spruceup 2024.8.20.post1

A module for lexible identification, visualization, and removal of outliers from large multiple sequence alignments

spruceup

        A  
       TCG  
      ACGTA  
        T 

Tools to discover, visualize, and remove outlier sequences in large multiple sequence alignments.

spruceup is a Python tool for biologists (bioinformaticians, phylogeneticists, evolutionary biologists) doing inference on phylogenomic sequence alignments. It allows discovery and removal of individual poorly aligned sequences or sequence fragments (alignment rows), which is different from the problem of poorly aligned sequence blocks (alignment columns) commonly addressed by alignment trimming software.

If you are using this program, please cite this publication:

Borowiec, M.L. (2019) Spruceup: fast and flexible identification, visualization, and removal of outliers from large multiple sequence alignments. Journal of Open Source Software, 4(42), 1635, https://doi.org/10.21105/joss.01635

This script uses numpy, scipy, matplotlib, psutil, tqdm, and treeswift.

Installation and requirements

You can download a zipped GitHub repository, clone it if you have git installed on your system, or install using pip from the Python Package Index. Cloning the repository will also download example files used in the tutorial below.

spruceup has been developed and tested under Ubuntu Linux. It requires you have Python version 3.6 or newer. You can use spruceup with early releases of Python 3.7, but as of today (12 August 2019), you may experience trouble installing it if using Python 3.7.4. Dependencies should be installed automatically. If your system does not have Python version 3.6 or newer you will need to download and install it. On Linux-like systems (including Ubuntu) you can install it from the command line using:

sudo apt-get install python3.6

It is highly recommended that you install spruceup in a new Python environment. If you are using the conda package manager, type:

conda create --name spruceup python=3.6

And follow instuctions on the prompt.

Note that at this point you only created a virtual environment called "spruceup". You haven't installed it yet. Now activate your environment (with conda you would type conda activate spruceup) and install. I recommend using pip (Python Package Index):

pip install spruceup

If you want to download example files, simply download and unzip this repository or, if you have git, clone it with:

git clone https://github.com/marekborowiec/spruceup.git

You can then install spruceup from within this cloned repository:

python setup.py install

If you do not wish to install conda, you can set up a new environment using a different package manager, such as venv:

python3 -m venv spruceup

There is a known issue when installing on Mac/OSX, which arises because matplotlib requires a framework build of Python. If you are working on OSX, you can install this with conda install python.app and use pythonw -m instead of python -m to run spruceup. See also here.

If you are unsuccessful in installing spruceup using these instructions on your system, try installing Docker. You can then pull an image containing installed spruceup here or using

docker pull marekborowiec/spruceup-miniconda

You can then use spruceup by invoking

docker run -w /data -v "/path/to/your/files/:/data" marekborowiec/spruceup-miniconda bash -c "source activate spruceup && python3 -m spruceup config_example.conf"

Replace the part of the command following -v and preceding : with the path to your files and config_example.conf with your configuration file name.

Interface

Once you successfully installed spruceup you will need 1) an alignment in FASTA, PHYLIP or NEXUS format, 2) (optional) a guide tree for your alignment in NEWICK format, and 3) configuration file to run the program. To run the program from the command line you can type:

python -m spruceup my-configuration-file.conf

Directory examples contains a template configuration file (/examples/config_example.conf). This file should be a template for your own analyses. It has the following fields:

[input]

The input category defines parameters of the input alignment and its type.

input_file_name is the file path of the alignment to be processed. This can be provided as a file name, if the config is in the same directory as alignment and spruceup command will be ran from there as well. Alternatively, you can provide the absolute path. This should be a relatively large, concatenated alignment. The spruceup algorithm works the better the more data it has, so if you have a phylogenomic dataset consisting of many single-locus alignments, you should first concatenate them. You can then split the processed/trimmed alignment back into single-locus alignments using another utility such as AMAS.

input_format indicates which format the alignment file is in. It can be one of five popular formats: fasta, phylip, phylip-int (interleaved PHYLIP), nexus, and nexus-int.

data_type tells the program whether your alignment contains amino acids (aa) or DNA nucleotides (nt).

