Inbred Based Populations"

In onemap: Construction of Genetic Maps in Experimental Crosses

knitr::opts_chunk$set(collapse = TRUE,
                      comment = "#>",
                      fig.width = 6,
                      fig.height = 6,
                      fig.align = "center",
                      dev = "png",
                      dpi = 36,
                      cache = TRUE)

Starting in version 2.0-0, OneMap can also deal with inbred-based populations, that is, populations that have homozygous parental lines in the genealogy (F2s, backcrosses, and RILs). As a consequence, linkage phases do not need to be estimated. Since version 2.3.0, phases are estimated for F2 populations to properly generate progeny haplotypes not only the recombination fraction.

In this vignette, we explain how to proceed with the analysis in an F2 population. The same procedure can be used for backcrosses and RILs as well, and therefore users should not have any difficulty in analyzing their data. However, there are a number of differences from genetic mapping in outcrossing species; please read the proper vignette.

If you are not familiar with R, we recommend first the reading of vignette Introduction to R. You do not need to be an expert in R to build your linkage map, but some concepts are necessary and will help you through the process.

There is a GitHub OneMap version which is constantly improved, we strongly recommend all users to try this version. In augusto-garcia/onemap GitHub page you can find instructions to install the package from GitHub and also more fancy tutorials.

Creating the data file

For F2s, backcrosses, and RILs, two input formats are accepted. The user can choose between the standard OneMap file format or the same raw file used by MAPMAKER/EXP (Lander et al., 1987). Therefore, one should have no difficulty in using data sets already available for MAPMAKER/EXP when deciding to try OneMap.

Both types of raw files can contain phenotypic information, but this will not be used during map construction, that requires only genotypic information (made available by molecular markers).

Creating MAPMAKER/EXP data file

The MAPMAKER/EXP raw file, combined with the map file produced by OneMap, can be readily used for QTL mapping using R/qtl (Broman et al., 2008) or QTL Cartographer (Wang et al., 2010), among others.

Here, we briefly present how to set up this data file. For more detailed information see the MAPMAKER/EXP manual (Lincon et al., 1993), available here.

The first line of your data file should be:

data type xxxx

where xxxx is one of the following data types:

| xxxx | Population type | |-----------------|-----------------------------| | f2 backcross | Backcross | | f2 intercross | F2 | | ri self | RIL, produced by selfing | | ri sib | RIL, produced by sib mating |

The second line should contain the number of individuals in the progeny, the number of markers, and the number of quantitative traits. So, for example, a valid line would be

10 5 2

for a data set with 10 individuals (yes, very small, but this is just an example), 5 markers, and 2 traits evaluated.

Then, the genotype information is included for each marker. The character * indicates the beginning of information of a marker, followed by the marker name. For instance, here is an example of such a file for an F2 population with 10 individuals, 5 markers, and 2 quantitative traits:

data type f2 intercross
10 5 2

*M1 A B H H A - B A A B
*M2 C - C C C - - C C A
*M3 D B D D - - B D D B
*M4 C C C - A C C A A C
*M5 C C C C C C C C C C

*weight 10.2 - 9.4 11.3 11.9 8.9 - 11.2 7.8 8.1 
*length 1.7 2.1 - 1.8 2.0 1.0 - 1.7 1.0 1.1

The codification for genotypes is the following:

| Code | Meaning | |-------|------------------------------------------------| | A | homozygous for allele A (from parent 1 - AA) | | B | homozygous for allele B (from parent 2 - BB) | | H | heterozygous carrying both alleles (AB) | | C | Not homozygous for allele A (Not AA) | | D | Not homozygous for allele B (Not BB) | | - | Missing data for the individual at this marker |

The symbols option (not included in this example), used in MAPMAKER/EXP files, is also accepted (please, see its manual for details).

The quantitative trait data should come after the genotypic data and has a similar format, except the trait values for each individual must be separated by at least one space, a tab, or a line break. A dash (-) indicates missing data.

This file must be saved in plain text format using a simple text editor such as notepad on Microsoft Windows. Historically, MAPMAKER/EXP uses the .raw extension for this file; however, you can use any other extensions, such as .txt.

If you want to see more examples about this file type, open mapmaker_example_bc.raw and mapmaker_example_f2.raw, both available with OneMap and saved in the directory extdata on your computer, in the folder where you installed OneMap (use system.file(package="onemap") to see where it is located on your computer).

Now, let us load OneMap:

library(onemap)

To save your project anytime, type:

save.image("C:/.../yourfile.RData")

if you are using Windows, otherwise, adapt the code. Notice that you need to specify where to save and the name of the file. You can also use the toolbar, of course.

Creating OneMap data file

The OneMap data file has few differences compared to MAPMAKER/EXP format. As MAPMAKER/EXP format, the input OneMap file is a text file, where the first line indicates the cross-type and the second line provides information about the number of individuals, the number of markers, and the number of quantitative traits. Here, the format also supports keeping physical markers location information. The followed numbers indicate the presence/absence (1/0) of chromosome and position information and the presence/absence(1/0) of phenotypic data.

The third line contains sample IDs. Then, the genotype information is included separately for each marker. The character * indicates the beginning of information input for a new marker, followed by the marker name. Next, there is a code indicating the marker type according to:

| Code | Type | |-----------|-------------------------------| | A.H.B | Codominant marker | | C.A | Dominant marker for allele B | | D.B | Dominant marker for allele A | | A.H | Marker for backcross | | A.B | Marker for ril self/sib cross |

Finally, after each marker type, comes the genotype data for the segregating population. Missing data are indicated with the character - (minus sign) and an empty space separates the information for each individual. Positions and phenotype information, if present, follows genotypic data with a similar structure. Details are found with the help of function read_onemap.

Here is an example of such file for 10 individuals, 5 markers (the two zeros in the second line indicate that there is no chromosome information, physical position information), and two phenotypic data, respectively). It is very similar to a MAPMAKER/EXP file, but has additional information about the cross_type.

data type f2 intercross
10 5 0 0 2
I1 I2 I3 I4 I5 I6 I7 I8 I9 I10
*M1 A.H.B  ab a - ab b a ab - ab b
*M2 A.H.B  a - ab ab - b a - a ab
*M3 C.A    c a a c c - a c a c
*M4 A.H.B  ab b - ab a b ab b - a
*M5 D.B    b b d - b d b b b d
*fen1 10.3 11.2 11.1 - 9.8 8.9 11.0 10.7 - 10.1
*fen2 42 49 - 45 51 42 28 32 38 40

In case you have physical chromosome and position information:

data type f2 intercross
10 5 1 1 2
I1 I2 I3 I4 I5 I6 I7 I8 I9 I10
*M1 A.H.B  ab a - ab b a ab - ab b
*M2 A.H.B  a - ab ab - b a - a ab
*M3 C.A    c a a c c - a c a c
*M4 A.H.B  ab b - ab a b ab b - a
*M5 D.B    b b d - b d b b b d
*CHROM    1 1 1 2 2
*POS    2391 3812 5281 1823 3848
*fen1 10.3 11.2 11.1 - 9.8 8.9 11.0 10.7 - 10.1
*fen2 42 49 - 45 51 42 28 32 38 40

The input file must be saved in text format, with extensions like .raw. It is a good idea to open the text files called onemap_example_f2.raw, onemap_example_bc, onemap_example_riself (available in extdata with OneMap and saved in the directory you installed it) to see how this file should be. You can see where OneMap is installed using the command system.file(package="onemap").

Importing data

From MAPMAKER/EXP file

Once you created your data file with raw data, you can use OneMap function read_mapmaker to import it to OneMap:

mapmaker_example_f2 <- read_mapmaker(dir="C:/workingdirectory", 
                                     file="your_data_file.raw")

The first argument is the directory where the input file is located, modify it accordingly. The second one is the data file name.

