Ultraplex is an all-in-one software package for processing and demultiplexing fastq files. It performs the following processing steps:
-
Removes poor quality bases (normally from the 3' end only)
-
Removes sequencing adaptors (eg Illumina universal sequencing adaptor)
-
Moves unique molecular identifiers (UMIs) to the read header for downstream deduplication
-
Detects 5’ and (optionally) 3’ barcodes for (combinatorial) demultiplexing
-
Writes out files for each barcode (or barcode combination)
Ultraplex was designed with speed and ease of use in mind. It is capable of demultiplexing an entire HiSeq lane, consisting of ~400 million reads, in just 20 minutes. It relies heavily on code from the excellent Cutadapt tool.
Ultraplex is primarily designed for the demultiplexing of sequencing data generated using in-house library preparation protocols, with custom adaptors (for example iCLIP libraries). If instead you are using a commercial library prep kit (eg Illumina Truseq or Lexogen Quantseq) then in all likelihood the sequencing facility will already have demultiplexed the files for you, so Ultraplex is probably not for you. Furthermore, if you are doing single cell RNA-seq or using, for example, Oxford Nanopore long-read sequencing, there are other softwares specifically designed for these purposes.
Build an ultraplex environment straight from our yaml:
conda env create -f environment.yml
Or add ultraplex to your existing environment (note python >= 3.7 is required)
conda install -c conda-forge -c bioconda ultraplex
Ultraplex is also available from python package manager pip.
pip install ultraplex
Ultraplex requires two inputs: a gzip-compressed fastq file, and a comma-separated file (csv) containing all the barcodes you have used. There are also multiple optional inputs, which are detailed below.
The simplest usage of Ultraplex is as follows:
ultraplex -i your_fastq_file.fastq.gz -b your_barcode_csv.csv
The fastq file should be a 4-line per read fastq. It should be in gzipped format. If your fastq is uncompressed, or compressed in a different format, convert it to gzipped format or Ultraplex will not work.
For the barcode csv, the first column contains a list of all the 5’ barcodes, and (optionally) the other column(s) contain the 3’ barcode(s) linked to that specific 5’ barcode. For example, the following csv has three 5’ barcodes, the second of which is linked to two 3’ barcodes:
NNNATGNN,
NNNCCGNN,ATG,TCA
NNNCACNN,
There are certain constraints on the barcode sequences that can be used. It is required that the length and position of the non-N nucleotides in the barcodes be consistent for all 5’ barcodes, and all 3’ barcodes linked to that specific 5’ barcode if used. 3’ barcodes linked to different 5’ barcodes do not have to be consistent with each other. 5’ barcodes do not have to be consistent with 3’ barcodes. Barcodes can have different numbers of Ns, provided that the non-N characters are all consistent.
For example, the 5’ barcodes...
NNN ATG NN
NNN CCG NNN
NNN ATG
...are all consistent because the non-N characters are all in positions 4-6 relative to the 5’ end of the read. However…
NN ATG NN
NNN ATGC NN
... are not consistent with the first three barcodes (or each other) because, relative to the 5’ end the barcodes are in positions 3-5 and 4-7 respectively.
The rules governing 3’ barcodes are the same, except that the positions are defined relative to the 3’ end of the read. For example...
NN ATG NNN
N CCG NNN
...are consistent (positions -6 to -4), but...
NN ATG NN
...is not (positions -5 to -3).
N characters denote positions which contain randomers. These are of no use for demultiplexing; instead, they allow the removal of PCR duplicates further downstream (for example by UMI-Tools). The bases detected in positions corresponding to the “N”s in the barcode are removed from the read and placed in the read header, after “rbc:” (which stands for “random barcode”). Note that if Ultraplex is run sequentially on the same fastqs (for particularly complex demultiplexing jobs), “rbc:” will only be added to the read header once (on the first run).
We have noticed that some programs save csvs in a format that is incompatible with Ultraplex. If unexpected errors emerge, try saving the csv as a plain comma-separated file, avoiding any operating system-specific encoding.
Optionally, one can add sample names to the barcodes csv. After each barcode, add a colon, followed by the sample name. Ensure that all sample names are unique. Also ensure that if 3’ barcodes are specified, no name is specified for the 5’ barcode. Some samples can be left unnamed. For example:
NNNATGNN:sample1,
NNNCCGNN,ATG:sample2,TCA:sample3,
NNNCACNN,
...is fine, but...
NNNATGNN:sample1,
NNNCCGNN:sample2,ATG:sample3,TCA:sample4,
NNNCACNN,
…is not, because sample2 is assigned to a 5’ barcode that is linked to 3’ barcodes, which doesn’t make sense.
Temporary files will still have barcodes in the filename - only once demultiplexing and compression has finished will the files be renamed to the sample names specified.
