ANGSD: Analysis of next generation Sequencing Data

Latest tar.gz version is (0.938/0.939 on github), see Change_log for changes, and download it here.

Genotype Likelihoods

From angsd
Revision as of 16:37, 7 December 2013 by Thorfinn (talk | contribs) (→‎Options)
Jump to navigation Jump to search

Many methods in ANGSD are based on genotype likelihoods, and ANGSD has 4 different genotype likelihood models implemented.

Genotype likelihoods and the four models are described in the Bottom.

<classdiagram> // [input|bam files;SOAP files{bg:orange}]->[sequence data]

[sequence data]->[genotype likelihoods|SAMtools;GATK;SOAPsnp;Kim et.al]
</classdiagram>

We also allow for dumping of the calculated genotype likelihoods in various formats that might be handy for some users.

Brief Overview

./angsd -GL 
	-> angsd version: 0.567	 build(Dec  7 2013 14:56:25)
	-> Analysis helpbox/synopsis information:
---------------------
analysisEstLikes.cpp:
	-GL=0: 
	1: SAMtools
	2: GATK
	3: SOAPsnp
	4: SYK
	-trim		0		(zero means no trimming)
	-tmpdir		angsd_tmpdir/	(used by SOAPsnp)
	-errors		(null)		(used by SYK)
	-minInd		0		(0 indicates no filtering)

Filedumping:
	-doGlf	0
	1: binary glf (10 log likes)	.glf.gz
	2: beagle likelihood file	.beagle.gz
	3: binary 3 times likelihood	.glf.gz
	4: text version (10 log likes)	.glf.gz

Options

-GL [int]

If your input is sequencing file you can estimate genotype likelhoods from the mapped reads. Four different methods are available.

  1. SAMtools model
  2. GATK model
  3. SOAPsnp model
  4. SYK model
-trim [0]

This will discards the ends of the reads when calculating the genotype likelihoods. This is for aDNA where we would expect to have increased error rates at the ends. You can use MapDamage2 to examine the damage pattern. Unless you are working with aDNA there is no reason to trim the ends.

-tmpdir angsd_tmpdir

SOAPsnp generates a mismatch matrix for each BAM file and based on this mismatch matrix it calculates the type specific errors for each position in the read. So for each BAM file it generates two files, to avoid cluttering up the working directory you can specify a folder that should be used. SOAPsnp assumes that all reads have the same length, if this is not the case this model might not be suited.

-errors

SYK model requires a file containing the type specific errors, as estimated from doError 1.

./angsd -bam bam.filelist -GL 1 -out outfile

GATK

-GL 2

options

-minQ [int]

default 13. The minimum allowed base quality score.

example

./angsd -bam bam.filelist -GL 2 -out outfile

soapSNP

-GL 3 When estimating GL with soapSNP we need to generate a calibration matrix. This is done automaticly if these doesn't exist. These are located in angsd_tmpdir/basenameNUM.count,angsd_tmpdir/basenameNUM.qual

options

-minQ [int]

default 13. The minimum allowed base quality score.

-tmpdir [int]

default angsd_tmpdir; The directory of the recalibration matrix.

example

./angsd -bam bam.filelist -GL 3 -out outfile -ref hg19.fa -minQ 0
#NB important to set -minQ to zero, ANGSD defaults to minQ 13

This first loop doesn't estimate anything else than the calibration matrix. So now we can do the analysis we want

./angsd -bam bam.filelist -GL 3 -out outfile -ref hg19.fa -doGlf 1

NB internally the max readlength is not allowed to exceed 256.

Kim et al.

-GL 4 Citation Citation

options

-error [filename]

A file with the estimated type specific error rates (see Error_estimation).

example

./angsd -bam bam.filelist -GL 4 -out outfile -error error.file 

Output genotype likelihoods

-doGlf [int]

Output the log genotype likelihoods to a file

0. don't dump anything (default)
1. binary all 10 llh
2. beagle text
3. beagle binary
4. textoutput of all 10 llhs.


Binary

Glf file in binary doubles. All 10 genotype likelihoods are printed to a file. For each printed site there are 10*N doubles where N is the number of individuals. The order of the 10 genotypes are alphabetical AA AC AG AT CC CG CT GG GT TT.

Beagle format

Beagle haplotype imputation and be performed directly on genotype likelhoods. To generate beagle input file use

-doGlf 2

In order to make this file the major and minor allele has the be inferred (-doMajorMinor). It is also a good idea to only use the polymorphic sites.


Example

In this example our input files are bam files. We use the samtools genotype likelihood methods. We use 10 threads. We infer the major and minor allele from the likelihoods and estimate the allele frequencies. We test for polymorphic sites and only outbut the ones with are likelhood ratio test statistic of minimum 24 (ca. p-value<1e-6).

./angsd -GL 1 -out genolike -nThreads 10 -doGlf 2 -doMajorMinor 1 -minLRT 24 -doMaf 2 -doSNP 2 -bam bam.filelist

output

The above command generates the file genolike.beagle.gz that can be use as input for the beagle software

marker  allele1 allele2 Ind0    Ind0    Ind0    Ind1    Ind1    Ind1    Ind2    Ind2    Ind2    Ind3    Ind3    Ind3 
1_14000023      1       0       0.941177        0.058822        0.000001        0.799685        0.199918        0.000397        0.666316        0.333155        0.000529 
1_14000072      2       3       0.709983        0.177493        0.112525        0.941178        0.058822        0.000000        0.665554        0.332774        0.001672
1_14000113      0       2       0.855993        0.106996        0.037010        0.333333        0.333333        0.333333        0.799971        0.199989        0.000040 
1_14000202      2       0       0.835380        0.104420        0.060201        0.799685        0.199918        0.000397        0.333333        0.333333        0.333333
...

Note that the above values sum to one per sites for each individuals. This is just a normalization of the genotype likelihoods in order to avoid underflow problems in the beagle software it does not mean that they are genotype probabilities.

simple text format

Theory

Genotype likelihoods are in this context the likelihood the data given a genotype. This is to be understood as we take all the information from our data for a specific position for a single individual, and we use this information to calculate the likelihood for our different genotypes. Since we assume diploid individuals it follows that we have 10 different genotypes.

0 1 2 3 4 5 6 7 8 9
AA AC AG AT CC CG CT GG GT TT

And we write the genotype likelihood as

GATK genotype likelihoods

In angsd we use the direct method of the first version of GATK (dragon). This is simply

where M is the sequencing depth is the observed base in read i, e is the probability of error calculated from the phredscaled qscore e.g.

SAMtools genotype likelihoods

This subsection with SAMtools gl are preliminary

Define:


SOAPsnp genotype likelihoods

SYK genotype likelihoods