McClintock: Meta-pipeline to identify transposable element insertions using next generation sequencing data
# INSTALL (Requires Conda)
git clone git@github.com:bergmanlab/mcclintock.git
cd mcclintock
conda env create -f install/envs/mcclintock.yml --name mcclintock
conda activate mcclintock
python3 mcclintock.py --install
python3 test/download_test_data.py
# RUN
python3 mcclintock.py \
-r test/sacCer2.fasta \
-c test/sac_cer_TE_seqs.fasta \
-g test/reference_TE_locations.gff \
-t test/sac_cer_te_families.tsv \
-1 test/SRR800842_1.fastq.gz \
-2 test/SRR800842_2.fastq.gz \
-p 4 \
-o /path/to/output/directory
- Getting Started
- Introduction
- Software Components
- Software Dependencies
- Installing McClintock
- McClintock Usage
- McClintock Input
- McClintock Output
- Run Examples
Many methods have been developed to detect transposable element (TE) insertions from whole genome shotgun next-generation sequencing (NGS) data, each of which has different dependencies, run interfaces, and output formats. Here, we have developed a meta-pipeline to install and run multiple methods for detecting TE insertions in NGS data, which generates output in the UCSC Browser extensible data (BED) format. A detailed description of the original McClintock pipeline and evaluation of the original six McClintock component methods on the yeast genome can be found in Nelson, Linheiro and Bergman (2017) G3 7:2763-2778.
The complete pipeline requires a fasta reference genome, a fasta consensus set of TE sequences present in the organism and fastq paired-end sequencing reads. Optionally if a detailed annotation of TE sequences in the reference genome has been performed, a GFF file with the locations of reference genome TE annotations and a tab delimited taxonomy file linking individual insertions to the TE family they belong to can be supplied (an example of this file is included in the test directory as sac_cer_te_families.tsv). If only single-end fastq sequencing data are available, then this can be supplied as option -1, however only ngs_te_mapper and RelocaTE will run as these are the only methods that handle single-ended data.
- ngs_te_mapper - Linheiro and Bergman (2012)
- RelocaTE - Robb et al. (2013)
- RelocaTE2 - Chen et al. (2017)
- TEMP - Zhuang et al. (2014)
- RetroSeq - Keane et al. (2012)
- PoPoolationTE - Kofler et al. (2012)
- PoPoolationTE2 - Kofler et al. (2016)
- TE-locate - Platzer et al. (2012)
- TEFLoN - Adrion et al. (2017)
McClintock is written in Python3 leveraging the SnakeMake workflow system and is designed to run on linux operating systems. Installation of software dependencies for McClintock is automated by Conda, thus a working installation of Conda is required to install McClintock. Conda can be installed via the Miniconda installer.
wget https://repo.continuum.io/miniconda/Miniconda3-latest-Linux-x86_64.sh -O $HOME//miniconda.sh
bash ~/miniconda.sh -b -p $HOME/miniconda # silent mode
echo "export PATH=\$PATH:\$HOME/miniconda/bin" >> $HOME/.bashrc # add to .bashrc
source $HOME/.bashrc
conda init
conda init
requires you to close and open a new terminal before it take effect
conda update conda
After installing and updating Conda, McClintock can be installed by: 1. cloning the repository, 2. creating the Conda environment, and 3. running the install script.
git clone git@github.com:bergmanlab/mcclintock.git
cd mcclintock
conda env create -f install/envs/mcclintock.yml --name mcclintock
- This installs the base dependencies (
Snakemake
,Python3
,BioPython
) needed to run the main McClintock script into the McClintock Conda environment
conda activate mcclintock
- This adds the dependencies installed in the McClintock conda environment to the environment
PATH
so that they can be used by the McClintock scripts. - This environment must always be activated prior to running any of the McClintock scripts
- NOTE: Sometimes activating conda environments does not work via
conda activate myenv
when run through a script submitted to a queueing system, this can be fixed by activating the environment in the script as shown below
CONDA_BASE=$(conda info --base)
source ${CONDA_BASE}/etc/profile.d/conda.sh
conda activate mcclintock
- For more on Conda: see the Conda User Guide
python3 mcclintock.py --install
- This command installs each of the TE insertion detection tools and installs a conda environment for each method.