Second input for paired end sequencing (-i2, --input_2): This allows paired end data to be demultiplexed. This is especially useful for when 3’ barcodes are used as the forward read will not detect the 3’ barcode for long reads, but the reverse read will. (Note that interleaved fastqs are not currently supported)
This allows you to specify an output directory to which the temporary and final files, and the log file, are saved. If the directory does not already exist, Ultraplex will create it automatically.
The number of threads used for multithreaded operation. Larger values result in a less-than-additive increase in speed, until the limits of your machine are reached. By default this is set to 4, however a small speed increase may be seen by setting this to 8 or 16.
This option allows the user to specify how many mismatches are permitted when detecting which 5’ barcode a read contains. If set to zero, then the 5’ barcode must match the expected barcode perfectly. By default, this value is set to one mismatch.
This option allows the user to specify how many mismatches are permitted when detecting which 3’ barcode a read contains. If set to zero, then the 3’ barcode must match the expected barcode perfectly. By default, this value is set to zero mismatches.
The 3’ adapter sequence to be removed. By default this is the Illumina universal sequencing adapter "AGATCGGAAGAGCACACGTCTGAA". For paired-end demultiplexing, this adapter only applies to the forward read (i.e. input 1, option -i)
The 3’ adapter to be removed for the reverse read (option -i2) for paired-end demultiplexing. By default this is “AGATCGGAAGAGCACACGTCTGAA”.
The minimum quality score for 3’ end trimming (this uses the Cutadapt functionality). By default this is set to 30 (0.1% error rate). However, in some circumstances (especially when 3’ adaptors are used) it may be desirable to reduce the stringency of this.
The minimum quality score used for 5’ end trimming. By default this is set to zero. It is strongly recommended that this is kept at 0, as 5’ end trimming will result in incorrect detection of 5’ barcodes.
These are two optional running modes that can increase the speed of the program. By default, Ultraplex writes compressed temporary files. Ultra mode instead writes uncompressed temporary files, then compresses at the end. Depending on the system, this may slightly increase performance, at the expense of requiring large amounts of free storage space.
Larger performance gains from ultra mode are seen when it is used in conjunction with sbatch compression mode. This mode is only compatible with high performance computing clusters which have SLURM job management. After temporary files are concatenated, they are compressed using sbatch commands, thus enabling the compression workload to be spread across multiple nodes of a supercomputing cluster. Using ultra mode and sbatch compression mode together may double overall performance. Sbatch compression mode is only relevant when using ultra mode.
NOTE - THIS HAS BEEN UPDATED. This option gives the minimum length of the read after the barcode(s) and UMI(s) have been removed. For paired end, both must be at least this length for the pair of reads to be written. It is no longer necessary to specify a minimum length of at least the length of your longest adaptor.
When demultiplexing using 3’ adapters using the forward read (i.e. when single-end sequencing has been used), it is possible to obtain spurious barcode matches from the sequence at the 3’ end of the read if this is not the true end of the insert (i.e. it is not adjacent to the sequencing adaptor). To prevent this, this option ensures that at least N bases of the sequencing adapter (specified by option -a) have been detected and trimmed. By default its value is 3.
A prefix that is added to the output files to help identify which sequencing run they are derived from. By default this is “demux”.
When Ultraplex is run, it first estimates how much free space is needed for the job to complete, and then checks whether there is enough space. If there is not, it will not run. This option overrides this behaviour, allowing Ultraplex to run even if it does not think there is enough free storage space.
When using very long barcodes, reference building can take a long time. Select this option to skip the reference building step, which for long barcodes will be faster overall.
This option keeps barcodes and UMIs in the read after demultiplexing - note that UMIs will still be added to the read header.
This option does not write out reads for which there was no barcode match, which may save time.
This is when you only have 3' barcodes. In this case, write each barcode on a new line in the barcode csv. Feel free to add names too like with 5' barcodes. The format of the barcode CSV should be exactly the same as if you only had 5' barcodes (see above). Barcode sequences should be as they would be in read 1 (just as above for 3' barcodes). Ultraplex uses the read 2 for demultiplexing here, and handles all reverse complementing internally.
This is used solely in conjunction with the option --three_prime_barcode. It is solely for TSOs that have a p5 sequence (i.e. give rise to the read 1). Give a sequence of Ns followed by Is. Ns are moved to the UMI. Is are trimmed from the read but not moved to the UMI. For example, if using a TSO of sequence (p5)NNNNNrGrGrG, you would specify --tso_seq NNNNNIII. This would move the five random bases to the UMI, but the three non-random bases from the rGs would be trimmed and ignored.
Ultraplex was primarily designed for small barcodes. To improve speed during demultiplexing, it builds a 'reference' dictionary of sequence-to-barcode matches prior to demultiplexing. For long barcodes, this can result in a lengthy wait before demultiplexing and extremely high RAM usage. It is therefore recommended to using the --dont_build_reference command when using barcodes longer than 8 nt.