Close assembly gaps using long-reads at high accuracy.
Keywords: bioinformatics, close-assembly-gaps, cluster, daligner, damapper, docker, dub, gap-filling, genome-assembly, long-reads, pacbio, singularity, snakemake
Long sequencing reads allow increasing contiguity and completeness of fragmented, short-read based genome assemblies by closing assembly gaps, ideally at high accuracy. DENTIST is a sensitive, highly-accurate and automated pipeline method to close gaps in (short read) assemblies with long reads.
API documentation: current, v4.0.0, v3.0.0, v2.0.0
First time here? Head over to the example and make sure it works.
Make sure Mamba (a frontend for Conda) is installed on your system. You can then use DENTIST like so:
# run the whole workflow on a cluster using Conda
snakemake --configfile=snakemake.yml --use-conda -jall
snakemake --configfile=snakemake.yml --use-conda --profile=slurm
The last command is explained in more detail below in the usage section.
Note: If you do not have mamba
installed, you may need to pass
--conda-frontend=conda
to Snakemake.
Make sure Mamba (a frontend for Conda) is installed on your system. Install DENTIST and all dependencies like so:
mamba create -n dentist -c a_ludi -c bioconda dentist-core
mamba activate dentist
mamba install -c conda-forge -c bioconda snakemake
# execute the workflow
snakemake --configfile=snakemake.yml --cores=all
More details on executing DENTIST can be found in the usage section.
Download the latest pre-built binaries from the releases section
and extract the contents. The pre-built binaries are stored in a subfolder
called bin
. Here are the instructions for v4.0.0
:
# download & extract pre-built binaries
wget https://github.com/a-ludi/dentist/releases/download/v4.0.0/dentist.v4.0.0.x86_64.tar.gz
tar -xzf dentist.v4.0.0.x86_64.tar.gz
# make binaries available to your shell
cd dentist.v4.0.0.x86_64
PATH="$PWD/bin:$PATH"
# check installation with
dentist -d
# Expected output:
#
#daligner (part of `DALIGNER`; see https://github.com/thegenemyers/DALIGNER) [OK]
#damapper (part of `DAMAPPER`; see https://github.com/thegenemyers/DAMAPPER) [OK]
#DAScover (part of `DASCRUBBER`; see https://github.com/thegenemyers/DASCRUBBER) [OK]
#DASqv (part of `DASCRUBBER`; see https://github.com/thegenemyers/DASCRUBBER) [OK]
#DBdump (part of `DAZZ_DB`; see https://github.com/thegenemyers/DAZZ_DB) [OK]
#DBdust (part of `DAZZ_DB`; see https://github.com/thegenemyers/DAZZ_DB) [OK]
#DBrm (part of `DAZZ_DB`; see https://github.com/thegenemyers/DAZZ_DB) [OK]
#DBshow (part of `DAZZ_DB`; see https://github.com/thegenemyers/DAZZ_DB) [OK]
#DBsplit (part of `DAZZ_DB`; see https://github.com/thegenemyers/DAZZ_DB) [OK]
#fasta2DAM (part of `DAZZ_DB`; see https://github.com/thegenemyers/DAZZ_DB) [OK]
#fasta2DB (part of `DAZZ_DB`; see https://github.com/thegenemyers/DAZZ_DB) [OK]
#computeintrinsicqv (part of `daccord`; see https://gitlab.com/german.tischler/daccord) [OK]
#daccord (part of `daccord`; see https://gitlab.com/german.tischler/daccord) [OK]
The tarball additionally contains the Snakemake workflow, example config files and this README. In short, everything you to run DENTIST.
Remark: the Singularity container may not work properly depending on your system. (see issue #30)
Make sure Singularity is installed on your system. You can then use the container like so:
# launch an interactive shell
singularity shell docker://aludi/dentist:stable
# execute a single command inside the container
singularity exec docker://aludi/dentist:stable dentist --version
# run the whole workflow on a cluster using Singularity
snakemake --configfile=snakemake.yml --use-singularity --profile=slurm
The last command is explained in more detail below in the usage section.
dub install dentist
or
git clone --recurse-submodules https://github.com/a-ludi/dentist.git
cd dentist
dub build
The following software packages are required to run dentist
:
Manage sequences (reads and assemblies) in 4bit encoding alongside auxiliary information such as masks or QV tracks
Find significant local alignments.