##########################
## Required ##
##########################
-r REFERENCE, --reference REFERENCE
A reference genome sequence in fasta format
-c CONSENSUS, --consensus CONSENSUS
The consensus sequences of the TEs for the species in
fasta format
-1 FIRST, --first FIRST
The path of the first fastq file from paired end read
sequencing or the fastq file from single read
sequencing
##########################
## Optional ##
##########################
-h, --help show this help message and exit
-2 SECOND, --second SECOND
The path of the second fastq file from a paired end
read sequencing
-p PROC, --proc PROC The number of processors to use for parallel stages of
the pipeline [default = 1]
-o OUT, --out OUT An output folder for the run. [default = '.']
-m METHODS, --methods METHODS
A comma-delimited list containing the software you
want the pipeline to use for analysis. e.g. '-m
relocate,TEMP,ngs_te_mapper' will launch only those
three methods
-g LOCATIONS, --locations LOCATIONS
The locations of known TEs in the reference genome in
GFF 3 format. This must include a unique ID attribute
for every entry
-t TAXONOMY, --taxonomy TAXONOMY
A tab delimited file with one entry per ID in the GFF
file and two columns: the first containing the ID and
the second containing the TE family it belongs to. The
family should correspond to the names of the sequences
in the consensus fasta file
-s COVERAGE_FASTA, --coverage_fasta COVERAGE_FASTA
A fasta file that will be used for TE-based coverage
analysis, if not supplied then the consensus sequences
of the TEs will be used for the analysis
-T, --comments If this option is specified then fastq comments (e.g.
barcode) will be incorporated to SAM output. Warning:
do not use this option if the input fastq files do not
have comments
-a AUGMENT, --augment AUGMENT
A fasta file of TE sequences that will be included as
extra chromosomes in the reference file (useful if the
organism is known to have TEs that are not present in
the reference strain)
--resume This option will attempt to use existing intermediate
files from a previous McClintock run
--install This option will install the dependencies of
mcclintock
--debug This option will allow snakemake to print progress to
stdout
--make_annotations This option will only run the pipeline up to the
creation of the repeat annotations
- Available methods to use with
-m/--methods
:trimgalore
: Runs Trim Galore to QC the fastq file(s) and trim the adaptors prior to running the component methodscoverage
: Estimates copy number based on normalized coverage and creates coverage plots for each TE in the fasta provided by-c/--consensus
or-s/coverage_fasta
if providedmap_reads
: Maps the reads to the reference genome. This is useful to ensure the BAM alignment file is produced regardless if another method requires it as inputngs_te_mapper
: Runs the ngs_te_mapper component methodrelocate
: Runs the RelocaTE component methodrelocate2
: Runs the RelocaTE2 component methodtemp
: Runs the TEMP component method (Paired-End Only)retroseq
: Runs the RetroSeq component method (Paired-End Only)popoolationte
: Runs the PoPoolation TE component method (Paired-End Only)popoolationte2
: Runs the PoPoolation TE2 component method (Paired-End Only)te-locate
: Runs the TE-locate component method (Paired-End Only)teflon
: Runs the TEFLoN component method (Paired-End Only)
- Reference FASTA (
-r/--reference
)- The genome sequence of the reference genome in FASTA format. The reads from the FASTQ file(s) will be mapped to this reference genome to predict TE insertions
- example
- Consensus FASTA (
-c/--consensus
)- A FASTA file containing a consensus sequence for each family
- example
- FASTQ File 1 (
-1/--first
)- Either the Read1 FASTQ file from a paired-end sequencing run or the FASTQ file from an unpaired sequencing run
- FASTQ File 2 (
-2/--second
)- The Read2 FASTQ file from a paired-end sequencing run. Not required if using unpaired data.
- Locations (
-g/--locations
)- A GFF file contianing the locations of the reference TEs.
- Each annotation should contain an
ID=
attribute that contains a unique identifier that does not match any other annotation. - If both the locations GFF and taxonomy TSV are not provided, McClintock will produce them using RepeatMasker with the consensus sequences
- example
- Taxonomy (
-t/--taxonomy
)- A Tab deliminted file that maps the unique reference TE to the family it belongs to.
