This is a compressed dictionary data structure for k-mers (strings of length k over the DNA alphabet {A,C,G,T}), based on Sparse and Skew Hashing.
The data structure is described in the following papers:
Please, cite these papers if you use SSHash.
For a dictionary of n k-mers, two basic queries are supported:
- i = Lookup(g), where i is in [0,n) if the k-mer g is found in the dictionary or i = -1 otherwise;
- g = Access(i), where g is the k-mer associated to the identifier i.
If also the weights of the k-mers (their frequency counts) are stored in the dictionary, then the dictionary is said to be weighted and it also supports:
- w = Weight(i), where i is a given k-mer identifier and w is the weight of the k-mer.
Other supported queries are:
- Membership Queries: determine if a given k-mer is present in the dictionary or not.
- Streaming Queries: stream through all k-mers of a given DNA file (.fasta or .fastq formats) to determine their membership to the dictionary.
- Navigational Queries: given a k-mer g[1..k] determine if g[2..k]+x is present (forward neighbourhood) and if x+g[1..k-1] is present (backward neighbourhood), for x = A, C, G, T ('+' here means string concatenation). SSHash internally stores a set of strings, called contigs in the following, each associated to a distinct identifier. If a contig identifier is specified for a navigational query (rather than a k-mer), then the backward neighbourhood of the first k-mer and the forward neighbourhood of the last k-mer in the contig are returned.
If you are interested in a membership-only version of SSHash, have a look at SSHash-Lite. It also works for input files with duplicate k-mers (e.g., matchtigs [4]). For a query sequence S and a given coverage threshold E in [0,1], the sequence is considered to be present in the dictionary if at least E*(|S|-k+1) of the k-mers of S are positive.
NOTE: It is assumed that two k-mers being the reverse complement of each other are the same.
The code is tested on Linux with gcc
and on Mac with clang
.
To build the code, CMake
is required.
Clone the repository with
git clone --recursive https://github.com/jermp/sshash.git
If you have cloned the repository without --recursive
, be sure you pull the dependencies with the following command before
compiling:
git submodule update --init --recursive
To compile the code for a release environment (see file CMakeLists.txt
for the used compilation flags), it is sufficient to do the following:
mkdir build
cd build
cmake ..
make -j
For a testing environment, use the following instead:
mkdir debug_build
cd debug_build
cmake .. -D CMAKE_BUILD_TYPE=Debug -D SSHASH_USE_SANITIZERS=On
make -j
SSHash uses by default the following 2-bit encoding of nucleotides.
A 65 01000.00.1 -> 00
C 67 01000.01.1 -> 01
G 71 01000.11.1 -> 11
T 84 01010.10.0 -> 10
a 97 01100.00.1 -> 00
c 99 01100.01.1 -> 01
g 103 01100.11.1 -> 11
t 116 01110.10.0 -> 10
If you want to use the "traditional" encoding
A 65 01000001 -> 00
C 67 01000011 -> 01
G 71 01000111 -> 10
T 84 01010100 -> 11
a 97 01100001 -> 00
c 99 01100011 -> 01
g 103 01100111 -> 10
t 116 01110100 -> 11
for compatibility issues with other software, then
compile SSHash with the flag -DSSHASH_USE_TRADITIONAL_NUCLEOTIDE_ENCODING=On
.
By default, SSHash uses a maximum k-mer length of 31.
If you want to support k-mer lengths up to (and including) 63,
compile the library with the flag -DSSHASH_USE_MAX_KMER_LENGTH_63=On
.
The repository has minimal dependencies: it only uses the PTHash library (for minimal perfect hashing), and zlib
to read gzip-compressed streams.
To automatically pull the PTHash dependency, just clone the repo with
--recursive
as explained in Compiling the Code.
If you do not have zlib
installed, you can do
sudo apt-get install zlib1g
if you are on Linux/Ubuntu, or
brew install zlib
if you have a Mac.
There is one executable called sshash
after the compilation, which can be used to run a tool.
Run ./sshash
as follows to see a list of available tools.
== SSHash: (S)parse and (S)kew (Hash)ing of k-mers =========================
Usage: ./sshash <tool> ...