Find alignment chains, i.e. sequences of significant local alignments possibly with unaligned gaps.
Discover tandem repeats.
Estimate coverage and compute QVs.
Compute reference-based consensus sequence for gap filling.
Please see their own documentation for installation instructions. Note, the
available packages on Bioconda are outdated and should not be used at the
moment but they are available using conda install -c a_ludi <dependency>
.
Please use the exact versions specified in the Conda recipe in case you experience troubles.
Before you start producing wonderful scientific results, you should skip over to the example section and try to run the small example. This will make sure your setup is working as expected.
TL;DR
wget https://github.com/a-ludi/dentist/releases/download/v4.0.0/dentist.v4.0.0.x86_64.tar.gz tar -xzf dentist.v4.0.0.x86_64.tar.gz cd dentist.v4.0.0.x86_64 # edit dentist.yml and snakemake.yml # execute with CONDA: snakemake --configfile=snakemake.yml --use-conda # execute with SINGULARITY: snakemake --configfile=snakemake.yml --use-singularity # execute with pre-built binaries: PATH="$PWD/bin:$PATH" snakemake --configfile=snakemake.yml
Install Snakemake version >=5.32.1 and prepare your working directory:
wget https://github.com/a-ludi/dentist/releases/download/v4.0.0/dentist.v4.0.0.x86_64.tar.gz
tar -xzf dentist.v4.0.0.x86_64.tar.gz
cp -r -t . \
dentist.v4.0.0.x86_64/snakemake/dentist.yml \
dentist.v4.0.0.x86_64/snakemake/Snakefile \
dentist.v4.0.0.x86_64/snakemake/snakemake.yml \
dentist.v4.0.0.x86_64/snakemake/envs \
dentist.v4.0.0.x86_64/snakemake/scripts
Next edit snakemake.yml
and dentist.yml
to fit your needs and optionally
test your configuration with
# see above for variants with pre-built binaries or Singularity
snakemake --configfile=snakemake.yml --use-conda --cores=1 -f -- validate_dentist_config
If no errors occurred the whole workflow can be executed using
# see above for variants with pre-built binaries or Singularity
snakemake --configfile=snakemake.yml --use-conda --cores=all
For small genomes of a few 100 Mbp this should run on a regular workstation.
One may use Snakemake’s --cores
to run independent jobs in parallel. Larger
data sets may require a cluster in which case you can use Snakemake’s
cloud or cluster facilities.
Please follow the setup steps from above except for the actual execution.
To make execution on a cluster easy DENTIST comes with examples files to make
Snakemake use SLURM via DRMAA, sbatch
or srun
found under
snakemake
. If your cluster does not use SLURM please modify
the profiles to suit your needs or read the documentation of
Snakemake. Another good starting point is the
Snakemake-Profiles project.
After you have selected an appropriate cluster profile, make it available to Snakemake, e.g.:
# choose appropriate file from `snakemake/profile-slurm.*.yml`
mkdir -p ~/.config/snakemake/slurm
cp ./snakemake/profile-slurm.submit-async.yml ~/.config/snakemake/slurm/config.yaml
Adjust the profile according to your cluster, e.g. you may need to specify
accounting information. Values defined in cluster.yml
can be used in the
profile as demonstrated in the examples. This file is also the place to modify
resource allocations and job names.
Now, you can execute the workflow like this:
snakemake --configfile=snakemake.yml --profile=slurm --use-conda
Snakemake will now start submitting jobs to your cluster until all the work is done. If something fails, you can execute the same command again to continue from the latest state of the workflow.
Please inspect the Snakemake workflow to get all the details. It might be
useful to execute Snakemake with the -p
switch which causes Snakemake to
print the shell commands. If you plan to write your own workflow management
for DENTIST please feel free to contact the maintainer!
Make sure you have Snakemake 5.32.1 or later installed.
You can also use the convenient Conda package to execute the rules. Just make sure you have Mamba installed.
First of all download the test data and workflow and switch to the
dentist-example
directory.
wget https://github.com/a-ludi/dentist/releases/download/v4.0.0/dentist-example.tar.gz
tar -xzf dentist-example.tar.gz
cd dentist-example
Execute the entire workflow on your local machine using all
cores:
# run the workflow
PATH="$PWD/bin:$PATH" snakemake --configfile=snakemake.yml --cores=all
# validate the files
md5sum -c checksum.md5
Execution takes approx. 7 minutes and a maximum of 1.7GB memory on my little laptop with an Intel® Core™ i5-5200U CPU @ 2.20GHz.
