ONT Artificial Deletion Filter-Delter

A tool to filter short artificial deletion variations by Oxford Nanopore Technologies (ONT) R9 and R10 flow cells and chemistries.

Requirements

The tool has been tested on Ubuntu 20.04 with 256GB RAM, 64 CPU cores and a NVIDIA GPU with 48GB RAM. The minimal requirements should be >= 64GB RAM and a NVIDIA GPU with >= 8GB RAM. Other operating systems like Windows or Mac were not tested.

ONT softwares like Guppy, Tombo, and ont-fast5-api should be pre-installed before generating Tombo-resquiggled single-read fast5 files. Users might run following commands to preprocess R9 fast5 files in shell terminal before running our pipeline. As these steps below need GPU support and might take a long time, our pipeline doesn't contain them.

#===basecalling the fast5 files===
ont-guppy/bin/guppy_basecaller -c ont-guppy/data/dna_r9.4.1_450bps_sup.cfg -i $fast5dir/barcode${barcode} -s guppy_sup_basecalled/barcode${barcode} -r --compress_fastq -x cuda:1,2 --gpu_runners_per_device 4 --chunks_per_runner 256 --num_callers 3 --fast5_out

#===preprocessing R9 fast5 files===
c=$(ls *.fast5 | wc -l)
declare -i count=$c-1
#multiread fast5 to single read fast5
multi_to_single_fast5 -i guppy_sup_basecalled/barcode${barcode}/workspace -s fast5_pass_single --threads 24 
#copy to new directory
cd fast5_pass_single
mkdir all_single_fast5s
for ((j=0;j<=$count;j=j+1))
do
  echo $j
  cp -r ./$j/*.fast5 all_single_fast5s
  rm -rf ./$j
done
#align to reference genome via tombo resquiggle
tombo resquiggle guppy_sup_basecalled/barcode${barcode}/workspace/fast5_pass_single/all_single_fast5s data/Refs/$refseq --processes 24 --overwrite --num-most-common-errors 5  --failed-reads-filename tombo_resquiggle_failed_fast5.txt

The fast5 files in the directory named all_single_fast5s could be employed in downstream workflow, which equals the input parameter Tombo_dir in shell command line or config yaml (details listed in Configure input parameters section).

Installation

The tool runs via Snakemake workflows. Users must install workflow dependencies including Snakemake (Version >= 7.3) and R before using the pipeline. The workflow dependencies, which stored in a file named environment.yaml, are listed as below:

channels:

• conda-forge
• bioconda
• anaconda

dependencies:

• snakemake-minimal >=7.3
• graphviz
• seaborn
• numpy
• pandas
• h5py
• scipy
• samtools =1.15
• r-essentials
• r-base
• bioconductor-shortread
• r-stringr
• r-dplyr
• r-vegan

Users are suggested to use Conda or Mamba to install these dependencies. After users download the working directory containing all the necessary files, the following shell command could install Delter in a conda environment in less than half an hour.

cd /path/to/Delter/working/directory
conda env create --name Delter --file environment.yaml

or

cd /path/to/Delter/working/directory
conda create --name Delter
conda activate Delter
conda install -c bioconda -y snakemake-minimal>=7.3
conda install -c anaconda -y numpy pandas 
conda install -c anaconda -y h5py seaborn
conda install -c conda-forge -y scipy
conda install -c bioconda -y samtools=1.15
conda install -c conda-forge -y r-essentials r-base
conda install -c conda-forge -y r-dplyr r-vegan r-stringr
conda install -c bioconda -y bioconductor-shortread

Activate and exit the environment

To activate the environment

conda activate Delter

To exit the environment (after finishing the usage of the pipeline), just execute

conda deactivate

Run the pipeline

The whole pipeline could handle ONT R9 and R10 sequencing data. The working directory contains file named Delter.config.yaml, which stores key input parameters for the workflow.

The Demo data could be accessed via figshare. User should modify the filepaths in the Delter.config.yaml in the snakemakeexample directory.

Configure input parameters for the workflow

There are two ways to configure input parameters for this workflow.

(1) Via shell command line

Users could define customized input paramaters using --config option in Snakemake command line.

Usage:
snakemake -s Delter.py --cores 8 --config Ref=refname Num=5 Vcf=path/to/VCF Refseq=path/to/refseq Outdir=path/to/outputdir Bam=path/to/sorted/bam Tombo_dir=path/to/tombo_processed/fast5 Subsample=2000 Flowcell=R9 Strategy=Direct MRPPthres=0.001 HomoQthres=23 OtherQthres=20.6
Ref=refname                             The value of #CHROM in vcf file, e.g., 'Ref=chr1'
Num=5                                   The number of bases up- and down-stream that are centered around the variation loci, default=5
Vcf=path/to/VCF                         The file path to vcf file, e.g., 'Vcf=/data/res/lofreq.vcf'
Refseq=path/to/refseq                   The file path to reference sequence, e.g., 'Refseq=/database/COVID-19.fa'
Outdir=path/to/outputdir                The file path storing the output results and intermediate files, e.g., 'Outdir=/data/res'
Bam=path/to/sorted/bam                  The file path to sorted bam files, e.g., 'Bam=/data/res/sorted.bam'
Tombo_dir=path/to/tombo_processed/fast5 The file path to tombo-resquiggled single fats5 files, e.g., 'Tombo_dir=/data/fast5'
Subsample=2000                          The number to subsample from reads covering variation loci, should be larger than 200, default=2000
Flowcell=R9                             The version of flow cell, should be R9 or R10, default=R9
Strategy=Direct                         The sequencing strategy, should be Amplicon or Direct, default=Direct
MRPPthres=0.001                         The threshold of MRPP A, default=0.001
HomoQthres=23                           The threshold of homo-dels, default=23
OtherQthres=20.6                        The threshold of other-dels, default=20.6

(2) Edit config.yaml

Users could also define customized input paramaters by editing config.yaml.