Available tools:
build build a dictionary
query query a dictionary
check check correctness of a dictionary
bench run performance tests for a dictionary
dump write super-k-mers of a dictionary to a fasta file
permute permute a weighted input file
compute-statistics compute index statistics
For large-scale indexing, it could be necessary to increase the number of file descriptors that can be opened simultaneously:
ulimit -n 2048
For the examples, we are going to use some collections
of stitched unitigs from the directory data/unitigs_stitched
.
Important note: The value of k used during the formation of the unitigs is indicated in the name of each file and the dictionaries must be built with that value as well to ensure correctness.
For example, data/unitigs_stitched/ecoli4_k31_ust.fa.gz
indicates the value k = 31, whereas data/unitigs_stitched/se.ust.k63.fa.gz
indicates the value k = 63.
For all the examples below, we are going to use k = 31.
(The directory data/unitigs_stitched/with_weights
contains some files with k-mers' weights too.)
In the section Input Files, we explain how such collections of stitched unitigs can be obtained from raw FASTA files.
./sshash build -i ../data/unitigs_stitched/salmonella_enterica_k31_ust.fa.gz -k 31 -m 13 --check --bench -o salmonella_enterica.index
This example builds a dictionary for the k-mers read from the file ../data/unitigs_stitched/salmonella_enterica_k31_ust.fa.gz
,
with k = 31 and m = 13. It also check the correctness of the dictionary (--check
option), run a performance benchmark (--bench
option), and serializes the index on disk to the file salmonella_enterica.index
.
To run a performance benchmark after construction of the index, use:
./sshash bench -i salmonella_enterica.index
To also store the weights, use the option --weighted
:
./sshash build -i ../data/unitigs_stitched/with_weights/salmonella_enterica.ust.k31.fa.gz -k 31 -m 13 --weighted --check --verbose
./sshash build -i ../data/unitigs_stitched/salmonella_100_k31_ust.fa.gz -k 31 -m 15 -l 2 -o salmonella_100.index
This example builds a dictionary from the input file ../data/unitigs_stitched/salmonella_100_k31_ust.fa.gz
(a pangenome consisting in 100 genomes of Salmonella Enterica), with k = 31, m = 15, and l = 2. It also serializes the index on disk to the file salmonella_100.index
.
To perform some streaming membership queries, use:
./sshash query -i salmonella_100.index -q ../data/queries/SRR5833294.10K.fastq.gz
if your queries are meant to be read from a FASTQ file, or
./sshash query -i salmonella_100.index -q ../data/queries/salmonella_enterica.fasta.gz --multiline
if your queries are to be read from a (multi-line) FASTA file.
./sshash build -i ../data/unitigs_stitched/salmonella_100_k31_ust.fa.gz -k 31 -m 13 -l 4 -s 347692 --canonical-parsing -o salmonella_100.canon.index
This example builds a dictionary from the input file ../data/unitigs_stitched/salmonella_100_k31_ust.fa.gz
(same used in Example 2), with k = 31, m = 13, l = 4, using a seed 347692 for construction (-s 347692
), and with the canonical parsing modality (option --canonical-parsing
). The dictionary is serialized on disk to the file salmonella_100.canon.index
.
The "canonical" version of the dictionary offers more speed for only a little space increase (for a suitable choice of parameters m and l), especially under low-hit workloads -- when the majority of k-mers are not found in the dictionary. (For all details, refer to the paper.)
Below a comparison between the dictionary built in Example 2 (not canonical) and the one just built (Example 3, canonical).
./sshash query -i salmonella_100.index -q ../data/queries/SRR5833294.10K.fastq.gz
./sshash query -i salmonella_100.canon.index -q ../data/queries/SRR5833294.10K.fastq.gz
The canonical dictionary can be twice as fast as the regular dictionary for low-hit workloads, even on this tiny example, for only +0.4 bits/k-mer.
./sshash permute -i ../data/unitigs_stitched/with_weights/ecoli_sakai.ust.k31.fa.gz -k 31 -o ecoli_sakai.permuted.fa
This command re-orders (and possibly reverse-complement) the strings in the collection as to minimize the number of runs in the weights and, hence, optimize the encoding of the weights.
