Overcoming uncollapsed haplotypes in long-read assemblies of non-model organisms

Supporting scripts for the paper "Overcoming uncollapsed haplotypes in long-read assemblies of non-model organisms" by Guiglielmoni et al., available in bioRxiv.

In this paper, we compare assemblies of PacBio and Nanopore reads with seven assemblers, and evaluate their efficiency at producing properly collapsed haploid assemblies. We also try to reduce uncollapsed haplotypes by using a subset of filtered reads, and haplotig-purging tools. We also present the tool HapPy to evaluate the correct haploid representation of a diploid genome.

Requirements

Long-read assemblers

Haplotigs-purging tools

Sampling depth subsets

BBmap

Evaluation

Other dependencies

Python packages:

Biopython

Datasets

We tested assemblies on two long-read datasets for the bdelloid rotifer Adineta vaga:

PacBio dataset (23.5 Gb, N50 = 11.6 kb) designated as pacbio_data
Nanopore dataset (17.5 Gb, N50 = 18.8 kb) designated as ont_data
Illumina dataset (11.4 Gb, 2*250 bp) designated as illumina_end1 and illumina_end2

Assembly of full datasets

PacBio reads

We ran the assemblers on the PacBio data in the following manner. Assemblies were repeated 5 times to test the robustness of the output assemblies.

Canu

canu -d corrected -p corrected genomeSize=100m useGrid=false -pacbio-raw pacbio_data
for i in {1..5}
do
    canu -d out -p out genomeSize=100m useGrid=false \
        -pacbio-corrected corrected/corrected.correctedReads.fasta
    cp out/out.contigs.fasta canu_assembly_${i}.fasta
done

Flye

for i in {1..5}
do
    flye -o out -g 100m --pacbio-raw pacbio_data
    cp out/assembly.fasta flye_assembly_${i}.fasta
done

NextDeNovo

echo pacbio_data > input.fofn
seq_stat input.fofn -g 100Mb -d 150 > stats.txt
for i in {1..5}
do
    nextDenovo run.cfg
done

Ra

for i in {1..5}
do
    ra -x pb pacbio_data > ra_assembly_${i}.fasta
done

Raven

for i in {1..5}
do
    raven pacbio_data > raven_assembly_${i}.fasta
done

Shasta

for i in {1..5}
do
    shasta --input pacbio_data --Reads.minReadLength 0 --assemblyDirectory out \
        --Assembly.consensusCaller Modal --Kmers.k 12
    cp out/Assembly.fasta shasta_assembly_${i}.fasta
done

wtdbg2

for i in {1..5}
do
    wtdbg2 -x rs -g 100m -i pacbio_data -fo out
    wtpoa-cns -i out.ctg.lay.gz -o out.ctg.fa
    minimap2 -x map-pb -a out.ctg.fa pacbio_data | samtools sort > out.ctg.bam
    samtools view out.ctg.bam | wtpoa-cns -d out.ctg.fa -i - -fo wtdbg2_assembly_${i}.fasta
done

Nanopore reads

We ran the assemblers on the Nanopore data in the following manner. Assemblies were repeated 5 times to test the robustness of the output assemblies.

Canu

canu -d corrected -p corrected genomeSize=100m useGrid=false -nanopore-raw ont_data
for i in {1..5}
do
    canu -d out -p out genomeSize=100m useGrid=false \
        -nanopore-corrected corrected/corrected.correctedReads.fasta
    cp out/out.contigs.fasta canu_assembly_${i}.fasta
done

Flye

for i in {1..5}
do
    flye -o out -g 100m --nano-raw ont_data
    cp out/assembly.fasta flye_assembly_${i}.fasta
done

NextDeNovo

echo ont_data > input.fofn
seq_stat input.fofn -g 100Mb -d 150 > stats.txt
for i in {1..5}
do
    nextDenovo run.cfg
done

Ra

for i in {1..5}
do
    ra -x ont ont_data > ra_assembly_${i}.fasta
done

Raven

for i in {1..5}
do
    raven ont_data > raven_assembly_${i}.fasta
done

Shasta

for i in {1..5}
do
    shasta --input ont_data --Reads.minReadLength 0 --assemblyDirectory out 
    cp out/Assembly.fasta shasta_assembly_${i}.fasta
done

wtdbg2

for i in {1..5}
do
    wtdbg2 -x ont -g 100m -i ont_data -fo out
    wtpoa-cns -i out.ctg.lay.gz -o out.ctg.fa
    minimap2 -x map-ont -a out.ctg.fa ont_data | samtools sort > out.ctg.bam
    samtools view out.ctg.bam | wtpoa-cns -d out.ctg.fa -i - -fo wtdbg2_assembly_${i}.fasta
done

Read filtering

The reads were filtered with a python script.

python3 filter_reads.py --min 15000 --fasta pacbio_data --out pacbio.min15000.fasta
python3 filter_reads.py --min 30000 --fasta ont_data --out ont.min30000.fasta

The assemblies were run on these filtered datasets in the same manner as for assemblies of the full datasets.

Haplotig purging

Assemblies were post-processed with HaploMerger2, purge_dups, purge_haplotigs (one replicate per assembler). L, M, H are selected depending on the histogram produced by purge_haplotigs hist.

