omicsDB

Home
Downloads
Documents
Contact

Home

SVLearn

A machine learning-based genotyping tool for structural variation of short reads

Update (0.0.5)

In the latest 0.0.5 release, SVLearn has introduced the --no-filter-overlaps option to the prepareAlt subcommand. When building the ALT genome, this option disables the filtering of overlapping SVs, ensuring that every input SV is included in the genotyping analysis.

Please cite

Yang, Q., Sun, J., Wang, X. et al. SVLearn: a dual-reference machine learning approach enables accurate cross-species genotyping of structural variants. Nat Commun 16, 2406 (2025).
https://doi.org/10.1038/s41467-025-57756-z

Documents

CNVcaller flow chart

CNVcaller pipeline includes three main steps. First, considering the population sequencing data may come from different platforms, the read-depth (RD) of each sample is counted and corrected individually. An original absolute copy number correction is used to modify the standard read alignments generated by BWA software to multi-hit alignments, as similar to mrsFAST format. After corrections and normalization, the comparable RDs of each sample is concentrated to a ~100 Mb intermediate file and output. This design avoids repeat calculation of a same individual in different populations.

In the second CNVR detection step, the RD files of all samples are piled up into a two dimensional population RD file. Multi-criteria are implied to remove the high-proportional noise caused by low sequencing quality or assembly bias. Individually, the RD of the candidate CNV window should significantly deviates from average. The piled-up candidate windows should also meet two population-level criteria: CNV allele frequency > 5% and the multi-sample RDs of adjacent windows are significantly correlated.

After merging the candidate CNV windows into a CNVR, the RDs of all samples in each CNVR are clustered by the mixture Gaussian model and deducing the integer copy number of each individual. This step is called genotyping as used in SNP detection. The final output is compatible with most SNP based population genetic algorithm.

Documents

Running the program

CNVcaller contains four steps consisting one perl script, two bash scripts and one python script.You need to set CNVcaller variables in the three bash scripts based on your environment.

Step 1. Indexing Reference Genome
The reference genome is segmented into overlapping sliding windows. The windows are indexed to form a reference database used in all samples. This commend will create the file referenceDB.windowsize in current directory by default.
```
 $ perl CNVReferenceDB.pl <ref>
 Required arguments
 <ref> Reference sequence
 Optional arguments
 -w the window size (bp) for all samples [default=800]
 -l the lower limit of GC content [default=0.2]
 -u the upper limit of GC content [default=0.7]
 -g the upper limit of gap content [default=0.5]
```
Argument details:

-w We recommend 400-1000bp window size for >10X coverage sequencing data, 1000-2000bp window size for <10X coverage sequencing data. Increasing the window size will reduce the noise at the cost of sensitivity.
Step 2: Individual RD processing
Count the reads of each window across genome from BAM file and generate a comparable read depth (RD) file of each individual. referenceDB.windowsize must be placed in current directory.

Three default directories RD_raw RD_absolute RD_normalized will be created in current directory in order, containing the raw read depth, read depth after absolute copy number correction and the final GC corrected normalized read depth of each sample. The name of the normalized RD file indicates the average RD (mean), STDEV of the RD and the gender (1=XX/ZZ, 2=XY/ZW) of this sample. The final read depths are normalized to one.

This step consumes about 500 MB for each individual, multiple tasks can be run in parallel. Shell script Individual.Process.sh is provided to complete these procedures.
```
 $ bash Individual.Process.sh -b <bam> -h <header> -d <dup> -s <sex_chromosome>
 Required arguments
 -b|--bam      alignment file in BAM format
 -h|--header   header of bam file, the prefix of output file
 -d|--dup      duplicated window record file used for absolute copy number correction
 -s|--sex      the name of sex chromosome
```
Argument details

-dup The duplicated window record files. We provide duplicated window record files for different species, such as human, goat, sheep, pig, cattle, chicken, maize, wheat, and soybean. If you work with other organisms, you will want to create duplicated window record file in order to use absolute copy number correction function of CNVcaller. Follow the instruction.