The result is saved to the file ecoli_sakai.permuted.fa
.
In this example for the E.Coli collection (Sakai strain) we reduce the number of runs in the weights from 5820 to 3723.
Then use the build
command as usual to build the permuted collection:
./sshash build -i ecoli_sakai.permuted.fa -k 31 -m 13 --weighted --verbose
The index built on the permuted collection optimizes the storage space for the weights which results in a 15.1X better space than the empirical entropy of the weights.
For reference, the index built on the original collection:
./sshash build -i ../data/unitigs_stitched/with_weights/ecoli_sakai.ust.k31.fa.gz -k 31 -m 13 --weighted --verbose
already achieves a 12.4X better space than the empirical entropy.
SSHash is meant to index k-mers from collections that do not contain duplicates nor invalid k-mers (strings containing symbols different from {A,C,G,T}). These collections can be obtained, for example, by extracting the maximal unitigs of a de Bruijn graph.
To do so, we can use the tool BCALM2. This tool builds a compacted de Bruijn graph and outputs its maximal unitigs. From the output of BCALM2, we can then stitch (i.e., glue) some unitigs to reduce the number of nucleotides. The stitiching process is carried out using the UST tool.
NOTE: Input files are expected to have one DNA sequence per line. If a sequence spans multiple lines (e.g., multi-fasta), the lines should be concatenated before indexing.
Below we provide a complete example (assuming both BCALM2 and UST are installed correctly) that downloads the Human (GRCh38) Chromosome 13 and extracts the maximal stitiched unitigs for k = 31.
mkdir DNA_datasets
wget http://ftp.ensembl.org/pub/current_fasta/homo_sapiens/dna/Homo_sapiens.GRCh38.dna.chromosome.13.fa.gz -O DNA_datasets/Homo_sapiens.GRCh38.dna.chromosome.13.fa.gz
~/bcalm/build/bcalm -in ~/DNA_datasets/Homo_sapiens.GRCh38.dna.chromosome.13.fa.gz -kmer-size 31 -abundance-min 1 -nb-cores 8
~/UST/ust -k 31 -i ~/Homo_sapiens.GRCh38.dna.chromosome.13.fa.unitigs.fa
gzip Homo_sapiens.GRCh38.dna.chromosome.13.fa.unitigs.fa.ust.fa
rm ~/Homo_sapiens.GRCh38.dna.chromosome.13.fa.unitigs.fa
The script scripts/download_and_preprocess_datasets.sh
contains all the needed steps to download and pre-process
the datasets that we used in [1].
For the experiments in [2] and [3], we used the datasets available on Zenodo.
Using the option -all-abundance-counts
of BCALM2, it is possible to also include the abundance counts of the k-mers in the BCALM2 output. Then, use the option -a 1
of UST to include such counts in the stitched unitigs.
For some example benchmarks, see the folder /benchmarks
.
Some more large-scale benchmarks below.
Pinus Taeda ("pine", GCA_000404065.3) and Ambystoma Mexicanum ("axolotl", GCA_002915635.2) are some of the largest genome assemblies, respectively counting 10,508,232,575 and 17,987,935,180 distinct k-mers for k = 31.
After running BCALM2 and UST, we build the indexes as follows.
./sshash build -i ~/DNA_datasets.larger/GCA_000404065.3_Ptaeda2.0_genomic.ust_k31.fa.gz -k 31 -m 20 -l 6 -c 7 -o pinus.m20.index
./sshash build -i ~/DNA_datasets.larger/GCA_000404065.3_Ptaeda2.0_genomic.ust_k31.fa.gz -k 31 -m 19 -l 6 -c 7 --canonical-parsing -o pinus.m19.canon.index
./sshash build -i ~/DNA_datasets.larger/GCA_002915635.3_AmbMex60DD_genomic.ust_k31.fa.gz -k 31 -m 21 -l 6 -c 7 -o axolotl.m21.index
./sshash build -i ~/DNA_datasets.larger/GCA_002915635.3_AmbMex60DD_genomic.ust_k31.fa.gz -k 31 -m 20 -l 6 -c 7 --canonical-parsing -o axolotl.m20.canon.index
The following table summarizes the space of the dictionaries.