PacBio assemblies

for i in canu flye ra raven shasta wtdbg2
do
    minimap2 -ax map-pb ${i}_assembly_1.fasta pb_data --secondary=no | samtools sort -o ${i}_all.ali.sorted.bam -T tmp.ali
    samtools index ${i}_all.ali.sorted.bam
    samtools faidx ${i}_assembly_1.fasta
    purge_haplotigs hist -b ${i}_all.ali.sorted.bam -g ${i}_assembly_1.fasta
    purge_haplotigs cov -i ${i}_all.ali.sorted.bam -l L -m M -h H -o ${i}_all.cov_stats.csv
    purge_haplotigs purge -g ${i}_assembly_1.fasta -c ${i}_all.cov_stats.csv -o ${i}_assembly_1.purged.fasta
done

Nanopore assemblies

for i in canu flye ra raven shasta wtdbg2
do
    minimap2 -ax map-ont ${i}_assembly_1.fasta ont_data --secondary=no | samtools sort -o ${i}_all.ali.sorted.bam -T tmp.ali
    samtools index ${i}_all.ali.sorted.bam
    samtools faidx ${i}_assembly_1.fasta
    purge_haplotigs hist -b ${i}_all.ali.sorted.bam -g ${i}_assembly_1.fasta
    purge_haplotigs cov -i ${i}_all.ali.sorted.bam -l L -m M -h H -o ${i}_all.cov_stats.csv
    purge_haplotigs purge -g ${i}_assembly_1.fasta -c ${i}_all.cov_stats.csv -o ${i}_assembly_1.purged.fasta
done

Impact of read-depth

Subsets were randomly sampled from PacBio and Nanopore datasets to reach a certain read-depth. The script reformat.sh from BBmap was used to generate these subsets. The target number of bases if based on the estimated haploid genome size of 102.3 Mb.

for i in {1..5}
do
    reformat.sh in=pacbio_data out=pacbio.subset.10X.0${i}.fasta samplebasestarget=1023000000
    reformat.sh in=pacbio_data out=pacbio.subset.20X.0${i}.fasta samplebasestarget=2046000000
    reformat.sh in=pacbio_data out=pacbio.subset.30X.0${i}.fasta samplebasestarget=3069000000
    reformat.sh in=pacbio_data out=pacbio.subset.40X.0${i}.fasta samplebasestarget=4092000000
    reformat.sh in=pacbio_data out=pacbio.subset.50X.0${i}.fasta samplebasestarget=5115000000
    reformat.sh in=pacbio_data out=pacbio.subset.60X.0${i}.fasta samplebasestarget=6138000000
    reformat.sh in=pacbio_data out=pacbio.subset.80X.0${i}.fasta samplebasestarget=8184000000
    reformat.sh in=pacbio_data out=pacbio.subset.100X.0${i}.fasta samplebasestarget=10230000000

    reformat.sh in=ont_data out=ont.subset.10X.0${i}.fasta samplebasestarget=1023000000
    reformat.sh in=ont_data out=ont.subset.20X.0${i}.fasta samplebasestarget=2046000000
    reformat.sh in=ont_data out=ont.subset.30X.0${i}.fasta samplebasestarget=3069000000
    reformat.sh in=ont_data out=ont.subset.40X.0${i}.fasta samplebasestarget=4092000000
    reformat.sh in=ont_data out=ont.subset.50X.0${i}.fasta samplebasestarget=5115000000
    reformat.sh in=ont_data out=ont.subset.60X.0${i}.fasta samplebasestarget=6138000000
    reformat.sh in=ont_data out=ont.subset.80X.0${i}.fasta samplebasestarget=8184000000
    reformat.sh in=ont_data out=ont.subset.100X.0${i}.fasta samplebasestarget=10230000000
done

The assemblies were run on these subsets in the same manner as for assemblies of the full datasets. For Canu assemblies, the parameter stopOnLowCoverage was set to 1 to allow runs on low-depth datasets.

Assembly evaluation

Basic statistics

Assembly size and N50 were computed with assembly-stats

assembly-stats assembly.fasta

BUSCO completeness

The BUSCO completeness was computed using the dataset metazoa odb10.

busco -i assembly.fasta -o busco_output -l metazoa_odb10 -m genome

k-mer completeness

The k-mer completeness is calculated against a low-error rate Illumina dataset.

kat comp -o kat_output 'illumina_end1 illumina_end2' assembly.fasta

tinycov

First we mapped the long reads to the assemblies, PacBio reads for PacBio assemblies, Nanopore reads for Nanopore assemblies.

minimap2 -ax map-pb assembly.fasta pacbio_data --secondary=no | samtools sort -o aligned.bam -T tmp.ali
samtools index mapping.bam

minimap2 -ax map-ont assembly.fasta ont_data --secondary=no | samtools sort -o aligned.bam -T tmp.ali
samtools index mapping.bam

Then the mapping data was used to study the coverage with tinycov.

tinycov covplot -r 20000 -t cov.txt aligned.bam

HapPy

The mapping data is used again for HapPy, that also relies on coverage.

main.py coverage -d happy_output aligned.bam
main.py estimate -S 102M -O happy_out/happy.stats happy_output/aligned.bam.hist

nadegeguiglielmoni/overcoming_uncollapsed_haplotypes_in_long_read_assemblies

Overcoming uncollapsed haplotypes in long-read assemblies of non-model organisms

Table of contents

Requirements

Long-read assemblers

Haplotigs-purging tools

Sampling depth subsets

Evaluation

Other dependencies

Datasets

Assembly of full datasets

PacBio reads

Canu

Flye

NextDeNovo

Ra

Raven

Shasta

wtdbg2

Nanopore reads

Canu

Flye

NextDeNovo

Ra

Raven

Shasta

wtdbg2

Read filtering

Haplotig purging

PacBio assemblies

Nanopore assemblies

Impact of read-depth

Assembly evaluation

Basic statistics

BUSCO completeness

k-mer completeness

tinycov

HapPy