In this example, an object named mapmaker_example_f2.raw was created. Notice that if you leave the argument dir blank, the file will be read from your current working directory. To set a working directory, see Introduction to R (Importing and Exporting Data).

mapmaker_example_f2 <- read_mapmaker(file= system.file("extdata/mapmaker_example_f2.raw", 
                                                       package = "onemap"))

For this example, we will use a simulated data set from an F2 population which is distributed along with OneMap. Because this particular data set is distributed along with the package, you can load it by typing:

data("mapmaker_example_f2")

To see what this data set is about, type:

mapmaker_example_f2

As you can see, the data consists of a sample of 200 individuals genotyped for 66 markers (36 co-dominant (AA, AB or BB), 15 dominant in one parent (Not AA or AA) and 15 dominant in the other parent (Not BB or BB) with 15% of missing data. You can also see that there is phenotypic information for one trait in the data set, that can be used for QTL mapping.

From OneMap raw file

The same procedure is made for the OneMap raw file, but, instead of using the function read_mapmaker we use read_onemap to read the OneMap format.

onemap_example_f2 <- read_onemap(dir="C:/workingdirectory", 
                                 inputfile = "your_data_file.raw")

In this example, an object named onemap_example_f2.raw was created. The data set containing the same markers and individuals of the mapmaker_example_f2.raw file. Would be a good idea to open these two files in a text editor and compare them to better understand the differences between the two kinds of input files. We can read the onemap_example_f2.raw using:

onemap_example_f2 <- read_onemap(inputfile= system.file("extdata/onemap_example_f2.raw", 
                                                        package = "onemap"))

Or, because this particular data are available together with the OneMap package:

data("onemap_example_f2")

To see what this data set is about, type:

onemap_example_f2

As you can see, the mean difference in the output object is that the read_onemap function keeps chromosome and position information. Because the objects mapmaker_example_f2 and onemap_example_f2 are practically the same, from now we will use only onemap_example_f2.

Importing data from the VCF file

If you are working with biallelic markers, as SNPs and indels (only codominant markers A.H.B), in VCF (Variant Call Format) files, you can import information from VCF to OneMap using onemap_read_vcfR function.

With the onemap_read_vcfR you can convert VCF file directly to onemap. The onemap_read_vcfR function keeps chromosome and position information for each marker at the end of the raw file.

We will use the same example file vcf_example_f2.vcf to show how it works.

Here we use the the vcfR package internally to help this conversion. The vcfR authors mentioned in their tutorials that RAM memory use is an important consideration when using the package. Depending of your dataset, the object created can be huge and occupy a lot of memory.

You can use onemap_read_vcfR function to convert the VCF file to onemap object. The parameters used are the vcf with the VCF file path, the identification of each parent (here, you must define only one sample for each parent) and the cross type.

vcf_example_f2 <- onemap_read_vcfR(vcf = system.file("extdata/vcf_example_f2.vcf.gz", package = "onemap"),
                                   parent1 = "P1", 
                                   parent2 = "P2", 
                                   cross = "f2 intercross")

Depending on your dataset, this function can take some time to run.

NOTE: From version 2.0.6 to 2.1.1005, OneMap had the vcf2raw function to convert vcf to .raw. Now, this function is defunct, but it can be replaced by a combination of onemap_read_vcfR and write_onemap_raw functions. See Exporting .raw file from onemap object session to further information about write_onemap_raw.

Visualization of raw data

Before building your linkage map, you should take a look at your data set. First, notice that by reading the raw data into OneMap, an object of classes onemap and f2 was produced:

data(vcf_example_f2)

class(onemap_example_f2)
class(vcf_example_f2)

In fact, functions read_mapmaker and read_onemap will produce objects of classes backcross, riself, risib or f2, according to the information in the data file for inbred-based populations. Therefore, you can use OneMap's version of function plot to produce a graphic with information about the raw data. It will automatically recognize the class of the object and produce the graphic. To see it in action, try:

plot(onemap_example_f2)
plot(vcf_example_f2)

The graphic is self-explanatory. If you want to save it, see the help for function plot.onemap:

?plot.onemap

This graphic shows that missing data is somehow randomly distributed; also, the proportion of dominant markers is relatively high for this data set. In OneMap's notation, codominant markers are classified as of B type; dominant ones, by C type (for details about this notation, see the vignette for outcrossing species). You can see the number of loci within each type using function plot_by_segreg_type:

plot_by_segreg_type(onemap_example_f2)
plot_by_segreg_type(vcf_example_f2)

So, as shown before, the object onemap_example_f2 has 36 codominant markers and 30 dominant ones and the vcf_example_f2has only codominant markers.

Graphical view of genotypes and allele depths (new!)

Function create_depth_profile generates dispersion graphics with x and y axis representing, respectively, the reference and alternative allele depths. The function is only available for biallelic markers in VCF files with allele counts information. Each dot represents a genotype for mks markers and inds individuals. If both arguments receive NULL, all markers and individuals are considered. Dots are colored according to the genotypes present in the OneMap object (GTfrom = onemap) or in the VCF file (GTfrom = vcf). An rds file is generated with the data in the graphic (rds.file). The alpha argument controls the transparency of the color of each dot. Control this parameter is a good idea when having a big amount of markers and individuals. The x_lim and y_lim control the axis scale limits, by default, it uses the maximum value of the counts.

Here is an example of the simulated dataset.

simu_f2_obj <- onemap_read_vcfR(vcf = system.file("extdata/vcf_example_f2.vcf.gz", package="onemap"), 
                                cross = "f2 intercross", 
                                parent1 = "P1", parent2 = "P2")
create_depths_profile(onemap.obj = simu_f2_obj,
                      vcfR.object =  system.file("extdata/vcf_example_f2.vcf.gz", package="onemap"), 
                      parent1 = "P1", 
                      parent2 = "P2", 
                      vcf.par = "AD", 
                      recovering = FALSE, 
                      mks = NULL, 
                      inds = NULL, 
                      GTfrom = "vcf", 
                      alpha = 0.1,
                      rds.file = "depths_f2.rds")

Selecting genotypes from vcf (GTfrom = "vcf") the colors are separated by VCF genotypes: 0/0 homozygote for the reference allele, 0/1 heterozygote, and 1/1 homozygote for the alternative allele. Depending on your VCF, you can also have phased genotypes, which are presented by pipe (|) instead of the bar (/).

Combining OneMap objects

If you have more than one dataset of markers, all from the same cross-type, you can use the function combine_onemap to merge them into only one onemap object.

In our example, we have two onemap objects:

• first: onemap_example_f2 (equivalent to mapmaker_example_f2) with 66 markers and 200 individuals
• second: vcf_example_f2 with 25 biallelic markers and 192 individuals.

The combine_function recognizes the correspondent individuals by the ID, thus, it is important to define exactly the same IDs to respective individuals in both raw files. Compared with the first file, the second file does not have markers information for 8 individuals. The combine_onemap will complete that information with NA.

In our examples, we have only genotypic information, but the function can also merge the phenotypic information if it exists.

comb_example <- combine_onemap(onemap_example_f2, vcf_example_f2)
comb_example

The function arguments are the names of the OneMap objects you want to combine.

Plotting markers genotypes from the outputted OneMap object, we can see that there are more missing data - (black vertical lines) for some individuals because they were missing in the second file.

plot(comb_example)

Find redundant markers

It is possible that there are redundant markers in your dataset, especially when dealing with too many markers. Redundant markers have the same genotypic information that others markers, usually because didn't happen recombination events between each other. They will not increase information on the map but will increase computational effort during the map building. Therefore, it is a good practice to remove them to build the map and, once the map is already built, they can be added again.

First, we use the function find_bins to group the markers into bins according to their genotypic information. In other words, markers with the same genotypic information will be in the same bin.

bins <- find_bins(comb_example, exact = FALSE)
bins

The first argument is the OneMap object and the exact argument specifies if only markers with exact same information will be at the same bin. Using FALSE at this second argument, missing data will not be considered and the marker with the lowest amount of missing data will be the representative marker on the bin.

Our example dataset has only two redundant markers. We can create a new OneMap object without them, using the create_data_bins function. This function keeps only the most representative marker of each bin from the bins object.

bins_example <- create_data_bins(comb_example, bins)
bins_example

The arguments for create_data_bins function are the OneMap object and the object created by find_bins function.