Dictionary | Pine | Axolotl | ||
---|---|---|---|---|
GB | bits/k-mer | GB | bits/k-mer | |
SSHash, regular | 13.21 | 10.06 | 22.28 | 9.91 |
SSHash, canonical | 14.94 | 11.37 | 25.03 | 11.13 |
To query the dictionaries, we use SRR17023415 fastq reads (23,891,117 reads, each of 150 bases) for the pine, and GSM5747680 multi-line fasta (15,548,160 lines) for the axolotl.
Timings have been collected on an Intel Xeon Platinum 8276L CPU @ 2.20GHz, using a single thread.
Dictionary | Pine | Axolotl | ||
---|---|---|---|---|
(>75% hits) | (>86% hits) | |||
tot (min) | avg (ns/k-mer) | tot (min) | avg (ns/k-mer) | |
SSHash, regular | 19.2 | 400 | 4.2 | 269 |
SSHash, canonical | 14.8 | 310 | 3.2 | 208 |
Below the complete query reports.
./sshash query -i pinus.m20.index -q ~/DNA_datasets.larger/queries/SRR17023415_1.fastq.gz
==== query report:
num_kmers = 2866934040
num_valid_kmers = 2866783488 (99.9947% of kmers)
num_positive_kmers = 2151937575 (75.0645% of valid kmers)
num_searches = 418897117/2151937575 (19.466%)
num_extensions = 1733040458/2151937575 (80.534%)
elapsed = 1146.58 sec / 19.1097 min / 399.933 ns/kmer
./sshash query -i pinus.m19.canon.index -q ~/DNA_datasets.larger/queries/SRR17023415_1.fastq.gz
==== query report:
num_kmers = 2866934040
num_valid_kmers = 2866783488 (99.9947% of kmers)
num_positive_kmers = 2151937575 (75.0645% of valid kmers)
num_searches = 359426304/2151937575 (16.7025%)
num_extensions = 1792511271/2151937575 (83.2975%)
elapsed = 889.779 sec / 14.8297 min / 310.359 ns/kmer
./sshash query -i axolotl.m21.index -q ~/DNA_datasets.larger/queries/Axolotl.Trinity.CellReports2017.fasta.gz --multiline
==== query report:
num_kmers = 931366757
num_valid_kmers = 748445346 (80.3599% of kmers)
num_positive_kmers = 650467884 (86.9092% of valid kmers)
num_searches = 124008258/650467884 (19.0645%)
num_extensions = 526459626/650467884 (80.9355%)
elapsed = 250.173 sec / 4.16955 min / 268.608 ns/kmer
./sshash query -i axolotl.m20.canon.index -q ~/DNA_datasets.larger/queries/Axolotl.Trinity.CellReports2017.fasta.gz --multiline
==== query report:
num_kmers = 931366757
num_valid_kmers = 748445346 (80.3599% of kmers)
num_positive_kmers = 650467884 (86.9092% of valid kmers)
num_searches = 106220473/650467884 (16.3299%)
num_extensions = 544247411/650467884 (83.6701%)
elapsed = 193.871 sec / 3.23119 min / 208.158 ns/kmer
Giulio Ermanno Pibiri - giulioermanno.pibiri@unive.it
- [1] Giulio Ermanno Pibiri. Sparse and Skew Hashing of K-Mers. Bioinformatics. 2022.
- [2] Giulio Ermanno Pibiri. On Weighted K-Mer Dictionaries. International Workshop on Algorithms in Bioinformatics (WABI). 2022.
- [3] Giulio Ermanno Pibiri. On Weighted K-Mer Dictionaries. Algorithms for Molecular Biology (ALGOMB). 2023.
- [4] Schmidt, S., Khan, S., Alanko, J., Pibiri, G. E., and Tomescu, A. I. Matchtigs: minimum plain text representation of k-mer sets. Genome Biology 24, 136. 2023.