Methods

Multi-sample transcriptome assembly protocol

1. Sequence reads from each dataset are aligned to the reference genome using STAR, HISAT2, or another RNA-Seq aligner
2. Read alignments (SAM/BAM/CRAM format) from each dataset are assembled into transfrags using Cufflinks, StringTie, or another ab initio transcriptome assembly tool.
3. Multi-sample assembly is performed on the collection of ab initio assemblies (GTF format) by TACO, resulting in a consensus merged transcriptome
4. Downstream analysis of gene expression, conservation, variation, and functionality can be performed

Results

Varying the number of input assemblies

Performance metrics for TACO, Cuffmerge, and StringTie when merging different numbers of input assemblies. Recall (i.e., sensitivity), precision and the F-measure for all three tools were assessed for splicing patterns, splice junctions, and bases. Twenty batches were randomly selected at each number of assemblies. Points represent the mean statistic across the 20 runs, error bars represent the 95% confidence interval.

Varying isoform fraction cutoff

The parameter --isoform-frac FRAC configures TACO to only report isoforms with abundance ratio more than FRAC of the most abundant transcript in the locus

The precision-recall (PR) plots depict the performance for the three merging tools at 50 different isoform fraction cutoffs ranging from 0.001-0.999. Statistics describe the merger of the 55 breast cancer cell lines in the CCLE.

Performance for highly expressed transcripts

The precision-recall (PR) plots depict the performance for the three merging tools across subsets of the top 500, 1k, 1.5k, 2k, 3k, 5k, 7k, 10k, 12k, 15k, 18k, 20k, 22k, 25k, 27k, and 30k highest expressed transcripts in the merger of 55 breast cancer cell lines from the CCLE

Fraction of read-through transcripts

Scatterplot depicting the fraction of read-through genes produced in the multi-sample assembly from the three merge tools. Gene designated as being a read-through if it contains at least one transcript isoform bearing exonic overlap with two separate protein-coding genes on the same strand. Fraction of read-throughs was determined for assemblies at 50 different isoform fraction cutoffs ranging from 0.001-0.999.

Download Code

Releases

Source Code

The TACO GitHub repository is located here.

Build Code

More information on the build procedure is located here.

Test Installation

Instructions on testing the installation are here.

Download References

For your convenience we are providing links for human references. Be sure to gunzip the file before running with TACO or TACO-refcomp.

• GENCODEv25 Human Transcriptome GTF
• GENCODEv25 Human Transcriptome Metadata
• GRCh38 Genome Fasta (Used for CPAT)

Test Installation

Unpack the downloaded binary package:

$ tar -xpvfz taco-<RELEASE>.tar.gz

$ cd taco-<RELEASE>/
      $ chmod +x taco_run

To test installatoin, download the following two test data files and place them in the same directory as the TACO executable:

• taco_test.gtf
• taco_test_files.txt

In the directory where you placed the two downloaded files and unpacked the binary file, type:

./taco_run --gtf-expr-attr expr taco_test_files.txt

Note: Our sample dataset has already been processed by TACO, which converts the attribute specified by ‑‑gtf-expr-attr (e.g. FPKM) into expr. Hence, we use the argument ‑‑gtf-expr-attr expr here.

You should see the following output:

      2016-03-30 14:54:02,097 pid=89792 INFO - taco version 0.5.0
      2016-03-30 14:54:02,097 pid=89792 INFO - ------------------------------------------------------------------------------
      2016-03-30 14:54:02,097 pid=89792 INFO - verbose logging:                   False
      2016-03-30 14:54:02,097 pid=89792 INFO - num processes:                     1
      2016-03-30 14:54:02,097 pid=89792 INFO - output directory:                  taco
      2016-03-30 14:54:02,097 pid=89792 INFO - min transfrag length:              200
      2016-03-30 14:54:02,098 pid=89792 INFO - min expression:                    0.5
      2016-03-30 14:54:02,098 pid=89792 INFO - reference GTF file:                None
      2016-03-30 14:54:02,098 pid=89792 INFO - guided assembly mode:              False
      2016-03-30 14:54:02,098 pid=89792 INFO - guided strand mode:                False
      2016-03-30 14:54:02,098 pid=89792 INFO - guided ends mode:                  False
      2016-03-30 14:54:02,098 pid=89792 INFO - GTF expression attribute:          expr
      2016-03-30 14:54:02,098 pid=89792 INFO - isoform fraction:                  0.05
      2016-03-30 14:54:02,098 pid=89792 INFO - max_isoforms:                      0
      2016-03-30 14:54:02,098 pid=89792 INFO - change point:                      True
      2016-03-30 14:54:02,098 pid=89792 INFO - change point pvalue:               0.001
      2016-03-30 14:54:02,098 pid=89792 INFO - change point fold change:          0.85
      2016-03-30 14:54:02,098 pid=89792 INFO - change point trim:                 True
      2016-03-30 14:54:02,098 pid=89792 INFO - path graph loss threshold:         0.1
      2016-03-30 14:54:02,098 pid=89792 INFO - path frac:                         0.0
      2016-03-30 14:54:02,099 pid=89792 INFO - max paths:                         0
      2016-03-30 14:54:02,099 pid=89792 INFO - Samples: 1
      2016-03-30 14:54:02,099 pid=89792 INFO - Aggregating GTF files
      2016-03-30 14:54:02,210 pid=89792 INFO - Sorting GTF
      2016-03-30 14:54:02,236 pid=89792 INFO - Indexing Loci
      2016-03-30 14:54:02,239 pid=89792 INFO - Assembling GTF files
      2016-03-30 14:54:02,239 pid=89792 INFO - Assembling in parallel using 1 processes
      2016-03-30 14:54:02,729 pid=89792 INFO - Merging output files
      2016-03-30 14:54:02,739 pid=89792 INFO - Removing temporary files
      2016-03-30 14:54:02,740 pid=89792 INFO - Done
      