Exporting .raw file from OneMap object

The functions onemap_read_vcfR generates new OneMap objects without use a input .raw file. Also, the function combine_onemap manipulates the information of the original .raw file and creates a new data set. In both cases, you do not have an input file .raw that contains the same information as your current onemap object If you want to create a new input file with the data set you are working on after using these functions, you can use the function write_onemap_raw.

write_onemap_raw(bins_example, file.name = "new_dataset.raw", cross="f2 intercross")

The file new_dataset.raw will be generated in your working directory. In our example, it contains markers from onemap_example_f2 and vcf_example_f2 data sets.

Segregation tests

Now, it should be interesting to see if markers are segregating following what is expected by Mendel's law. You first need to use function test_segregation using as argument an object of class onemap.

f2_test <- test_segregation(bins_example)

This will produce an object of class onemap_segreg_test:

class(f2_test)

You cannot see the results if you simply type the object name; use OneMap's version of the print function for objects of class onemap_segreg_test:

f2_test

(Nothing is shown!)

print(f2_test)

This shows the results of the Chi-square test for the expected Mendelian segregation pattern of each marker locus. This depends of course on the marker type, because codominant markers can show heterozygous genotypes. The appropriate null hypothesis is selected by the function. The proportion of individuals genotyped is also shown.

To declare statistical significance, remember that you should consider that multiple tests are being performed. To guide you in the analysis, function Bonferroni_alpha shows the alpha value that should be considered for this number of loci if applying Bonferroni's correction with global alpha of 0.05:

Bonferroni_alpha(f2_test)

You can subset object f2_test to see which markers are distorted under Bonferroni's criterion, but it is easier to see the proportion of markers that are distorted by drawing a graphic using OneMap's version of the function plot for objects of class onemap_segreg_test:

plot(f2_test)

The graphic is self-explanatory: p-values were transformed by using -log10(p-values) for better visualization. A vertical line shows the threshold for tests if Bonferroni's correction is applied. Significant and non-significant tests are identified. In this particular example, no test was statistically significant, so none will be discarded.

Please, remember that Bonferroni's correction is conservative, and also that discarding marker data might not be a good approach to your analysis. This graphic is just to suggest a criterion, so use it with caution.

You can see a list of markers with non-distorted segregation using function select_segreg:

select_segreg(f2_test)

To get a list of distorted ones (none in this example):

select_segreg(f2_test, distorted = TRUE)

It is not recommended, but you can define a different threshold value by changing the threshold argument of the function select_segreg.

For the next steps will be useful to know the numbers of each marker with segregation distortion, so then you can keep those out of your map building analysis. These numbers refer to the lines where markers are located on the data file.

To access the corresponding number for of these markers you can change the numbers argument:

no_dist <- select_segreg(f2_test, distorted = FALSE, numbers = TRUE) #to show the markers numbers without segregation distortion
no_dist

dist <- select_segreg(f2_test, distorted = TRUE, numbers = TRUE) #to show the markers numbers with segregation distortion
dist

Estimating two-point recombination fractions

After visualizing raw data and checking for segregation distortion, let us now estimate recombination fractions between all pairs of markers (two-point tests). This is necessary to allow us to test which markers are linked. At this point, you should pay no attention if markers show segregation distortion or not, that is, simply use all of them.

twopts_f2 <- rf_2pts(input.obj = bins_example)

There are two optional arguments in function rf_2pts: LOD and max.rf which indicate the minimum LOD Score and the maximum recombination fraction to declare linkage (they default to 3.0 and 0.5, respectively).

The default for the recombination fraction is easy to understand, because if max.rf < 0.5 we could state that markers are linked. The LOD Score is the statistic used to evaluate the significance of the test for max.rf = 0.50. This needs to take into consideration the number of tests performed, which of course depends on the number of markers. Function suggest_lod can help users to find an initial value to use for their linkage test. For this example:

(LOD_sug <- suggest_lod(bins_example))

Thus, one should consider using LOD = 4.145858 for the tests. Please, notice that this is just a guide, not a value to take without any further consideration. For now, we will keep the default values, but later will show that results do not change in our example by using LOD = 3 or LOD = 4.145858.

If you want to see the results for a single pair of markers, say M12 and M42, use:

print(twopts_f2, c("M12", "M42"))

Since version 2.3.0, we estimate phases for F2 populations too, so here you can see the recombination fractions and LOD values for each possible phase. For RILs and backcross, you will obtain only one value.

This was possible because OneMap has a version of the print function that can be applied to objects of class rf_2pts:

class(twopts_f2)

However, objects of this type are too complex to print if you do not specify a pair of markers:

print(twopts_f2)

Strategies for this tutorial example

In this example we follow two different strategies:

• Using only recombinations information.

• Using the recombinations and also the reference genome information, once our example has CHROM and POS information for some of the markers.

First, we will apply the strategy using only recombinations information. In the second part of this tutorial, we show a way to use also reference genome information.

Using only recombinations information

Assigning markers to linkage groups

To assign markers to linkage groups, first, use the function make_seq to create a (un-ordered) sequence with all markers:

mark_all_f2 <- make_seq(twopts_f2, "all")

Function make_seq is used to create sequences from objects of several different classes. Here, the first argument is of class rf_2pts and the second argument specifies which markers one wants to use ("all" indicates that all markers will be analyzed). The object mark_all_f2 is of class sequence:

class(mark_all_f2)

If you want to form groups with a subset of markers, say M1, M3 and M7, use:

mrk_subset <- make_seq(twopts_f2, c(1, 3, 7))

In this case, it was easy because marker names and order in the objects (indicated in vector c(1, 3, 7)) are closely related, that is, you can easily know the position of markers in the object once you know their names. However, this is not true for real data sets, where markers do not have simple names such as M1 or M2.

A good example is to use the vector of markers without segregation distortion that we selected when applying the Chi-square tests.

mark_no_dist_f2 <- make_seq(twopts_f2, no_dist)

In our example, there are no markers with segregation distortion, then the object mark_no_dist_f2 is equivalent to mark_all_f2.

Forming the groups

OneMap has two functions to perform the markers grouping. The first presented here is the group function:

LGs_f2 <- group(mark_all_f2)
LGs_f2

This will show the linkage groups which are formed with criteria defined by max.rf and LOD. These criteria are applied as thresholds when assigning markers to linkage groups. If not modified, the same values used for the object twopts (from two-point analysis) will be maintained (so, LOD = 3.0 and max.rf = 0.5 in this example).

Users can easily change the default values. For example, using LOD suggested by suggest_lod (rounded up):

(LGs_f2 <- group(mark_all_f2, LOD = LOD_sug, max.rf = 0.5))

In our case, nothing different happens. (The parentheses above are just to avoid typing LGs_f2 in a new row to have the object printed).

We can see that the markers were assigned to only one linkage group. This division isn't so trustful, mostly if you are also working with dominant markers.

Another option is the group_upgma function. It is an adapted version of MAPpoly grouping function.

LGs_upgma <- group_upgma(mark_all_f2, expected.groups = 5, inter = F)
plot(LGs_upgma)

You can define the expected number of groups in the expected.groups argument and check how the markers are split in the plotted dendrogram. Using argument inter=TRUE you can change interactively the number of groups defined by the red squares in the graphic.

If you have reference genome information for some of your markers, you can separate the groups using it and the group_seq function (further information in Using the recombinations and the reference genome information) to add the markers without reference genome information. We can also confirm the separation of linkage groups in the ordering procedure.

Notice the class of object LGs_f2 and LGs_upgma:

class(LGs_f2)
class(LGs_upgma)

Ordering markers within linkage groups

After assigning markers to linkage groups, the next step is to order the markers within each group.

First, let us choose the mapping function used to display the genetic map. We can choose between Kosambi or Haldane mapping functions. To use Haldane, type

set_map_fun(type = "haldane")

To use Kosambi's function:

set_map_fun(type = "kosambi")

If none is set, kosambi function is applied.

Linkage group 1

First, you must extract it from the object of class group. Let us extract the group 1 using function make_seq:

LG1_f2 <- make_seq(LGs_f2, 1)

The first argument is an object of class group and the second is a number indicating which linkage group will be extracted. In this case, the object LGs_f2, generated by function group, is of class group. In fact, this function can handle different classes of objects.