- This file should contain two columns, the first corresponding to the refence TE idendifier which should match the
ID=
attribute from the locations GFF(-g
). The second column contains the reference TE's family which should match the name of a sequence in the consensus fasta (-c
) - example
- Coverage FASTA (
-s/--coverage_fasta
)- A fasta file of TE sequences to be used for the coverage analysis.
- By default, McClintock estimates the coverage and creates coverage plots of the consensus TE sequences (
-c
). This option allows you to use a custom set of TEs for the coverage estimations and plots.
- Augment FASTA (
-a/--augment
)- A FASTA file of TE sequences that will be included as extra chromosomes in the reference genome file (
-r
) - Some methods leverage the reference TE sequences to find non-reference TEs insertions. The augment FASTA can be used to augment the reference genome with additional TEs that can be used to locate non-reference TE insertions that do not have a representative in the reference genome.
- A FASTA file of TE sequences that will be included as extra chromosomes in the reference genome file (
The results of McClintock component methods are output to the directory <output>/<sample>/results
.
- Summary files from the run can be located at
<output>/<sample>/results/summary/
. - Each component method has raw output files which can be found at
<output>/<sample>/results/<method>/unfiltered/
. - Raw results are standardized into a bed format and can be found in
<output>/<sample>/results/<method>/*.bed
where<output>/<sample>/results/<method>/*.nonredundant.bed
has any redundant predictions removed. - Standardized results are filtered by parameters defined in the
config
files for each method. These config files can be found in/path/to/mcclintock/config/
and can be modified if you want to adjust default filtering parameters.
- McClintock generates a summary report that contains information on how the run was executed, read mapping information, QC information, and a summary of component method predictions
- This page also links to the pages that summarize the predictions from each method: all predictions by method, predictions for each family, predictions for each contig.
<output>/<sample>/results/summary/html/<method>.html
- The HTML report also summarizes reference and non-reference predictions for all families.
<output>/<sample>/results/summary/html/families.html
- A page is also generated for each family, which summarizes the coverage for the family consensus sequence and the family-specific predictions from each component method.
<output>/<sample>/results/summary/html/<family>.html
<output>/<sample>/results/summary/data/run/summary_report.txt
: Summary Report of McClintock run. Contains information on the McClintock command used, when and where the script was run, details about the mapped reads, and table that shows the number of TE predictions produced from each method.<output>/<sample>/results/summary/data/run/te_prediction_summary.txt
: A comma-delimited table showing reference and non-reference predictions for each component method<output>/<sample>/results/summary/data/families/family_prediction_summary.txt
: a comma-delimited table showing TE predictions (all, reference, non-reference) from each method for each TE family<output>/<sample>/results/summary/data/coverage/te_depth.txt
: (Only produced if coverage module is run) a comma-delimited table showing normalized depth for each consensus TE or TE provided in coverage fasta.- All tables and plots contain a link to the raw data so that users can manually filter or visualize it with other programs.
<fastq>_trimming_report.txt
: Information on parameters used and statistics related to adapter trimming with cutadapt. Provides an overview of sequences removed via the adapter trimming process.<fastq>_fastqc.html
: FastQC report of the trimmed fastq files. Provides information on the results of steps performed by FastQC to assess the quality of the trimmed reads.<fastq>_fastqc.zip
: FastQC graph images and plain-text summary statistics compressed into a single.zip
file
plots/*.png
: Coverage plots showing the normalized read coverage across each TE either from the consensus fasta (-c
) or the coverage fasta (-s
) if provided. Coverage of uniquely mapping reads (MAPQ > 0) is in dark gray, while coverage of all reads (MAPQ >= 0) is in light gray. Raw coverage at each postion in a TE is normalized to the average mapping depth at unique regions of the hard-masked reference genome. The average normalized coverage is shown as a black line, and is estimated from the central region of each TE omitting regions at the 5' and 3' ends equal to the average read length to prevent biases due to mapping at TE edges.te-depth-files/*.allQ.cov
: Raw read coverage at each position in a TE sequence. (Output ofsamtools depth
)te-depth-files/*.highQ.cov
: coverage of mapped reads with MAPQ > 0 at each position, omitting multi-mapped reads.