Build Instructions TACO is written in Python with C extensions for performance critical code

Required Dependencies:

• Python 2.7
• Numpy 1.9.2+
• Cython 0.23.4+

Optional Dependencies (to generate single file binaries):

• PyInstaller 3.0+ (3.2.1+ recommended)

Building TACO on your system:

Source files can be obtained from download section. Unpack the release file (.tar.gz or .zip) into source directory taco-X.Y.Z. For example:

$ tar -zxvf taco-X.Y.Z.tar.gz

If pyinstaller is installed, the build.sh bash script can be used to compile, install, and generate executable binary files. (note: it might be necessay to chmod +x build.sh to enable the build file to be executed). e.g.:

$ cd taco-<RELEASE>
$ ./build.sh

To install TACO and run without creating an executable, run the following:

$ cd taco-<RELEASE>
$ python setup.py build

$ cd build/lib*/taco
$ python taco_run.py -h
$ python taco_refcomp.py -h

$ export PYTHONPATH=<insatll_dir>/build/lib*/taco:$PYTHONPATH

Manual

$ ./taco_run <options> <gtf_files.txt>

/path/to/my/assemblies/mcf7.gtf
            /path/to/my/assemblies/bt474.gtf
            /path/to/my/assemblies/hela.gtf

$ ./taco_refcomp -o <output_directory> -r <reference_gtf> -t <test_gtf> --cpat (optional flag to run coding potential prediction)
                      

FAQ

Q. Can you compare a TACO generated assembly to an annotated reference??

A. We provide a reference comparison tool (taco_refcomp) that can be found bundled with the taco tool. This tool can be used to compare any transcriptome to a reference. The tool also supports running the CPAT coding potential tool. If CPAT is used, please cite the CPAT manuscript.

Q. How much memory does TACO require?

A. TACO memory usage scales with the size of the input and the number of parallel processes. Specifically, TACO partitions the input transfrags into independent loci, and transfrags within each loci are stored in memory. Running TACO with multiple processes allows it to assemble multiple loci in parallel at the expense of additional memory overhead. To examine peak memory usage, we tested TACO on >10,000 datsets. Some of the largest loci contained over 10 million transfrags each, and memory per process ranged from 3-10GB.

Q. How much time does TACO take to run?

A. TACO runtime also scales with the size of the input and the number of parallel processes. We can offer several examples: 1) TACO processed 55 breast cancer cell line datasets using a Mac OSX laptop computer with 4 cores in about 10 minutes, and 2) TACO processed ~1000 datasets from the CCLE on a 64-core Linux workstation in about 2 hours and 30 minutes.

Q. Why do I receive a WARNING message that my sort command does not support the --parallel flag?

A. TACO uses the operating system's command line sort to concatenate and sort the input GTF files before beginning assembly. For large jobs (~1000 samples), sorting the GTF can consume a significant portion of total runtime. Newer versions of GNU coreutils provide a parallelized version of sort. TACO checks to see if sort can be parallelized and defaults to single-threaded sort if this is not supported.

Contact

Please send any questions/comments regarding tool to yniknafs@med.umich.edu.

If using TACO, please cite Niknafs et al., Nature Methods, 2017.

Niknafs, Y. S., Pandian, B., Iyer, H. K., Chinnaiyan, A. M., & Iyer, M. K. (2017).
TACO produces robust multisample transcriptome assemblies from RNA-seq. Nat Meth, 14(1), 68–70.
              

Transcriptome Assemblies

Cufflinks ab initio assemblies of the RNA-Seq datasets from the Cancer Cell Line Encyclopedia (CCLE) are provided for user download. RNA-Seq data were aligned using STAR prior to running Cufflinks.

Batches

Run instructions

First, untar/uncompress the downloaded dataset file:

tar -zxvf ccle_cufflinks.tar.gz

Files will uncompress into a directory e.g. ccle_cufflinks. Within this directory is the file list_of_files.txt which is a list relative paths to the GTF files.

Change your directory to uncompressed directory:

cd ccle_cufflinks

Run TACO with the following command:

./taco_run --gtf-expr-attr FPKM -o my_output_dir list_of_files.txt

Individual assemblies

References

MIT License

Copyright (c) 2015-2018 Matthew Iyer, Yashar Niknafs, Balaji Pandian,
Chinnaiyan Lab, University of Michigan Medical School.

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.