If you type:

LG1_f2

you will see which markers are comprised in the sequence. But notice that no parameters have been estimated so far (the function says Parameters not estimated). This refers to the fact that so far we only attributed markers to linkage groups, but we did not perform any analysis for them as a group - only as pairs. (Does it seem complicated? Do not worry, you will understand the details in a moment).

Notice the class of object LG1_f2:

class(LG1_f2)

To order markers in this group, you can use a two-point based algorithm such as Seriation (Buetow and Chakravarti, 1987), Rapid Chain Delineation (Doerge, 1996), Recombination Counting and Ordering (Van Os et al., 2005) and Unidirectional Growth (Tan and Fu, 2006):

LG1_rcd_f2 <- rcd(LG1_f2, hmm = FALSE)
LG1_rec_f2 <- record(LG1_f2, hmm = FALSE)
LG1_ug_f2 <- ug(LG1_f2, hmm = FALSE)

Argument hmm defines if the function should run the HMM chain multipoint approach to estimate the genetic distances given the marker order provided by the two-points ordering algorithm. We set here the argument hmm=FALSE because we just want to obtain the marker order. We are not yet estimating the genetic distances. We suggest to use hmm=TRUE only when you already decided which order is the best because the HMM chain is the most computationally intensive step in the map building (mainly if F2 intercross population). You can use rf_graph_table to check the ordering quality (see details below) and make editions in the marker order using drop_marker. After, you can use map or map_avoid_unlinked functions to estimate the genetic distances (check session Map estimation for an arbitrary order).

The algorithms provided different results (results not printed in this vignette). For an evaluation and comparison of these methods, see Mollinari et al. (2009).

In this particular case, seriation will return an expected error:

LG1_ser_f2 <- seriation(LG1_f2, hmm = F) # Will return an error (can not be used in this case)

Another method that has been demonstrated to be very efficient in ordering markers uses the multidimensional scaling (MDS) approach. OneMap has the mds_onemap function that makes an interface with the MDSMap package to apply this approach to order markers. This is particularly useful if you are dealing with many markers (up to 20). The method also provides diagnostics graphics and parameters to find outliers to help users to filter the dataset. You can find more information in MDSMap vignette. Our mds_onemap does not present all possibilities of analysis that the MDSMap package presents. See help page ?mds_onemap to check which ones we implemented.

LG1_mds_f2 <- mds_onemap(input.seq = LG1_f2, hmm = F)

If you only specify the input sequence, mds_onemap will use the default parameters. It will also generate the MDSMap input file in out.file file. You can use out.file in the MDSMap package to try other parameters too. The default method used is the principal curves, know more about using ?mds_onemap and reading the MDSMap vignette.

Besides these algorithms use a two-point approach to order the markers, if you set hmm=TRUE a multipoint approach is applied to estimate the genetic distances after the order is estimated. Thus, it can happen that some markers are not considered linked when evaluated by multipoint information, and the function will return an error like this:

ERROR: The linkage between markers 1 and 2 did not reach the OneMap default criteria. They are probably segregating independently

You can automatically remove these markers setting argument rm_unlinked = TRUE. The marker will be removed, and the ordering algorithms will be restarted. Warning messages will inform which markers were removed. If you don't get warning messages, it means that any marker needed to be removed. This is our case in this example, but if you obtain an error or warning running your dataset, you already know what happened.

NOTE: (new!) If you are working with f2 intercross mapping population and have many markers (more than 60), we suggest to first use hmm=FALSE to check the ordering and after speed up mds using BatchMap parallelization approach. See section Speed up analysis with parallelization for more information.

By now, we ordered our group in several ways, but, which one result in the best order? We can check it by plotting the color scale recombination fraction matrix and see if the generated maps obey the colors patterns expected.

Plotting the recombination fraction matrix

It is possible to plot the recombination fraction matrix and LOD Scores based on a color scale using the function rf_graph_table. This matrix can be useful to make some diagnostics about the map.

Ordered markers are presented on both axes. Hotter the colors lower the recombination fraction between markers related to each cell. Good map orders have hot colors in the diagonal and it gradually turns to blue at the superior left and inferior right corners. This pattern means that markers that have low recombination fractions are really positioned closely on the map.

Let's see the maps we have until now:

rf_graph_table(LG1_rcd_f2)
rf_graph_table(LG1_rec_f2)
rf_graph_table(LG1_ug_f2)
rf_graph_table(LG1_mds_f2)

With default arguments, the graphic cells represent the recombination fractions. If you change the argument to graph.LOD = TRUE, LOD score values are plotted. The color scale varies from red (small distances and big LODs) to dark blue. You can also change the number of colors from the rainbow palette with argument n.colors, add/remove graphic main and axis title (main and lab.xy), and shows marker numbers, instead of names in the axis (mrk.axis).

We can see that any of the algorithms gave an optimal ordering, there are many red cells out of the diagonal, the color pattern is broke in all graphics, but we need to do an effort to find which one of the algorithms gave the best result. Then, we continue the analysis by doing hand manipulations.

Here, ug and record approaches are the ones that gave results more close to the expected pattern. We can also see that probably there is more than one group, because of the big gaps. Let's see the map generated by the ug algorithm with more details using the interactive mode of the rf_graph_table function:

rf_graph_table(LG1_ug_f2, inter = TRUE, html.file = "test.html")

An interactive version of the graphic will pop up (not shown here) in your internet browser end generated an HTML file in your work directory. Hover the mouse cursor over the cell corresponding to two markers, you can see some useful information about them.

Markers from 89 to 11 seem to form a separated group (we will call this group LG1). Markers from 23 to 55 will be our LG2. Markers from 39 to 50 show strong evidence that constitutes another separated group, including markers (LG3).

Let's separate them and order again using the same method:

# New LG1 will be this separated group
pos11 <- which(LG1_ug_f2$seq.num == 11) # Find position of marker 11
mksLG1 <- LG1_ug_f2$seq.num[1:pos11] # From marker 89 to 11

# LG2
pos23 <- which(LG1_ug_f2$seq.num == 23)
pos55 <- which(LG1_ug_f2$seq.num == 55)
mksLG2 <- LG1_ug_f2$seq.num[pos23:pos55]

# LG3
pos39 <- which(LG1_ug_f2$seq.num == 39)
mksLG3 <- LG1_ug_f2$seq.num[pos39:length(LG1_ug_f2$seq.num)] # use the position to find the 39 marker and take all the markers from there to the end of sequence

# Ordering again LG1
LG1 <- make_seq(twopts_f2, mksLG1)
LG1_ug2_f2 <- ug(LG1, hmm = F)

rf_graph_table(LG1_ug2_f2) # Now it is better

# Ordering LG2
LG2 <- make_seq(twopts_f2, mksLG2)
LG2_ug_f2 <- ug(LG2, hmm = F)

rf_graph_table(LG2_ug_f2)

# Ordering LG3
LG3 <- make_seq(twopts_f2, mksLG3)
LG3_ug_f2 <- ug(LG3, hmm = F)

rf_graph_table(LG3_ug_f2)

Besides the groups now is better separated, the order is still not the best we can do. With fewer markers in each group, we can apply a multipoint strategy to better order those markers, starting with group 1.

Linkage group 1

When possible (i.e., when groups have a small number of markers, in general up to 10 or 11), one should select the best order by comparing the multipoint likelihood of all possible orders between markers (exhaustive search). This procedure is implemented in the function compare. Although feasible for up to 10 or 11 markers, with 7 or more markers it will take a couple of hours until you see the results (depending of course on the computational resources available).

All our groups have more than 7 markers, so using the function compare is infeasible. Thus we will apply a heuristic that shows reliable results. First, we will choose a moderate number of markers, say 6, to create a framework using the function compare, and then we will position the remaining markers into this framework using the function try_seq. The way we choose these initial markers in inbred-based populations is somewhat different from what we did for outcrossing populations, where there is a mixture of segregation patterns (see the vignette for details).

