gerbenvoshol / lesv

High accuracy structural variations caller for Nanopore noisy long reads

lesv

lesv is a high accuracy structural variantions caller for Nanopore noisy long reads. For a 45x human raw read dataset, lesv achieves 93%, 91% and 92% of precesion, recall and f1 metrics, respectively. For a 40x human ultra-long raw read >100k dataset, lesv achieves 94%, 97%, and 95% of precision, recall and f1 metrics, respectively. Details can be found in the Benchmarks below.

lesv currently only supports Deletions and Insertions Structural Variants. More types of SVs will be implemented subsequently.

Installation

Installation and usage of lesv can be found here.

Benchmarks

Experimental Setup

Our experiments are all conducted on a computer with 48 2.50GHz CPUs and 512GB memory. The operating system is Ubuntu 16.04.

Datasets

We carry out benchmarks on two datasets.

• The first dataset is generated using our in-house sequencing and will be released soon. We call it HG002-normal-45x.

• The second dataset consists of 40x 100kb+ ultra-long raw reads. It is extracted from the publically released data HG002_ucsc_ONT_lt100kb.fastq.gz We call itHG002-100k-45x.

Statistical Information of HG002-normal-45x

sequences:              10,546,295
residues:               134,084,883,764
max:                    69,999
min:                    1
avg:                    12,713
median:                 10,663

(N10, L10):             42,081    266,833   
(N20, L20):             33,018    629,889   
(N25, L25):             29,853    843,677   
(N30, L30):             27,150    1,079,309 
(N40, L40):             22,679    1,620,717 
(N50, L50):             18,994    2,267,411 
(N60, L60):             15,861    3,040,812 
(N70, L70):             13,216    3,967,872 
(N75, L75):             12,055    4,499,182 
(N80, L80):             10,982    5,081,991 
(N90, L90):             8,276     6,454,807 

Statistical Information of HG002-100k-40x

sequences:              744,007
residues:               120,002,274,309
max:                    2,432,464
min:                    120,142
avg:                    161,291
median:                 140,392

(N10, L10):             299,074    29,408    
(N20, L20):             214,579    77,867    
(N25, L25):             193,630    107,388   
(N30, L30):             179,214    139,657   
(N40, L40):             160,609    210,697   
(N50, L50):             148,861    288,466   
(N60, L60):             140,430    371,568   
(N70, L70):             133,932    459,135   
(N75, L75):             131,128    504,417   
(N80, L80):             128,576    550,634   
(N90, L90):             124,057    645,684  

Sequencing Error Distribution

Error Rate	HG002-normal-45x	HG002-100k-40x
1%	0.23%	0.59%
2%	0.21%	0.39%
3%	0.26%	0.73%
4%	0.52%	1.74%
5%	1.10%	3.43%
6%	2.31%	5.03
7%	3.51%	6.34%
8%	4.85%	6.17%
9%	7.00%	5.28%
10%	8.52%	4.39%
10-15%	32.28%	13.35%
15-20%	14.33%	8.02%
20-25%%	8.66%	9.39%
25-30%%	6.85%	14.37%
>30%	9.36%	20.76%

Download Reference Genome

Download hg19 reference with decoys and map non-ACGT characters to N following the same way as PacBio CCS 15kb dataset.

curl -s ftp://ftp-trace.ncbi.nih.gov/1000genomes/ftp/technical/reference/phase2_reference_assembly_sequence/hs37d5.fa.gz > ref/human_hs37d5.fasta.gz
gunzip ref/human_hs37d5.fasta.gz
sed -i '/^[^>]/ y/BDEFHIJKLMNOPQRSUVWXYZbdefhijklmnopqrsuvwxyz/NNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNN/' ref/human_hs37d5.fasta

We further extract the 24 chrosomes from human_hs37d5.fasta and store them in hg19_sv_ref.fasta.

Download the Repeat Annotations

curl -s https://raw.githubusercontent.com/PacificBiosciences/pbsv/master/annotations/human_hs37d5.trf.bed > ref/human_hs37d5.trf.bed

Download the giab SV results

FTPDIR=ftp://ftp-trace.ncbi.nlm.nih.gov/giab/ftp/data/AshkenazimTrio/analysis/
curl -s ${FTPDIR}/NIST_SVs_Integration_v0.6/HG002_SVs_Tier1_v0.6.bed > giab/HG002_SVs_Tier1_v0.6.bed
curl -s ${FTPDIR}/NIST_SVs_Integration_v0.6/HG002_SVs_Tier1_v0.6.vcf.gz > giab/HG002_SVs_Tier1_v0.6.vcf.gz