unfiltered/<reference>_insertions.bed
: BED file containing raw 0-based intervals corresponding to TSDs for non-reference predictions and 0-based intervals corresponding to the reference TEs. Reference TE intervals are inferred from the data, not from the reference TE annotations. Strand information is present for both non-reference and reference TEs.<reference>_ngs_te_mapper_nonredundant.bed
: BED file containing 0-based intervals corresponding to TSDs for non-reference predictions and 0-based intervals corresponding to the reference TEs. This file contains the same predictions fromunfiltered/<reference>_insertions.bed
with the bed line name adjusted to match the standard McClintock naming convention. By default, no filtering is performed on the rawngs_te_mapper
predictions aside from removing redundant predictions. However, the config file: (/path/to/mcclintock/config/ngs_te_mapper/ngs_te_mapper_post.py
) can be modified to increase the minimum read support threshold if desired.
unfiltered/combined.gff
: GFF containing 1-based TSDs for non-reference predictions and 1-based intervals for reference TEs. The reference intervals are based on the reference TE annotations.<reference>_relocate_nonredundant.bed
: BED file containing predictions fromunfiltered/combined.gff
converted into 0-based intervals with bed line names matching the standard McClintock naming convention. By default, no filtering is performed on the raw predictions aside from removing redundant predictions. However, the config file: (/path/to/mcclintock/config/relocate/relocate_post.py
) can be modified to increase the minimum left and right prediction support thresholds for both reference and non-reference predictions.
unfiltered/repeat/results/ALL.all_ref_insert.gff
: GFF file containing reference TE predictions with 1-based coordinates. The final column also contains read counts supporting the junction (split-read) and read counts supporting the insertion (read pair).unfiltered/repeat/results/ALL.all_nonref_insert.gff
: GFF file containing non-reference TE predictions with 1-based coordinates. The final column also contains read counts supporting the junction (split-read) and read counts supporting the insertion (read pair).<reference>_relocate2_nonredundant.bed
: BED file containing all reference and non-reference predictions fromALL.all_ref_insert.gff
andALL.all_nonref_insert.gff
. Coordinates are adjusted to be 0-based. By default, no filtering is performed on split-read and split-pair evidence. However, the config file: (/path/to/mcclintock/config/ngs_te_mapper/ngs_te_mapper_post.py
) can be modified to increase the default threshold for these values.
unfiltered/<reference>.absence.refined.bp.summary
: Tab-delimited table containing reference TEs that are predicted to be absent from the short read data. Position intervals are 1-based.unfiltered/<reference>.insertion.refined.bp.summary
: Tab-delimited table containing non-reference TE predictions. Position intervals are 1-based.<reference>_temp_nonredundant.bed
: BED file containing all reference TEs not reported as absent by TEMP in theunfiltered/<reference>.absence.refined.bp.summary
file. Also contains non-reference TE predictionsunfiltered/<reference>.insertion.refined.bp.summary
formatted as a bed line using the McClintock naming convention. Positions for both reference and non-reference predictions are 0-based. Non-reference predictions fromunfiltered/<reference>.insertion.refined.bp.summary
are only added to this file if the prediction has read support on both ends ("1p1") and has a sample frequency of > 10%. These filtering restrictions can be modified in the config file: (/path/to/mcclintock/config/TEMP/temp_post.py
). Non-reference TEs with split-read support at both ends are marked in the bed line name with "sr" and theJunction1
andJunction2
columns fromunfiltered/<reference>.insertion.refined.bp.summary
are used as the start and end positions of the TSD in this file (converted to 0-based positions). If the non-reference TE prediction does not have split-read support on both ends of the insertions, the designation "rp" is added to the bed line name and theStart
andEnd
columns fromunfiltered/<reference>.insertion.refined.bp.summary
are used as the start and end positions of the TSD in this file (converted to 0-based). Note: TEMP reference insertions are labelednonab
in the bed line name since they are inferred by no evidence of absence to contrast them from reference insertions detected by other components that are inferred from evidence of their presence.