In our scenario, we recommend two methods:

1. Randomly choose a number of markers and calculate the multipoint likelihood of all possible orders (using the function compare). If the LOD Score of the second-best order is greater than a given threshold, say, 3, then take the best order to proceed with the next step. If not, repeat the procedure.

2. Use some two-point based algorithm to construct a map; then, take equally spaced markers from this map. Then, create a framework of ordered markers using the function compare. Next, try to map the remaining markers, one at a time, beginning with co-dominant ones (most informative ones), then add the dominant ones.

You can do this procedure manually, in a similar way as done for outcrossing species (see the vignette for details). However, this procedure is automated in function order_seq, which we will use here (it will take some time to run):

LG1_f2_ord <- order_seq(input.seq = LG1_ug2_f2, n.init = 5,
                        subset.search = "twopt",
                        twopt.alg = "rcd", THRES = 3)

The first argument is an object of class sequence (LG1_ug2_f2). n.init = 5 means that five markers will be used in the compare step. The argument subset.search = "twopt" indicates that these five markers should be chosen by using a two-point method, which will be Rapid Chain Delineation, as indicated by the argument twopt.alg = "rcd". THRES = 3 indicates that the try_seq step will only add markers to the sequence which can be mapped with LOD Score greater than 3.

Check the order obtained by this procedure:

LG1_f2_ord # Results not shown in this vignette

Markers 5, 66, and 2 could not be safely mapped to a single position (LOD Score > THRES in absolute value). The output displays the safe order and the most likely positions for markers not mapped, where *** indicates the most likely position, and * corresponds to other plausible positions. (If you are familiar with MAPMAKER/EXP, you will recognize the representation).

To get the safe order, use:

LG1_f2_safe <- make_seq(LG1_f2_ord, "safe")

and to get the order with all markers (i.e., including the ones not mapped to a single position), use:

(LG1_f2_all <- make_seq(LG1_f2_ord, "force"))

which places markers 5, 66, and 2 into their most likely positions.

Although some old publications presented maps with only safe orders, we see no reason not to use the option force and recommend it for users. In the next, steps we can make modifications to the map based on more information.

The order_seq function can perform two rounds of the try_seq step, first using THRES and then THRES - 1 as the threshold. This generally results in safe orders with more markers mapped but takes longer to run. To do this, type (it will take some time to run):

LG1_f2_ord <- order_seq(input.seq = LG1_ug2_f2, n.init = 5,
                        subset.search = "twopt",
                        twopt.alg = "rcd", THRES = 3,
                        touchdown = TRUE)

The output is too big to be included here, so please try it to see what happens. In short, for this particular sequence, the touchdown the step could additionally map markers 5 and 66, but this depends on the dataset. Let us continue our analysis using the order with all markers as suggested by the function order_seq:

(LG1_f2_final <- make_seq(LG1_f2_ord, "force"))

rf_graph_table(LG1_f2_final)

Finally, to check for alternative orders, use the ripple_seq function:

ripple_seq(LG1_f2_final, ws = 5, LOD = 3)

The second argument, ws = 5, means that subsets (windows) of five markers will be permuted sequentially (5! orders for each window), to search for other plausible orders. The LOD argument means that only orders with LOD Score smaller than 3 will be printed.

The output shows sequences of five numbers, because ws = 5. They can be followed by an OK, if there are no alternative orders with LOD Scores smaller than LOD = 3 in absolute value, or by a list of alternative orders.

In this example, the first two windows showed alternative orders with LOD smaller than LOD = 3. However, the best order was that obtained with the order_seq function (LOD = 0.00). If there were an alternative order more likely than the original, one should check the difference between them and, if necessary, change the order with (for example) functions drop_marker (see Section about using an arbitrary order) and try_seq, or simply by typing the new order. For that, use LG2_f2_final$seq.num to obtain the original order; then make the necessary changes (by copying and pasting) and use the function map to reestimate the genetic map for the new order.

Linkage group 2

Here we will use the multipoint approach to re-order the LG2. The approach is the automatic usage of the try algorithm:

LG2_f2_ord <- order_seq(input.seq = LG2_ug_f2, n.init = 5,
                        subset.search = "twopt",
                        twopt.alg = "rcd", THRES = 3,
                        touchdown = TRUE, rm_unlinked = TRUE)

The second round of try_seq added markers 23, 29, 67, 44, 68, 36, 40, 26, 63, 31, 17, 12, 75, 74, 58, 35, 13, 6, 70, 72, 30, 69, 1, 46, 42, 3, 27 and 55 (try it; results not shown).

Get the order with all markers:

(LG2_f2_final <- make_seq(LG2_f2_ord, "force"))

rf_graph_table(LG2_f2_final)

Our heatmap is already very good with an automatic approach, but I may want to see what happens if I remove a marker and try to reposition it. To remove a marker use function drop_marker:

LG2_edit <- drop_marker(LG2_f2_final, 23) # removing marker 23

After, you need to re-estimate the parameters using the same previous order. For that, use the map function:

LG2_edit_map <- map(LG2_edit) 

Warning: If you find an error message like:

Error in as_mapper(.f, ...) : argument ".f" is missing, with no default

It's because the map function has a very common name and you can have in your environment other functions with the same name. In the case of the presented error, R is using the map function from purrr package instead of OneMap, to solve this, simply specify that you want the OneMap function with :: command from stringr package:

library(stringr)
LG2_edit_map <- onemap::map(LG2_edit) 

NOTE: (new!) If you are working with f2 intercross mapping population and have many markers (more than 60), we suggest to speed up map using BatchMap parallelization approach. See section Speed up analysis with parallelization for more information.

See what happened:

rf_graph_table(LG2_edit_map)

And try to reposition the marker:

(LG2_temp <- try_seq(input.seq = LG2_edit_map, mrk = 23))

The result shows us that the best position for this marker (with higher LOD) is the 1 (as before). Let's now include the marker at this position using:

LG2_f2_final <- make_seq(LG2_temp, 1)

rf_graph_table(LG2_f2_final)

Check the final map (results not shown):

ripple_seq(LG2_f2_final, ws = 5)

Print it:

LG2_f2_final

This is the final version of the map for this linkage group.

Linkage group 3

Automatic usage of try algorithm.

LG3_f2_ord <- order_seq(input.seq = LG3_ug_f2, n.init = 5,
                        subset.search = "twopt",
                        twopt.alg = "rcd", THRES = 3,
                        touchdown = TRUE)

A careful examination of the graphics can be a good source of information about how markers were placed.

Now, get the order with all markers:

(LG3_f2_final <- make_seq(LG3_f2_ord, "force"))

Check heatmap:

rf_graph_table(LG3_f2_final)

Markers 34, 39, 50, 56, 20, 24 and 64 seem to broke the color pattern. We will remove them and use try_seq to find better locations:

LG3_edit <- drop_marker(LG3_f2_final, c(34,39,50,56, 20,24, 64))
LG3_edit_map <- order_seq(LG3_edit) # We remove several markers maybe it's better to order again
LG3_edit_map <- make_seq(LG3_edit_map, "force")

rf_graph_table(LG3_edit_map)

Adding one by one we can see how each one changes the map log-likelihood, map size, and the color pattern in the heatmap and judge if we will remove them.