Running Configurations

HG002-normal-45x configurations

PROJECT=hg002_short
RAW_READS=/data1/cy/ontsv/HG002_NA24385_Son/read_list.txt
REFERENCE=/data1/cy/hg19/hg19_sv_ref.fasta
TRF_FILE=/data1/cy/run_ccs/ref/human_hs37d5.trf.bed
THREADS=48

# split long read into short subreads
MAX_SUBSEQ_SIZE=50000
SUBSEQ_OVLP_SIZE=0
MIN_LAST_SUBSEQ_SIZE=20000

# reference mapping options
MAP_OPTIONS="-max_query_vol_res 8g -query_batch_size 2g -max_target_seqs 5 -max_hsps 10"

# sv read options
SVR_MIN_SEQ_SIZE=3000
SVR_MIN_SVE_PERC_IDENTITY=70.0
SVR_MAX_OVERHANG=300

# sv signature options
SVSIG_MIN_INDEL_SIZE=40

HG002-100k-40x configuations

PROJECT=hg002_100k
RAW_READS=/data1/cy/data/human-long-fltd.fasta.gz
REFERENCE=/data1/cy/hg19/hg19_sv_ref.fasta
TRF_FILE=/data1/cy/run_ccs/ref/human_hs37d5.trf.bed
THREADS=48

# split long read into short subreads
MAX_SUBSEQ_SIZE=50000
SUBSEQ_OVLP_SIZE=0
MIN_LAST_SUBSEQ_SIZE=20000

# reference mapping options
MAP_OPTIONS="-kmer_size 19 -kmer_window 20 -max_query_vol_res 8g -query_batch_size 2g -max_target_seqs 5 -max_hsps 10"

# sv read options
SVR_MIN_SEQ_SIZE=3000
SVR_MIN_SVE_PERC_IDENTITY=80.0
SVR_MAX_OVERHANG=300

# sv signature options
SVSIG_MIN_INDEL_SIZE=40

Experimental Results

HG002-normal-45x results

The following shell script is used for evaluating the SV results on HG002-normal-45x:

$ cat evaluate_hg002_short.sh
pbsv_vcf="/data1/cy/run_short_ont/hg002_short/7-sv-call/pbsv/hg002_short.pbsv.vcf.gz"
/data1/cy/run_ccs/truvari/truvari.py -f /data1/cy/hg19/hg19_sv_ref.fasta -b giab/HG002_SVs_Tier1_v0.6.vcf --includebed giab/HG002_SVs_Tier1_v0.6.bed -o bench-pbsv --passonly --giabreport -r 1000 -p 0.00 -c ${pbsv_vcf}

    "TP-base": 8815,
    "TP-call": 8815,
    "FP": 669,
    "FN": 826,
    "base cnt": 9641,
    "call cnt": 9484,
    "base size filtered": 6309,
    "call size filtered": 2387,
    "base gt filtered": 0,
    "call gt filtered": 0,
    "precision": 0.9294601433994095,
    "recall": 0.9143242402240431,
    "f1": 0.9218300653594771

HG002-100k-40x results

The following shell script is used for evaluating the SV results on HG002-100k-40x:

$ cat evaluate_hg002_long.sh
pbsv_vcf="/data1/cy/run_long_ont/hg002_100k/7-sv-call/pbsv/hg002_100k.pbsv.vcf.gz"
/data1/cy/run_ccs/truvari/truvari.py -f /data1/cy/hg19/hg19_sv_ref.fasta -b giab/HG002_SVs_Tier1_v0.6.vcf --includebed giab/HG002_SVs_Tier1_v0.6.bed -o bench-pbsv --passonly --giabreport -r 1000 -p 0.00 -c ${pbsv_vcf}

    "TP-base": 9340,
    "TP-call": 9340,
    "FP": 591,
    "FN": 301,
    "base cnt": 9641,
    "call cnt": 9931,
    "base size filtered": 6309,
    "call size filtered": 1962,
    "base gt filtered": 0,
    "call gt filtered": 0,
    "precision": 0.9404893766992246,
    "recall": 0.9687791722850326,
    "f1": 0.9544246883302677

PacBio CCS 15kb dataset results

We reproducing the benchmarks using the same way as in PacBio CCS 15kb dataset. We obtain a similar but slightly different metrics.

    "TP-base": 9441,
    "TP-call": 9441,
    "FP": 685,
    "FN": 200,
    "base cnt": 9641,
    "call cnt": 10126,
    "base size filtered": 6309,
    "call size filtered": 4274,
    "base gt filtered": 0,
    "call gt filtered": 0,
    "precision": 0.932352360260715,
    "recall": 0.9792552639767659,
    "f1": 0.9552284109880103