reference panel.

-ou outfile2

Output file outfile2 has the list of subjects who were in samplefile but not in outfile1. These individuals will have their sequence data imputed.

-v

Verbose setting. This flag is optional and if present warnings for duplicate individuals in the samplefile and for missing kinship coefficients are printed to the error stream (typically the screen). By default these warnings are not printed.

Format notes

• IDs should uniquely identify individuals. If you have IDs that have both a family ID and an individual ID, we recommended combining them into a single ID of the form ”familyID,individualID”, i.e. a comma separating the two IDs. Be sure there is no white-space in the final ID.
• If you have a duplicate ID in the samplefile, kinselect will issue a warning only if you specify the -v flag on the command line. Otherwise, kinselect will silently ignore the duplicate.
• The kinshipfile may have kinship coefficients for subjects who are not in the samplefile. We do need a value, though, for every pair of individuals who are in the samplefile. If a pair is not found in the kinshipfile kinselect will silently assume the pair is unrelated (i.e. their kinship coefficient is zero). If the -v flag has been given on the command line, kinselect will print a warning that the pair was not found in the file. Note that a subject’s self-kinship coefficient (i.e. related to the inbreeding coefficient) is not needed and may be omitted from the file.

Example

In the example directory there are a sample kinship file and study sample file. If we want to select the best three people from the study sample file to sequence, we would run the following command:

kinselect -s study -i small.kin -n 3 -or small_ref_panel.txt -ou imp_panel.txt -v

The -v flag is optional and will print out warning messages about missing individuals. Note that in the file study are two individuals W and X that are not present in the kinship file, and the individual A appears twice. Kinselect issues warnings, but still selects three individuals who should be sequenced. Note that one of the individuals chosen to be sequenced is (assumed to be) unrelated to all the other subjects. This happens because the algorithm determines that after two of the related individuals are selected for sequencing there is more value to be gained by sequencing an unrelated individual than one of the related ones, even though the unrelated individual will not be informative for imputing the sequence data of any other subject (when imputation is based on IBD).

Kinpute

Kinpute will use IBD information to impute, or improve the imputation of, sequence data in a sample. To do this kinpute requires the sequence data of the individuals in the reference panel (this should be an “internal” reference panel of subjects who are related to the imputation sample, as opposed to one of the standard population reference panels), a file with probabilities of the genotypes of the sequence data in the imputation panel, and a file with IBD information between the subjects in the reference panel and the imputation panel. The input genotype probabilities normally is the output from another imputation program, such as IMPUTE2. This file is optional, and if not present kinpute will use the allele frequency in the internal reference panel to generate a genotype probability before using IBD to improve this estimate.

IBD estimates at every SNP in a framework set or markers can provide useful information for imputing genotypes. Clearly, the more IBD that exists between the reference panel and the imputation panel, the better kinpute will perform. In regions where there is no IBD, kinpute will output posterior genotype probabilities that are the same as the input (prior) genotype probabilities. Kinpute works best when multiple individuals in the reference panel are IBD with the region being imputed, particularly if the reference panel individuals also share some IBD at that location. IBD estimates come in the form of conditional probabilities of the nine condensed identity states at each SNP given the observed genotypes at a framework set of markers. These probabilities can be computed using the =IBDLD= software package.

Command line arguments

-ibd ibdfile

The ibdfile contains the probabilities of the nine condensed identity states for every SNP in the framework set for every pair of reference panel individuals and every imputation panel individual with each of the reference panel individuals.

-map mapfile

The map of the framework set of SNPs. This should be the same file as given to IBDLD.

-u samplefile

The samplefile has the list of individuals to be imputed. There must be one individual per line and in the format ”familyID,individualID”, where the familyID and individualID match those values in the ibdfile.

-r refpanelfile

The refpanelfile has the list of individuals who will be used as the reference panel for imputation. There must be only one individual per line and in the format ”familyID,individualID”, where the familyID and individualID match those values in the ibdfile.

-seq sequencefile

The sequencefile has the sequence data for every individual in the reference panel. The sequence data must be formatted as a tped file, see PLINK documentation.

-prior priorprobfile

This flag is optional. If it is not set kinpute will assume prior genotype probabilities based on the allele frequencies in the reference panel. If it is set the priorprobfile sets the prior genotype probabilities for every genotype that will be imputed. Normally, this will be the output from a standard imputation method that provides “soft calls”, i.e. probabilities for each of the three genotypes. The format is the same as the genotype file format used by IMPUTE2. This means that the imputation output from IMPUTE2 can be used directly as the priorprobfile.

-o outputfile

The outputfile has the posterior probabilities of the genotypes of the individuals in the samplefile. The format is identical to the format of the priorprobfile.

Format notes

• We recommend using =IBDLD= to create the ibdfile. The ibdfile format is the same as the output file format from IBDLD when using the ”-ibd 9 --ibdtxt” flags.
• In the ibdfile the familyID and individualID are separated by white space. These IDs must match the reformatted ID ”familyID,individualID” that appears in the refpanelfile and samplefile.
• The order of the individuals in the samplefile is assumed to be the order in the priorprobfile. It is also the order of the individuals in the outputfile.
• Kinpute also assumes the order of individuals in the refpanelfile is the order of the individuals in the sequencefile, i.e. the tped file.
• Kinpute does not need a tfam file, as is required by Plink, only the tped file.
• The markers in the sequencefile and priorprobfile should be ordered by physical position and no markers should have the same position as another marker.

Example

In the example directory are files that can be used to run an example imputation analysis. In this example we assume that IBDLD has already been run and you are doing imputation on a single chromosome. In general, kinpute should be run on different chromosome separately. We also assume that IMPUTE2 was run on the imputation sample resulting in the prior genotype probability file prior_gprob.txt. To perform imputation on the individuals in study_panel.txt you would use the following:

kinpute -ibd ibd.ibdtxt -map map.txt -u study_panel.txt -r ref_panel.txt \
-seq seq.tped -prior prior_gprob.txt -o imputed_genos

To perform imputation without having run IMPUTE2 (or some other imputation method) first, execute the same command but without the -prior flag:

kinpute -ibd ibd.ibdtxt -map map.txt -u study_panel.txt -r ref_panel.txt
-seq seq.tped  -o imputed_genos