LG3_edit <- try_seq(LG3_edit_map, 34)
LG3_edit_temp <- make_seq(LG3_edit, 1) # Not included
LG3_edit <- try_seq(LG3_edit_map, 39)
LG3_edit_temp <- make_seq(LG3_edit, 3) 
LG3_edit_map <- LG3_edit_temp # include
LG3_edit <- try_seq(LG3_edit_map, 50)
LG3_edit_temp <- make_seq(LG3_edit, 22) # Not included
LG3_edit <- try_seq(LG3_edit_map, 56)
LG3_edit_temp <- make_seq(LG3_edit, 22) # Not included 
LG3_edit <- try_seq(LG3_edit_map, 20)
LG3_edit_temp <- make_seq(LG3_edit, 22) 
LG3_edit_map <- LG3_edit_temp # include
LG3_edit <- try_seq(LG3_edit_map, 24)
LG3_edit_temp <- make_seq(LG3_edit, 22) 
LG3_edit_map <- LG3_edit_temp # include
LG3_edit <- try_seq(LG3_edit_map, 64)
LG3_edit_temp <- make_seq(LG3_edit, 23) # Not included

LG3_f2_final <- LG3_edit_map

Check the final map (not shown):

ripple_seq(LG3_f2_final, ws = 5)

The fifth window presented alternative orders that seem better than the current one. Let's try to substitute the current order with the best presented by ripple. See that, instead of 59-49-76-28-80 we have 59-49-76-80-28. Let's find this window and substitute the sequence:

idx <- which(LG3_f2_final$seq.num == 59) 
new_seq <- LG3_f2_final$seq.num
new_seq[idx:(idx+4)] <- c(59, 49, 76, 80, 28)
LG3_edit_seq <- make_seq(twopts_f2, new_seq)

Now, we estimate the distances for this already known order using map:

LG3_edit_map <- onemap::map(LG3_edit_seq)

Print it:

LG3_f2_final <- LG3_edit_map

rf_graph_table(LG3_f2_final)

Using the recombinations and the reference genome information

In our example, we have reference genome chromosome and position information for some of the markers, here we will exemplify one method of using this information to help build the genetic map.

With the CHROM information in the input file, you can identify markers belonging to some chromosome using the function make_seq with the rf_2pts object. For example, assign the string "1" for the second argument to get chromosome 1 makers. The output sequence will be automatically ordered by POS information.

CHR1 <- make_seq(twopts_f2, "1")
CHR1
CHR2 <- make_seq(twopts_f2, "2")
CHR3 <- make_seq(twopts_f2, "3")

Adding markers with no reference genome information

According to CHROM information we have three defined linkage groups, now we can try to group the markers without chromosome information to them using recombination information. For this, we can use the function group_seq:

CHR_mks <- group_seq(input.2pts = twopts_f2, seqs = "CHROM", unlink.mks = mark_all_f2,
                     repeated = FALSE)

The function works as the function group but considering pre-existing sequences. Setting seqs argument with the string "CHROM", it will consider the pre-existing sequences according to CHROM information. You can also indicate other pre-existing sequences if it makes sense for your study. For that, you should inform a list with objects of class sequences, as the example:

CHR_mks <- group_seq(input.2pts = twopts_f2, seqs = list(CHR1=CHR1, CHR2=CHR2, CHR3=CHR3), 
                     unlink.mks = mark_all_f2, repeated = FALSE)

In this case, the command had the same effect as the previous, because we indicate chromosome sequences, but others sequences can be used.

The unlink.mks argument receives an object of class sequence, this defines which markers will be tested to group with the sequences in seqs. In our example, we will indicate only the markers with no segregation distortion, using the sequence mark_no_dist. It is also possible to use the string "all" to test all the remaining markers at the rf_2pts object.

In some cases, the same marker can group into more than one sequence, those markers will be considered repeated. We can choose if we want to remove or not (FALSE/TRUE) them of the output sequences, with the argument rm.repeated. Anyway, their numbers will be informed at the list repeated in the output object.

In the example case, there are no repeated markers. However, if they exist, it could indicate that their groups actually constitute the same group. Also, genotyping errors can generate repeated markers. Anyway, they deserve better investigations.

We can access detailed information about the results just by printing:

CHR_mks

Also, we can access the numbers of repeated markers with:

CHR_mks$repeated

In the same way, we can access the output sequences:

CHR_mks$sequences$CHR1
# or
CHR_mks$sequences[[1]]

For this function, optional arguments are LOD and max.rf, which define thresholds to be used when assigning markers to linkage groups. If none is provided (default), criteria previously defined for the object rf_2pts are used.

In our example, all markers without genomic information grouped with all chromosomes, then this approach did not give us more information about grouping and from here we could follow the same strategy did in Ordering markers within linkage groups.

Map estimation for an arbitrary order

If you have some information about the order of the markers, for example, from a reference genome or previously published paper, you can define a sequence of those markers in a specific order (using the function make_seq) and then use the function map to estimate the final genetic map (based on multi-point analysis). For example, let us assume that we know that the following markers are ordered in this sequence: M47, M38, M59, M16, M62, M21, M20, M48 and M22. In this case, select them from the two-point analysis, and use function map:

LG3seq_f2 <- make_seq(twopts_f2, c(47, 38, 59, 16, 62, 21, 20, 48, 22))

LG3seq_f2_map <- map(LG3seq_f2)

Warning: If you find an error message like:

Error in as_mapper(.f, ...) : argument ".f" is missing, with no default

It's because the map function has a very common name and you can have in your environment another function with the same name. In the case of the presented error, R is using the map function from purrr package instead of OneMap, to solve this, simply specify that you want the OneMap function with :: command from stringr package:

library(stringr)
(LG3seq_f2_map <- onemap::map(LG3seq_f2))

NOTE: (new!) If your map population is F2 intercross and your sequence has many markers (more than 60), we suggest to speed up map using BatchMap parallelization approach. See section Speed up analysis with parallelization for more information.

This is a subset of the first linkage group. When used this way, the map function searches for the best combination of phases between markers and prints the results.

To see the correspondence between marker names and numbers, use function marker_type:

marker_type(LG3seq_f2_map)

If one needs to add or drop markers from a predefined sequence, functions add_marker and drop_marker can be used. For example, to add markers M18, M56 and M50 at the end of LG3seq_f2_map:

(LG3seq_f2_map <- add_marker(LG3seq_f2_map, c(18, 56, 50)))

Removing markers M59 and 21 from LG3seq_f2_map:

(LG3seq_f2_map <- drop_marker(LG3seq_f2_map, c(59, 21)))

Drawing the genetic map

Once all linkage groups were obtained using both strategies, we can draw a map for each strategy using the function draw_map. Since version 2.1.1007, OneMap has a new version of draw_map, called draw_map2. The new function draws elegant linkage groups and presents new arguments to personalize your draw.

If you prefer the old function, we also keep it. Follow examples of how to use both of them.

Draw_map

We can draw a genetic map for all linkage groups using the function draw_map. First, we have to create a list of ordered linkage groups:

maps_list <- list(LG1_f2_final, LG2_f2_final, LG3_f2_final)

Then use function draw_map for this list:

draw_map(maps_list, names = TRUE, grid = TRUE, cex.mrk = 0.7)

We also can draw a map for a specific linkage group:

draw_map(LG1_f2_final, names = TRUE, grid = TRUE, cex.mrk = 0.7)

Function draw_map draws a very simple graphic representation of the genetic map. More recently, we developed a new version called draw_map2 that draws a more sophisticated figure. Furthermore, once the distances and the linkage phases are estimated, other map figures can be drawn by the user with any appropriate software.

Draw_map2

The same figures did with draw_map can be done with the draw_map2 function. But it has different capacities and arguments. Here are some examples, but you can find more options on the help page ?write_map2.

Let's draw all three groups built:

draw_map2(LG1_f2_final, LG2_f2_final, main = "Only linkage information", 
          group.names = c("LG1", "LG2", "LG3"), output = "map.eps")

NOTE: Check the GitHub vignette version to visualize the graphic.

You can include all sequence objects referring to the groups as the first arguments. The main argument defines the main title of the draw and group.names define the names of each group. If no output file and file extension is defined, the draw will be generated at your working directory as map.eps. The eps extension is only the default option but there are others that can be used. You can have access to a list of them on the help page.

We also can draw a map for a specific linkage group:

draw_map2(LG1_f2_final, col.group = "#58A4B0", col.mark = "#335C81", output = "map_LG1.pdf")

NOTE: Check the GitHub vignette version to visualize the graphic.

You can also change the default colors using the col.group and col.mark arguments.

With argument tag you can highlight some markers at the figure according to your specific purpose.

Speed up analysis with parallelization (new!)

Warning: Only available for outcrossing and f2 intercross populations.