distances_object_file is the file name of an existing distance object. By default this saves to the directory where you run the spruceup command and so it is better to provide absolute path here. Because sometimes you will want to adjust cutoffs or cutoff criterion and computing distances is the most time-consuming part of the analysis, spruceup saves a json format file with distances from each analysis. By default this is blank, but if you do have a distance file from a previous analysis and you want to trim your alignment with new cutoffs, supply the json file name here. spruceup will then run with new trimming cutoffs and/or criterion but without the need to re-calculate distances.

guide_tree is a phylogram or cladogram NEWICK format file to be used as a guide tree. File name or absolute path. The tree can be inferred using any method and can be fully resolved or contain polytomies. If you do not supply a guide tree the program will still run but without phylogeny it will have less information to identify misaligned sequences. This is particularly important in vartiable algnments with distantly related samples (individual sequences representing individual, taxon, OTU, etc.), where it is more difficult to distinguish genuinely variable sequences from misaligned fragments. In cases where you suspect and are mainly concerned with samples with spuriously long terminal branches it is adivsable to supply topology-only (cladogram) guide tree and/or run the program without a guide tree.

[analysis]

The analysis category defines parameters used to analyze and clean up your alignment.

cores how many CPU cores to use for distance calculations.

distance_method chooses to compute uncorrected p-distance with uncorrected or Jukes-Cantor-corrected distance with jc.

window_size chooses how many characters (aa/nt) to include in a window in which distances will be calculated. Default value that works well for most alignments is 20.

overlap indicates how many characters (aa/nt) each sliding window will be overlapping with preceding window. Overlap of 10 and window size of 20 means that each new window will move 10 positions down the alignment and overlap by 10 characters with the preceding window. Decreasing the overlap will decrease computational burden because fewer windows will be created. Default value is 10 (half of default window size of 20) but you may want to go lower, to half of window size or even 0 (non-overlapping windows) if your alignment is very large and you want to decrease compute time and memory usage and don't mind sacrificing some precision.

fraction signifies proportion of samples that will be used to calculate average distance in each window. With fraction set to 1.0 distances for each sample will be calculated against all other samples in the alignment. With fraction set to 0.5 distances for each sample will be calculated against a random draw representing 50% of samples in the alignment. Lowering this number will help to speed up calculations in alignments with large numbers of taxa.

criterion chooses how outlier distances will be determined. weibull_min means that a Weibull distribution will be fitted to your distance data for each sample and cutoffs will be determined by specifying quantile of observations above which sequence will be considered outliers. If you are using mean, simple multiple of those values computed for each sample will be considered cutoffs for identifying outliers.

cutoffs specifies multiple values considered as cutoffs. If you are using mean as your criterion (default), use multiples of those values, for example 2,3,5 etc. Defaults of 2,3,5,10,15,18,20,25,30 for mean criterion should be a useful starting point for most datasets. You can always trim with additional criteria and cutoffs after the initial analysis (see point 5 below under Interpreting the output). If using weibull_min criterion, use fractions of 1, for example 0.9,0.995 etc. Default values that should work for most alignments are 0.9,0.95,0.97,0.99,0.995,0.999. If the alignment contains many saturated or poorly aligned sites, a low setting may result in huge amount of data being trimmed from the original alignment. This is time-consuming and you may want to trim your alignment with a more stringent 'block' method before using spruceup or remove lower cutoff values from the list.

manual_cutoffs is an optional setting that allows manual modifications to cutoffs for individual samples. Important: the cutoff here is the absolute distance value on the x axis of your output plots where you want to draw the cutoff line, not where you expect the value of your weibull_min or mean to fall on that plot. It may prove useful if only one or a few samples have a significant proportion of poorly aligned sequences, skewing their overall cutoff such that they are not being flagged. If you find that this is case, however, you should probably rather be checking your data and pipeline for errors!

[output]

The output category tells the program how and where to save your analysis results. Remember that by default any file will be written to directory from which you executed spruceup command. To change this behavior provide full paths.

output_file_aln is the name for your trimmed output alignment(s). The actual name saved on your machine will have a prefix signifying cutoff value used.

output_format file format for the trimmed alignment. Choose from fasta, phylip, phylip-int, nexus, or nexus-int.

report name of files containing information on which sequences were flagged as outliers. The actual name will have a prefix signifying cutoff value used.

log is the name of the log with all analysis screen output.