Bam: "/public/data1/yefq/data/fast5/20220703_WGA_twist/processed/20230426_Guppy621_comparison/Sce20_guppy_sup_aligned.softclip_trimmed.endtrim10_minimap2_align.mapped.sorted.bam"
Ref: "Zymo_Saccharomyces_cerevisiae_Seq5_ref"
Num: "5"
Tombo_dir: "/public/data1/yefq/data/fast5/20220703_WGA_twist/processed/20230426_guppy_sup_basecalled/Sce20/workspace/fast5_pass_single/all_single_fast5s"
Subsample: "2000"
Vcf: "/public/data1/yefq/data/fast5/20220703_WGA_twist/processed/20230426_Guppy621_comparison/Sce20_guppy_sup_aligned.test.vcf" 
Refseq: "/public/data1/yefq/data/Refs/Zymo_Saccharomyces_cerevisiae_Seq5_ref.fa" 
Outdir: "/public/data1/yefq/data/fast5/20220703_WGA_twist/processed/20230426_Guppy621_comparison/snakemake-tutorial/data/test" 
Flowcell: "R9"
Strategy: "Direct"
MRPPthres: "0.001"
HomoQthres: "23"
OtherQthres: "20.6"

Users should note that, config values can be overwritten via the command line even when it has deen defined in the config.yaml.

Start a run

Once the work directory and configuration files are set up, users can run the pipeline as easy as invoking:

cd /path/to/Delter/working/directory
conda activate Delter
snakemake -s Delter.py --cores 8

Other Snakemake-related parameters like --cores and --configfile could be checked via

snakemake -h

Output

There are several outputs according to the indexes used.

(1) target.upstream_downstream.bases.comparison.result.txt. When the workflow used MRPP A, the main output is target.upstream_downstream.bases.comparison.result.txt, which contains (1) the variation locus position, (2) group1 (plus.match or minus.match, corresponding to forward-aligned reads supporting the reference allele and reverse-aligned reads supporting the reference allele), (3) group2 (plus.del or minus.del, corresponding to forward-aligned reads supporting the non-reference allele and reverse-aligned reads supporting the non-reference allele), (4) the number of reads supporting group1 (should always be around or higher than 400 in direct sequencing), (5) the number of reads supporting group2 (should always be around or higher than 400 in direct sequencing), (6) the mean current measurements of upstream and downstream config["Num"] bases centered around variation locus of group1, (7) the mean current measurements of upstream and downstream config["Num"] bases centered around variation locus of group2, (8) P values between current measurements of group1 and group2, (9) MRPP P values, (10) MRPP A statistic, users could compare this value against the pre-set threshold (WTA/Amplicon sequencing: 0.01; direct sequencing: 0.001) in our article to decide whether the variation locus is artificial.

(2) fq.Qscore.info.txt. For R9 or R10 data, when sequencing depth is low, Q score might be used to identify artificial deletions. The main output is fq.Qscore.info.txt, which contains (1) the variation locus position, (2) group1 (corresponding to forward-aligned reads supporting the non-reference allele and reverse-aligned reads supporting the non-reference allele), (3) group2 (plus.match or minus.match, corresponding to corresponding to forward-aligned reads supporting the reference allele and reverse-aligned reads supporting the reference allele), (4) the number of reads supporting group1 (should always be ≥20), (5) the number of reads supporting group2 (should always be ≥20), (6) the mean Q scores of upstream and downstream config["Num"] bases centered around variation locus of group1, users could compare this value against the pre-set threshold in our article to decide whether the variation locus is artificial, (7) the mean Q scores of upstream and downstream config["Num"] bases centered around variation locus of group2, (8) the P values between group1 and group2.

(3) variant.info.txt. This file stores basic information of each variation output by VCF, which is used by the workflow. The 2th-10th columns are identical to VCF. Users should note that DP4 could be lower than DP, and choosing to use MRPP A or Q score mainly depends on DP4 field. The 12th-17th columns represent the location of deletion (homo or non-homo), the 1-based strating position, the 0-based starting position of homo or non-homo region, the 0-based ending position of homo or non-homo region, the deletion length output by Variation caller, and the length of homo or non-homo region (= 15th-14th+1). Our workflow use a strict criteria to extract reads with and without deletions. For example, if a deletion lacks 3 bases relative to the reference, then reads supporting the non-reference allele should only contain 3-base deletions. Therefore, some deletions may be omitted due to undesirable read numbers.

(4) MRPP.filtered.txt or Qscore.filtered.txt. Users should note that before comparing the results to pre-set thresholds, they are strongly recommended to filter the 4th and 5th columns in the target.upstream_downstream.bases.comparison.result.txt or/and fq.Qscore.info.txt according to our article, or the result may be biased due to low sequencing depth. We have provided two accessory scripts bundled in the workflow to pre-filter the results based on sequencing depth and then to compare with user-defined threshold(s). Position(s) with flags marked as "FP" are predicted artificial deletions. If neither of the above two files are generated, it indicates that none of potential variations could be removed due to very low sequencing depth or very low threshold(s).