As already mentioned, OneMap uses HMM multipoint approach to estimate genetic distances, a very robust method, but it can take time to run if you have many markers. In 2017, Schiffthaler et. al release an OneMap fork with modifications in CRAN and in GitHub with the possibility of parallelizing the HMM chain dividing markers in batches and use different cores for each phase. Their approach speeds up our HMM and keeps the genetic distances estimation quality. It allows dividing the job into a maximum of four cores according to the four possible phases for outcrossing and f2 mapping populations. We add this parallelized approach to the functions: map, mds_onemap, seriation, rcd, record and ug. For better efficiency it is important that batches are composed of 50 markers or more, therefore, this approach is only recommended for linkage groups with many markers.

The parallelization is here available for all types of operational systems, however, we suggest setting argument parallelization.type to FORK if you are not using Windows system. It will improve the procedure speed.

Here we will show an example of how to use the BatchMap approach in some functions that requires HMM. For this, we simulated a dataset with a group with 300 markers (we don't want this vignette to take too much time to run, but usually maps with markers from high-throughput technologies result in larger groups). Before start, you can see the time spent for each approach (See also Session Info) in this example:

data(parallel_results_out)
time_spent <-time_spent/(60*60)
colnames(time_spent) <- c("Without parallelization (h)", "With parallelization (h)" )
knitr::kable(time_spent)

| | Without parallelization (h)| With parallelization (h)| |:----------|---------------------------:|------------------------:| |rcd | 0.6801889| 0.1260436| |record_map | 1.8935892| 0.3330297| |ug_map | 1.1002725| 0.2256356| |mds_onemap | 2.1478900| 0.4443492| |map | 2.1042114| 0.6156722|

# Simulation using onemapUTILS
run_pedsim(chromosome = "Chr1", n.marker = 300, tot.size.cm = 100, centromere = 50,
           n.ind = 200, mk.types = c("A.H.B", "C.A", "D.B"),
           n.types = rep(100,3), pop = "F2", 
           name.mapfile = "mapfile.txt", name.founderfile="founderfile.gen",
           name.chromfile="sim.chrom", name.parfile="sim.par",
           name.out="simParall_f2")

# Do the conversion

pedsim2raw(cross="f2 intercross", genofile = "simParall_f2_genotypes.dat", 
           parent1 = "P1", parent2 = "P2", out.file = "simParall_f2.raw", 
           miss.perc = 25)

# Import to R environment as onemap object

simParallel <- read_onemap("simParall_f2.raw")
plot(simParallel)

simParallel <- read_onemap(system.file("extdata/simParall_f2.raw", package = "onemap")) # dataset available only in onemap github version

# Calculates two-points recombination fractions
twopts <- rf_2pts(simParallel)

seq_all <- make_seq(twopts, "all")

# There are no redundant markers
find_bins(simParallel)

# There are no distorted markers
print(test_segregation(simParallel)) # Not shown

To prepare the data with defined bach size we use function pick_batch_sizes. It selects a batch size that splits the data into even groups. Argument size defines the batch size next to which an optimum size will be searched. overlap defines the number of markers that overlap between the present batch and next. This is used because pre-defined phases at these overlap markers in the present batch are used to start the HMM in the next batch. The around argument defines how much the function can vary around the defined number in size to search for the optimum batch size.

Some aspects should be considered to define these arguments because if the batch size were set too high, there would be less gain in execution time. If the overlap size would be too small, phases would be incorrectly estimated and large gaps would appear in the map, inflating its size. In practice, these values will depend on many factors such as population size, marker quality, and species. BatchMap authors recommended trying several configurations on a subset of data and select the best performing one.

batch_size <- pick_batch_sizes(input.seq = seq_all, 
                               size = 80, 
                               overlap = 30, 
                               around = 10)

batch_size

Speed up two-points ordering approaches

To use the parallelized approach you just need to include the arguments when using the functions:

time_spent <- data.frame("without-parallelization"= rep(0,5), "with-parallelization" =rep(0,5))
rownames(time_spent) <- c("rcd", "record_map", "ug_map", "mds_onemap", "map")

# Without parallelization 
time <- system.time(rcd_map <- rcd(input.seq = seq_all))
time_spent$without.parallelization[1] <- time[3]

# With parallelization 
time <- system.time(rcd_map_par <- rcd(input.seq = seq_all,
                                       phase_cores = 4, 
                                       size = batch_size, 
                                       overlap = 30))

time_spent$with.parallelization[1] <- time[3]

# Without parallelization 
rcd_map <- rcd(input.seq = seq_all)

# With parallelization 
rcd_map_par <- rcd(input.seq = seq_all,
                   phase_cores = 4, 
                   size = batch_size, 
                   overlap = 30)

a <- rf_graph_table(rcd_map, mrk.axis = "none")
b <- rf_graph_table(rcd_map_par, mrk.axis = "none")

p <- ggarrange(a,b , common.legend = TRUE,
               labels = c("rcd", "rcd + parallel"), 
               vjust = 0.2,
               hjust= -1.4,
               font.label = list(size = 10),
               ncol=2, nrow=1)

ggsave(p, filename = "rcd.jpg")

# Without parallelization 
time <- system.time(record_map <- record(input.seq = seq_all))
time_spent$without.parallelization[2] <- time[3]

# With parallelization 
time <- system.time(record_map_par <- record(input.seq = seq_all,
                                             phase_cores = 4, 
                                             size = batch_size, 
                                             overlap = 30))
time_spent$with.parallelization[2] <- time[3]

# Without parallelization 
record_map <- record(input.seq = seq_all)

# With parallelization 
record_map_par <- record(input.seq = seq_all,
                         phase_cores = 4, 
                         size = batch_size, 
                         overlap = 30)

a <- rf_graph_table(record_map, mrk.axis = "none")
b <- rf_graph_table(record_map_par, mrk.axis = "none")

p <- ggarrange(a,b , common.legend = TRUE,
               labels = c("record", "record + parallel"), 
               vjust = 0.2,
               hjust= -0.8,
               font.label = list(size = 10),
               ncol=2, nrow=1)

ggsave(p, filename = "record.jpg")

# Without parallelization 
time <- system.time(ug_map <- ug(input.seq = seq_all))
time_spent$without.parallelization[3] <- time[3]

# With parallelization
time <- system.time(ug_map_par <- ug(input.seq = seq_all,
                                     phase_cores = 4, 
                                     size = batch_size, 
                                     overlap = 30))

time_spent$with.parallelization[3] <- time[3]

# Without parallelization 
ug_map <- ug(input.seq = seq_all)

# With parallelization
ug_map_par <- ug(input.seq = seq_all,
                 phase_cores = 4, 
                 size = batch_size, 
                 overlap = 30)

a <- rf_graph_table(ug_map, mrk.axis = "none")
b <- rf_graph_table(ug_map_par, mrk.axis = "none")

p <- ggarrange(a,b , common.legend = TRUE,
               labels = c("ug", "ug + parallel"), 
               vjust = 0.2,
               hjust= -1.6,
               font.label = list(size = 10),
               ncol=2, nrow=1)

ggsave(p, filename = "ug.jpg")

Speed up mds_onemap

# Without parallelization ok
time <- system.time(map_mds <- mds_onemap(input.seq = seq_all))
time_spent$without.parallelization[4] <- time[3]

# With parallelization
time <- system.time(map_mds_par <- mds_onemap(input.seq = seq_all, 
                                              phase_cores = 4, 
                                              size = batch_size, 
                                              overlap = 30))

time_spent$with.parallelization[4] <- time[3]

# Without parallelization ok
map_mds <- mds_onemap(input.seq = seq_all)

# With parallelization
map_mds_par <- mds_onemap(input.seq = seq_all, 
                          phase_cores = 4, 
                          size = batch_size, 
                          overlap = 30)

a <- rf_graph_table(map_mds, mrk.axis = "none")
b <- rf_graph_table(map_mds_par, mrk.axis = "none")

p <- ggarrange(a,b , common.legend = TRUE,
               labels = c("mds", "mds + parallel"), 
               vjust = 0.2,
               hjust= -1,
               font.label = list(size = 10),
               ncol=2, nrow=1)

ggsave(p, filename = "mds.jpg")

Speed up map estimation for an arbitrary order (map function)