unfiltered/<reference>.call.PE.vcf
: VCF file containing non-reference TE predictions. Non-reference TEs are annotated as 1-based intervals in the POS column and two consecutive coordinates in the INFO field. No predictions are made for reference TEs. Strand information is not provided.<reference>_retroseq_nonredundant.bed
: BED file containing non-reference TE predictions fromunfiltered/<reference>.call.PE.vcf
with a Breakpoint confidence threshold of >6 are retained in this file. This filtering threshold can be changed by modifying the config file: (/path/to/mcclintock/config/retroseq/retroseq_post.py
). The position interval reported in theINFO
column ofunfiltered/<reference>.call.PE.vcf
are converted to 0-based positions and used as the start and end positions in the bed lines in this file.
unfiltered/te-poly-filtered.txt
: Tab-delimited table with non-reference and reference TE predictions and support values. Predictions are annotated as 1-based intervals on either end of the predicted insertion, and also as a midpoint between the inner coordinates of the two terminal spans (which can lead to half-base midpoint coordinates)<reference>_popoolationte_nonredundant.bed
: BED file containing only TE predictions with read support on both ends ("FR") and with percent read support >0.1 for both ends were retained in this file. The entire interval between the inner coordinates of the of the two terminal spans (not the midpoint) was converted to 0-based coordinates. Filtering parameters can be modified in the config file: (/path/to/mcclintock/config/popoolationte/popoolationte_post.py
)
unfiltered/teinsertions.txt
: Tab-delimited table with TE predictions and TE frequency values (ratio of physical coverage supporting a TE insertion to the total physical coverage). PoPoolationTE2 does not indicate which predictions are reference and non-reference TEs. Also, only a single position is reported for each prediction, so the TSD is not predicted. Predictions may only have support from one side of the junction ("F" or "R") or both sides ("FR"). Prediction coordinates are 1-based.<reference>_popoolationte2_nonredundant.bed
: BED file containing all of the predictions fromunfiltered/teinsertions.txt
that have support on both ends ("FR") and have a frequency >0.1. The filtering criteria can be modified in the config file: (/path/to/mcclintock/config/popoolationte2/popoolationte2_post.py
). If predictions overlap a TE in the reference genome, that reference TE is reported in this file using the positions of the reference TE annotation (not the position reported by PoPoolationTE2). If the prediction does not overlap a reference TE, it is designated a non-reference TE insertion_non-reference_
. The coordinates for all predictions are adjusted to be 0-based.
unfiltered/te-locate-raw.info
: A tab-delimited table containing reference ("old") and non-reference ("new") predictions using 1-based positions. TSD intervals are not predicted for non-reference TEs, instead a single position is reported.<reference>_telocate_nonredundant.bed
: BED file containing all reference and non-reference predictions fromunfiltered/te-locate-raw.info
. Coordinates for both reference and non-reference TE predictions are converted to a 0-based interval. The reference TE end position is extended by thelen
column inunfiltered/te-locate-raw.info
. Non-reference TE predictions are a single position as TE-Locate does not predict the TSD size.
unfiltered/genotypes/sample.genotypes.txt
: A tab-delimited table containing all of the breakpoints and support information for insertion predictions. Predictions are treated as reference predictions if they contain a TE ID in column 7.
# from: https://github.com/jradrion/TEFLoN
C1: chromosome
C2: 5' breakpoint estimate ("-" if estimate not available)
C3: 3' breakpoint estimate ("-" if estimate not available)
C4: search level id (Usually TE family)
C5: cluster level id (Usually TE order or class)
C6: strand ("." if strand could not be detected)
C7: reference TE ID ("-" if novel insertion)
C8: 5' breakpoint is supported by soft-clipped reads (if TRUE "+" else "-")
C9: 3' breakpoint is supported by soft-clipped reads (if TRUE "+" else "-")
C10: read count for "presence reads"
C11: read count for "absence reads"
C12: read count for "ambiguous reads"
C13: genotype for every TE (allele frequency for pooled data, present/absent for haploid, present/absent/heterozygous for diploid) #Note: haploid/diploid caller is under construction, use "pooled" for presence/absence read counts
C14: numbered identifier for each TE in the population
<reference>_teflon_nonredundant.bed
: BED file containing all reference and non-reference predictions fromunfiltered/genotypes/sample.genotypes.txt
. Reference predictions use the coordinates for the TE with the reference ID from column 7. By default, only non-reference predictions with both breakpoints (C2 and C3) are kept in this file. Non-reference predictions must also have at least 3 presence reads (C10) and an allele frequency greater than0.1
(C13). These filtering restrictions can be changed by modifying the TEFLoN config file:/path/to/mcclintock/config/teflon/teflon_post.py
Some test data is provided in the test/
directory, though the fastQ files must be downloaded using the test/download_test_data.py
script.