Make sure Mamba (a frontend for Conda) is installed on your
system. Execute the workflow without explicit installation by adding
--use-conda
to the call to Snakemake:
# run the workflow
snakemake --configfile=snakemake.yml --use-conda --cores=all
# validate the files
md5sum -c checksum.md5
Note: If you do not have mamba
installed, you may need to pass
--conda-frontend=conda
to Snakemake.
Remark: the Singularity container may not work properly depending on your system. (see issue #30)
Execute the workflow inside a convenient Singularity image by adding
--use-singularity
to the call to Snakemake:
# run the workflow
snakemake --configfile=snakemake.yml --use-singularity --cores=all
# validate the files
md5sum -c checksum.md5
Please follow the instructions “Executing on a Cluster” above.
DENTIST comprises a complex pipeline of with many options for tweaking. This section points out some important parameters and their effect on the result or performance.
The default parameters are rather conservative, i.e. they focus on correctness of the result while not sacrificing too much sensitivity.
We also provide a greedy sample configuration
(snakemake/dentist.greedy.yml
) which
focuses on sensitivity but may introduce more errors. Warning: Use with
care! Always validate the closed gaps (e.g. manual inspection).
In any case, the workflow creates an intermediate assembly
workdir/{output_assembly}-preliminary.fasta
that contains all closed gaps,
i.e. before validation. It is accompanied by an AGP and BED file. You may
inspect these file for maximum sensitivity.
While the list of all commandline parameters is a good
reference, it does not provide an overview of the important parameters.
Therefore, we provide this shorter list of important and influential
parameters. Please also consider adjusting the performance parameter in the
workflow configuration (snakemake/snakemake.yml
).
--read-coverage
: This is the preferred way of providing values to
--max-coverage-reads
, --max-improper-coverage-reads
and
--min-coverage-reads
. See below how their values are derived from
--read-coverage
.
Ideally, the user provides the haploid read coverage which, can be inferred using a histogram of the alignment coverage across the assembly. Alternatively, the average raw read coverage can be used which is the number of base pairs in the reads divided by the number of base pairs in the assembly.
--ploidy
: Combined with --read-coverage
, this parameters is the preferred
way of providing --min-coverage-reads
.
We use the Wikipedia definition of ploidy, as “the number of complete sets of chromosomes in a cell” (https://en.wikipedia.org/wiki/Ploidy)
--max-coverage-reads
, --max-improper-coverage-reads
:
These parameters are used to derive a repeat mask from the ref vs. reads
alignment. If the coverage of (improper) alignments is larger than the given
theshold it will be considered repetitive. If supplied, default values are
derived from --read-coverage
as follows:
The maximum read coverage C_max
is calculated from the global read
coverage C
(provided via –read-coverage) such that the probability of
observing more than C_max
alignments in a unique (non-repetitive) genomic
region is very small (see paper, Methods section
and Supplementary Table 2). In practice, this probability is approximated
via
C_max = floor(C / log(log(log(b * C + c) / log(a))))
where
a = 1.65
b = 0.1650612
c = 5.9354533
To further increase the sensitivity, DENTIST searches for smaller
repeat-induced local alignments. To this end, we define an alignment as
proper if there are at most 100 bp (adjustable via
–proper-alignment-allowance) of unaligned sequence on either end of the
read. All other alignments, where only a smaller substring of the read
aligns, are called improper. Improper alignments are often indicative of
repetitive regions. Therefore, DENTIST considers genomic regions, where the
number of improper read alignments is higher than a threshold to be
repetitive. By default, this threshold equals half the global
read coverage C. (see paper, Methods section).
In practice, a smoothed version of max(4, x/2)
is used to provide better
performance for very low read coverage. The maximum improper read coverage
I_max
is computed as
I_max = floor(a*x + exp(b*(c - x)))
where
a = 0.5
b = 0.1875
c = 8
--dust-{reads,ref}
, --daligner-{consensus,reads-vs-reads,self}
,
--damapper-ref-vs-reads
, --datander-ref
, --daccord
:
These options allow passing parameters to the respective tools. They may have
dramatic influence on the result. The default settings work well for PacBio
CLR reads and should also work well with raw Nanopore data.