Because we simulate this dataset we know the correct order. We can use map_overlapping_batches to estimate genetic distance in this case. This is equivalent to map, but with the parallelized process.

batch_map <- map_overlapping_batches(input.seq = seq_all,
                                     size = batch_size,
                                     phase_cores = 4,
                                     overlap = 30, 
                                     rm_unlinked = TRUE)

Similarly with map, using argument rm_unlinked = TRUE the function will return a vector with marker numbers without the problematic marker. To repeat the analysis removing automatically all problematic markers use map_avoid_unlinked:

# Without parallelization
time <- system.time(batch_map <- map_avoid_unlinked(input.seq = seq_all))
time_spent$without.parallelization[5] <- time[3]

# With parallelization
time <- system.time(batch_map_par <- map_avoid_unlinked(input.seq = seq_all,
                                                        size = batch_size,
                                                        phase_cores = 4,
                                                        overlap = 30))

time_spent$with.parallelization[5] <- time[3]

# Without parallelization
batch_map <- map_avoid_unlinked(input.seq = seq_all)

# With parallelization
batch_map_par <- map_avoid_unlinked(input.seq = seq_all,
                                    size = batch_size,
                                    phase_cores = 4,
                                    overlap = 30)

a <- rf_graph_table(batch_map, mrk.axis = "none")
b <- rf_graph_table(batch_map_par, mrk.axis = "none")

p <- ggarrange(a,b , common.legend = TRUE,
               labels = c("map", "map + parallel"), 
               vjust = 0.2,
               hjust= -1,
               font.label = list(size = 10),
               ncol=2, nrow=1)

ggsave(p, filename = "map.jpg")

As you can see in the above maps, heuristic ordering algorithms do not return an optimal order result, mainly if you don't have many individuals in your population. Because of the erroneous order, generated map sizes are not close to the simulated size (100 cM) and their heatmaps don't present the expected color pattern. Two of them get close to the color pattern, they are the ug and the MDS method. They present good global ordering but not local. If you have a reference genome, you can use its position information to rearrange local ordering.

Export estimated progeny haplotypes (new!)

Function progeny_haplotypes generates a data.frame with progeny phased haplotypes estimated by OneMap HMM. For progeny, the HMM results in probabilities for each possible genotype, then the generated data.frame contains all possible genotypes. If most_likely = TRUE the most likely genotype receives 1 and the rest 0 (if there are two most likely both receive 0.5), if most_likely = FALSE genotypes probabilities will be according to the HMM results. You can choose which individual to be evaluated in ind. The data.frame is composed of the information: individual (ind) and group (grp) ID, position in centimorgan (pos), progeny homologs (homologs), and from each parent the allele came (parents).

(progeny_haplot <- progeny_haplotypes(LG2_f2_final, most_likely = TRUE, ind = 2, group_names = "LG2_final"))

You can also have a view of progeny estimated haplotypes using plot. It shows which markers came from each parent's homologs. position argument defines if haplotypes will be plotted by homologs (stack) or alleles (split). split option is a good way to better view the likelihoods of each allele.

plot(progeny_haplot, position = "stack")

plot(progeny_haplot, position = "split")

Final remarks

At this point, it should be clear that any potential OneMap user must have some knowledge about genetic mapping and also the R language, because the analysis is not done with only one mouse click. In the future, perhaps a graphical interface will be made available to make this software is a lot easier to use.

We do hope that OneMap is useful to researchers interested in genetic mapping in outcrossing or inbred-based populations. Any suggestions and critics are welcome.

Session Info

sessionInfo()

References

Adler, J. R in a Nutshell A Desktop Quick Reference, 2009.

Broman, K. W., Wu, H., Churchill, G., Sen, S., Yandell, B. qtl: Tools for analyzing QTL experiments R package version 1.09-43, 2008.

Buetow, K. H., Chakravarti, A. Multipoint gene mapping using seriation. I. General methods. American Journal of Human Genetics 41, 180-188, 1987.

Doerge, R.W. Constructing genetic maps by rapid chain delineation. Journal of Agricultural Genomics 2, 1996.

Garcia, A.A.F., Kido, E.A., Meza, A.N., Souza, H.M.B., Pinto, L.R., Pastina, M.M., Leite, C.S., Silva, J.A.G., Ulian, E.C., Figueira, A. and Souza, A.P. Development of an integrated genetic map of a sugarcane Saccharum spp. commercial cross, based on a maximum-likelihood approach for estimation of linkage and linkage phases. Theoretical and Applied Genetics 112, 298-314, 2006.

Haldane, J. B. S. The combination of linkage values and the calculation of distance between the loci of linked factors. Journal of Genetics 8, 299-309, 1919.

Jiang, C. and Zeng, Z.-B. Mapping quantitative trait loci with dominant and missing markers in various crosses from two inbred lines. Genetica 101, 47-58, 1997.

Kosambi, D. D. The estimation of map distance from recombination values. Annuaire of Eugenetics 12, 172-175, 1944.

Lander, E. S. and Green, P. Construction of multilocus genetic linkage maps in humans. Proc. Natl. Acad. Sci. USA 84, 2363-2367, 1987.

Lander, E.S., Green, P., Abrahanson, J., Barlow, A., Daly, M.J., Lincoln, S.E. and Newburg, L. MAPMAKER, An interactive computing package for constructing primary genetic linkage maps of experimental and natural populations. Genomics 1, 174-181, 1987.

Lincoln, S. E., Daly, M. J. and Lander, E. S. Constructing genetic linkage maps with MAPMAKER/EXP Version 3.0: a tutorial and reference manual. A Whitehead Institute for Biomedical Research Technical Report 1993.

Matloff, N. The Art of R Programming. 2011. 1st ed. San Francisco, CA: No Starch Press, Inc., 404 pages.

Margarido, G. R. A., Souza, A.P. and Garcia, A. A. F. OneMap: software for genetic mapping in outcrossing species. Hereditas 144, 78-79, 2007.

Mollinari, M., Margarido, G. R. A., Vencovsky, R. and Garcia, A. A. F. Evaluation of algorithms used to order markers on genetics maps. Heredity 103, 494-502, 2009.

Oliveira, K.M., Pinto, L.R., Marconi, T.G., Margarido, G.R.A., Pastina, M.M., Teixeira, L.H.M., Figueira, A.M., Ulian, E.C., Garcia, A.A.F., Souza, A.P. Functional genetic linkage map on EST-markers for a sugarcane (Saccharum spp.) commercial cross. Molecular Breeding 20, 189-208, 2007.

Oliveira, E. J., Vieira, M. L. C., Garcia, A. A. F., Munhoz, C. F.,Margarido, G. R.A., Consoli, L., Matta, F. P., Moraes, M. C., Zucchi, M. I., and Fungaro,M. H. P. An Integrated Molecular Map of Yellow Passion Fruit Based on Simultaneous Maximum-likelihood Estimation of Linkage and Linkage Phases J. Amer. Soc. Hort. Sci. 133, 35-41, 2008.

Tan, Y., Fu, Y. A novel method for estimating linkage maps. Genetics 173, 2383-2390, 2006.

Van Os H, Stam P, Visser R.G.F., Van Eck H.J. RECORD: a novel method for ordering loci on a genetic linkage map. Theor Appl Genet 112, 30-40, 2005.

Voorrips, R.E. MapChart: software for the graphical presentation of linkage maps and QTLs. Journal of Heredity 93, 77-78, 2002.

Wang S., Basten, C. J. and Zeng Z.-B. Windows QTL Cartographer 2.5. Department of Statistics, North Carolina State University, Raleigh, NC, 2010. https://brcwebportal.cos.ncsu.edu/qtlcart/

Wu, R., Ma, C.X., Painter, I. and Zeng, Z.-B. Simultaneous maximum likelihood estimation of linkage and linkage phases in outcrossing species. Theoretical Population Biology 61, 349-363, 2002a.

Wu, R., Ma, C.-X., Wu, S. S. and Zeng, Z.-B. Linkage mapping of sex-specific differences. Genetical Research 79, 85-96, 2002b.

onemap documentation built on Nov. 26, 2022, 9:05 a.m.