python test/download_test_data.py
- The test data provided is a UCSC sacCer2 yeast reference genome, an annotation of TEs in the yeast reference genome from Carr, Bensasson and Bergman (2012), and a pair of fastq files from SRA.
python3 mcclintock.py \
-r test/sacCer2.fasta \
-c test/sac_cer_TE_seqs.fasta \
-g test/reference_TE_locations.gff \
-t test/sac_cer_te_families.tsv \
-1 test/SRR800842_1.fastq.gz \
-2 test/SRR800842_2.fastq.gz \
-p 4 \
-o /path/to/output/directory
- change
/path/to/output/directory
to a real path where you desire the McClintock output to be created. - you can also increase
-p 4
to a higher number if you have more CPU threads available.
- By default, McClintock runs all component methods with the data provided.
- If you only want to run a specific component method, you can use the
-m
flag to specify which method to run
python3 mcclintock.py \
-r test/sacCer2.fasta \
-c test/sac_cer_TE_seqs.fasta \
-g test/reference_TE_locations.gff \
-t test/sac_cer_te_families.tsv \
-1 test/SRR800842_1.fastq.gz \
-2 test/SRR800842_2.fastq.gz \
-p 4 \
-m temp \
-o /path/to/output/directory
- Yoy can also specify multiple methods to run by writing a comma-separated list of the methods after the
-m
flag
python3 mcclintock.py \
-r test/sacCer2.fasta \
-c test/sac_cer_TE_seqs.fasta \
-g test/reference_TE_locations.gff \
-t test/sac_cer_te_families.tsv \
-1 test/SRR800842_1.fastq.gz \
-2 test/SRR800842_2.fastq.gz \
-p 4 \
-m temp,ngs_te_mapper,retroseq \
-o /path/to/output/directory
- When running McClintock on multiple samples that use the same reference genome and consensus TEs, it is advised to pregenerate the TE locations GFF and a TE Taxonomy TSV. Otherwise, these will be redundantly created by mcclintock for each sample
- If you lack a TE locations GFF and a TE Taxonomy TSV, you can run McClintock with the
--make_annotations
flag to produce these files in advance.
python3 mcclintock.py \
-r test/sacCer2.fasta \
-c test/sac_cer_TE_seqs.fasta \
-p 4 \
-o <output> \
--make_annotations
- With the
--make_annotations
flag, McClintock will produce the reference TE locations GFF and taxonomy file using RepeatMasker, then exit the run.- Reference TE locations GFF:
<output>/<reference_name>/reference_te_locations/unaugmented_inrefTEs.gff
- TE Taxonomy TSV:
<output>/<reference_name>/te_taxonomy/unaugmented_taxonomy.tsv
- Reference TE locations GFF:
- You can then use the
--resume
flag for future runs with the same reference genome and output directory without having to redundantly generate them for each run.
python3 mcclintock.py \
-r test/sacCer2.fasta \
-c test/sac_cer_TE_seqs.fasta \
-1 /path/to/sample1_1.fastq.gz \
-2 /path/to/sample1_2.fastq.gz \
-p 4 \
-o <output> \
--resume
python3 mcclintock.py \
-r test/sacCer2.fasta \
-c test/sac_cer_TE_seqs.fasta \
-1 /path/to/sample2_1.fastq.gz \
-2 /path/to/sample2_2.fastq.gz \
-p 4 \
-o <output> \
--resume
## etc ##
- Individual samples can be run in a serial manner as shown in the example above, or run in parallel, such as through separate jobs on a HPC cluster. License
Copyright 2014-2020 Preston Basting, Michael G. Nelson, Shunhua Han, and Casey M. Bergman
Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:
- Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
- Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.