In-depth discussion of each tool goes beyond the scope of this document, please refer to the respective documentations (DBdust, daligner, damapper, datander, daccord).
--max-insertion-error
: Strong influence on quality and sensitivity. Lower
values lead to lower sensitivity but higher quality. The maximum recommended
value is 0.05
.
--min-anchor-length
: Higher values results in higher accuracy but lower
sensitivity. Especially, large gaps cannot be closed if the value is too
high. Usually the value should be at least 500
and up to 10_000
.
--min-reads-per-pile-up
: Choosing higher values for the minimum number of
reads drastically reduces sensitivity but has little effect on the quality.
Small values may be chosen to get the maximum sensitivity in de novo
assemblies. Make sure to throughly validate the results though.
--min-spanning-reads
: Higher values give more confidence on the
correctness of closed gaps but reduce sensitivity. The value must be well
below the expected coverage.
--allow-single-reads
: May be used under careful consideration in
combination with --min-spanning-reads=1
. This is intended for one of the
following scenarios:
--existing-gap-bonus
: If DENTIST finds evidence to join two contigs that
are already consecutive in the input assembly (i.e. joined by N
s) then it
will preferred over conflicting joins (if present) with this bonus. The
default value is rather conservative, i.e. the preferred join almost always
wins over other joins in case of a conflict.
--join-policy
: Choose according to your needs:
scaffoldGaps
: Closes only gaps that are marked by N
s in the
assembly. This is the default mode of operation. Use this if you do not
want to alter the scaffolding of the assembly. See also
--existing-gap-bonus
.scaffolds
: Allows whole scaffolds to be joined in addition to the
effects of scaffoldGaps
. Use this if you have (many) scaffolds that
are not yet full chromosome-scale.contigs
: Allows contigs to be rearranged freely. This is especially
useful in de novo assemblies before applying any other scaffolding
methods as it increases the contiguity thus increasing the chance that
large-scale scaffolding (e.g. Bionano or Hi-C) finds proper joins.--min-coverage-reads
, --min-spanning-reads
, --region-context
:
DENTIST validates closed gaps by mapping the reads to the gap-closed
assembly. It requires for each gap and the base pairs down- and upstream
(--region-context
) are (1) covered by at least --min-coverage-reads
reads
at every position and (2) are spanned by at least --min-spanning-reads
reads. Thus, increasing any of these numbers makes the valid gaps more
robust but may reduce their number.
If --min-coverage-reads
is not provided, it will be derived from
--read-coverage
(see above) and --ploidy
. Given (haploid) read coverage
C
and ploidy p
, the minimum read coverage C_min
is calculated as
C_min = C / (2 * p)
This corresponds to 50% of the long read coverage expected to be sequenced from a haploid locus (see paper, Methods section).
In the examples PacBio long reads are assumed but DENTIST can be run using any
kind of long reads. Currently, this is either PacBio or Oxford Nanopore reads.
For using none-PacBio reads, the reads_type
in snakemake.yml
must be set
to anything other than PACBIO_SMRT
. The recommendation is to use
OXFORD_NANOPORE
for Oxford Nanopore. These names are borrowed from the NCBI.
Further details on the rationale can found in this issue.
Cluster job schedulers can become unresponsive or even crash if too many jobs with short running time are submitted to the cluster. It is therefore advisable to adjust the workflow accordingly. We tried to provide a default configuration that works in most cases as is but the application scenarios can be very diverse and manual adjustments may become necessary. Here is a small guide which config parameters influence the number of jobs and how much resources they consume.
threads_per_process
: Sets the maximum number of threads/cores a single job
may use. A single-threaded job will always allocate a single core but
thread-parallel steps, e.g. the sequence alignments, will use up to
threads_per_process
if snakemake has been provided enough cores
via --cores
.-s<block_size:uint>
: The assembly and reads FAST/A files are converted into
Dazzler DBs. These DBs store the sequence in a 2-bit encoding and have
additional features like tracks (similar to BED files). Also they are split
into blocks of <block_size>
Mb. Alignments are calculated on the basis of
these blocks which enables easy distribution onto the cluster. The larger the
block size the longer are the alignment jobs and the more memory they require
but also the number of jobs is reduced. Experience shows that the block size
should be between 200Mb and 500Mb.propagate_batch_size
: The repeat masks are homogenized by propagating them
from the assembly to the reads and back again. Usually these jobs are very
short because the propagation is parallelized over the blocks of the reads
DB. To reduce the number of jobs both propagation directions are grouped
together and submitted in batches of propagate_batch_size
read blocks.