-sThe gender of this individual will be determines by the ratio of RD of the given sex chromosome and the RD of the other autosomes. The name of X or Z chromosome should be given for the XY or ZW genomes.

Example, to convert ERR340328.bam to normalized copy number using 1000bp window size.
```
 bash Individual.Process.sh -b ERR340328.bam -h ERR340328 -d link -s X
```

Step 3: CNVR detection

The normolized RD files of all samples are piled up into a two-dimensional population RD file. The integrated CNVR are detected by scanning the population RD file with aberrantly RD, CNV allele frequency and significantly correlation with adjacent windows. The adjacent candidate windows showing high correlation will be further merged.

 $ bash CNV.Discovery.sh -l <RDFileList> -e <excludedFileList> -f <frequency> -h <homozygous> -r <pearsonCorrelation> -p <primaryCNVR> -m <mergedCNVR>
 Required arguments
 -l|--RDFileList          individual normalized read depth file list
 -e|--excludedFileList    list of samples exclude from CNVR detection
 -f|--frequency           minimum frequency of gain/loss individuals for candidate CNV window definition [recommend 0.1]
 -h|--homozygous          number of homozygous gain/loss individuals for candidate CNV window definition [recommend 3]
 -r|--pearsonCorrelation  minimum of Pearson’s correlation coefficient between the two adjacent non-overlapping windows
                          0.5 for sample size (0, 30]
                          0.4 for sample size (30, 50]
                          0.3 for sample size (50, 100]
                          0.2 for sample size (100, 200]
                          0.15 for sample size (200, 500]
                          0.1 for sample size (500,+∞)
 -p|--primaryCNVR         primary CNVR result
 -m|--mergedCNVR          merged CNVR result

Argument details

-e The samples in this list will be exclude from CNVR detection, and their copy numbers are deduced based on the CNVR boundaries defined by other samples. This option is applicable to the outgroup or the poor quality precious samples. An empty file means all individuals are included in the CNVR detection.

-f/-h Windows satisfied any of this two conditions will be selected as candidate CNV windows.

-r The adjacent windows with significant correlation will be merged in to one call. The recommend value is significant at p=0.01 level. Raise this index will increase the detection accuracy with a decrease of sensitivity.

Example, run CNV.Discovery.sh on all your individual normalized RD files for discovering CNV. An example of normalized read depth file list -l:

 RD_normalized/ERR340328_mean_70.81_SD_10.84_sex_1
 RD_normalized/ERR340329_mean_62.00_SD_10.52_sex_1
 RD_normalized/ERR340330_mean_135.66_SD_13.96_sex_1
 RD_normalized/ERR340331_mean_128.76_SD_15.27_sex_1
 RD_normalized/ERR340333_mean_69.30_SD_10.19_sex_1
 RD_normalized/ERR340334_mean_132.30_SD_14.59_sex_1
 RD_normalized/ERR340335_mean_73.50_SD_10.16_sex_1
 RD_normalized/ERR340336_mean_72.52_SD_10.03_sex_1
 RD_normalized/ERR340338_mean_124.12_SD_13.24_sex_1
 RD_normalized/ERR340340_mean_131.00_SD_14.74_sex_1

 bash CNV.Discovery.sh -l list -e exclude_list -f 0.1 -h 3 -r 0.5 -p primaryCNVR -m mergeCNVR

Step 4: Genotyping
Clustering the input samples into genotypes using Gaussian mixture modes. The output contain a genotype VCF -a VCF format file containing the input site descriptions, additional site-specific information and a called genotype for each input sample.
```
 $ python Genotype.py --cnvfile <input> --outprefix <outfile prefix>
 Required arguments:
 --cnvfile      merged CNVR file
 --outprefix    prefix of out files
 Optional arguments:
 --nproc        number of process will be used, default is one.
```
Example
```
 python Genotype.py --cnvfile mergeCNVR --outprefix Genotype
```

Home

SVLearn

Update (0.0.5)

Please cite

Downloads

Requirements

Installation

Trained Models

Demonstration

Documents

Documents

Argument details:

Argument details

Argument details

Documents