Increasing propagate_batch_size
reduces the number of submitted jobs and
increases the run time per job. It has no effect on the memory requirements.batch_size
: In the collect
step DENTIST identifies candidates for gap
closing each consisting of a pile up of reads. From these pile ups
consensus sequences are computed and validated in the process
step. Each
job process batch_size
pile ups. Increasing batch_size
reduces the
number of submitted jobs and increases the run time per job. It has no
effect on the memory requirements.validation_blocks
: The preliminarily closed gaps are validated by analyzing
how the reads align to each closed gap. The validation is conducted in
independent jobs for validation_blocks
many blocks of the gap-closed
assembly. Decreasing validation_blocks
reduces the number of submitted
jobs and increases the run time and memory requirements per job. The memory
requirement is proportional to the size of the read alignment blocks.ProtectedOutputException
Snakemake has a built-in facility to protect files from
accidental overwrites. This is meant to avoid overwriting precious results
that took many CPU hours to produce. If executing a rule would overwrite a
protected file, Snakemake raises a ProtectedOutputException
, e.g.:
ProtectedOutputException in line 1236 of /tmp/dentist-example/Snakefile:
Write-protected output files for rule collect:
workdir/pile-ups.db
File "/usr/lib/python3.9/site-packages/snakemake/executors/__init__.py", line 136, in run_jobs
File "/usr/lib/python3.9/site-packages/snakemake/executors/__init__.py", line 441, in run
File "/usr/lib/python3.9/site-packages/snakemake/executors/__init__.py", line 230, in _run
File "/usr/lib/python3.9/site-packages/snakemake/executors/__init__.py", line 155, in _run
Here workdir/pile-ups.db
is the protected file that caused the error. If you
are sure of what you are doing, you can simply raise the protection by chmod
-R +w ./workdir
and execute Snakemake again. Now, it will overwrite any files.
If you have no internet connection on your compute nodes or even the cluster
head node and want to use Singularity for execution, you will need to download
the container image manually and put it to a location accessible by all jobs.
Assume /path/to/dir
is such a location on your cluster. Then download the
container image using
# IF internet connection on head node
singularity pull --dir /path/to/dir docker://aludi/dentist:stable
# ELSE (on local machine)
singularity pull docker://aludi/dentist:stable
# copy dentist_stable.sif to cluster
scp dentist_stable.sif cluster:/path/to/dir/dentist_stable.sif
When the image is in place you will need to adjust your configuration in
snakemake.yml
:
dentist_container: "/path/to/dir/dentist_stable.sif"
Now, you are ready for execution.
Note, if you want to use Conda without internet connection, you can just use the
pre-compiled binaries instead because they are just what Conda will install.
Be sure to adjust your PATH
accordingly, e.g.:
PATH="$PWD/bin:$PATH" snakemake --configfile=snakemake.yml --profile=slurm
DBshow -n
This error message may appear in DENTIST’s log files. It is a known bug that
will be fixed in a future release. In the meantime avoid FASTA headers that
contain a literal " :: "
.
Arne Ludwig, Martin Pippel, Gene Myers, Michael Hiller. DENTIST — using long reads for closing assembly gaps at high accuracy. GigaScience, Volume 11, 2022, giab100. https://doi.org/10.1093/gigascience/giab100
DENTIST is being developed by Arne Ludwig <ludwig@mpi-cbg.de> at the Planck Institute of Molecular Cell Biology and Genetics, Dresden, Germany.
Contributions are warmly welcome. Just create an issue or pull request on GitHub. If you submit a pull request please make sure that:
dub test
runs successfully.It is recommended to install the Git hooks included in the repository to avoid premature pull requests. You can enable all shipped hooks with this command:
git config --local core.hooksPath .githooks/
If you do not want to enable just a subset use
ln -s .githooks/{hook} .git/hooks
. If you want to audit code changes before
they get executed on your machine you can you cp .githooks/{hook} .git/hooks
instead.
This project is licensed under MIT License (see LICENSE).