Examples and interpretation of results

To use spruceup you will need to run the spruceup script from the command line and provide the name of your configuration file as the argument. If you downloaded and installed spruceup from source, the example alignment and config files can be found in examples subdirectory:

cd ./examples
python -m spruceup config_example.conf

The example alignment is a subset of empirical, anonymized ultraconseved element or UCE data set generated from insects.

Once you run the script, the sequence alignment will be divided into a number of windows of the size and overlap you specified. The script will then compute distances for each sample (alignment row: the sequence representing an individual, taxon, OTU etc.) in each window. This is done all-by-all by default or all-by-fraction of samples, if specified. You will see some messages along the way, including a progress bar that will display the number of iterations (windows) and remaining time, as we as the maximum amount of memory used for distance calculation.

Once all distances are calculated, criterion and cutoffs settings will determine which windows are considered outliers and should be trimmed out of the alignment. When using the weibull_min criterion, specifying a quantile of 0.99 means that any sequence window that lies above 99th percentile of distances of a given sample to other sample in that window will be deemed an outlier and should be removed. In theory, setting of 0.99 should mean that 1% of all sequence windows will be removed from each sample. In practice, this is not true because real-life sequence data does not perfectly fit into the fitted density distribution. Cutoff being constant, certain samples may have many outlier (misaligned) sequence fragments and more than 1% of sequence data removed, while others may have no misaligned fragments and no outliers.

You can now go back to your configuration file and try other cutoffs or methods without the need to re-calculate distances (unless you would like to use different correction or scaling). Simple load the generated json file (see below) with the distances_object_file option under [input] category.

Interpreting output

spruceup produces several types of output:

1. Report files ending with suffix -report.txt, one of which is written for each cutoff specified, which indicated by the prefix (e.g 0.95, 0.99 and so forth). These files contain the distance cutoff value for each sample and which sequence windows were determined to be outliers and removed.

2. Trimmed alignment files ending with suffix -trimmed.fas/phylip/nexus, again, one for each cutoff value. These are the alignments with outlier windows removed.

3. Distance distribution png plots, one for each sample. These images can be used to examine the distribution of distances for each taxon, its fit to density distribution, and cutoff value placed on each sample given a cutoff.

An example of distances plot is below. The header is the name of the sample. The x-axis indicates distance to other samples, ranging from 0 to the maximum distance that was found for the sample. The y-axis specifies relative number of windows. Blue bars comprise the histogram of distances. Orange line is the fitted Weibull distribution (only shown when using the weibull_min criterion) and the vertical dashed line indicates the cutoff above which any window will be deemed an outlier and removed.

The first example plot below shows an sample with relatively smooth distance distribution and few sequence windows with extreme values, none of which are greater than 0.25. You may not be able to see individual window distances as visible histogram bars since the distributions comprise thousands (in this small example) to hundreds of thousands of distances.

The following plot shows a sample with less smooth distribution, overall sequences fewer sequences due to missing data, as indicated by the heigth of the histogram bars, and many outlier windows.

Both examples have been processed under the same cutoffs values, 0.95,0.97,0.99 quantiles of fitted distribution but in the poorly aligned data the last falls outside of computed distance values. These plots, combined with visual examinaton of report files and alignments should serve you as a guide on what criterion and cutoff values make most sense for your dataset.

4. Log file

This is a log file that will contain the same information that appears on the terminal screen, excluding progress bar. It will log what steps were taken by the program, their timing, show information about how many/what proportion of sites were trimmed at each cutoff and output file names that were written.

5. Distances Python object file

Calculating distances with spruceup is often the most time- and memory-consuming part of the process, although trimming very large numbers of positions from the alignment can also take a long time. Because of this spruceup writes a json format file each time you run an analysis from scratch, allowing you to load it up later and trim with different criterion or cutoff values. Note that distances will be specific for each window size, overlap, and taxon fraction and you will need to re-run the whole analysis if you want to adjust these parameters. Note that the json file can be quite large at >150MB per 100,000 windows and 100 taxa.

Testing

Note: As of version 2024.7.20 and newer, the tests are not working because spruceup completely changed the way distances are computed. Stay tuned as I update the tests.

Tests written for spruceup code use the Python's standard library module unittest and are integrated with setuptools. This means that if you downloaded the spruceup source code, you can run tests from the top spruceup directory after installing with:

python setup.py install
python setup.py test

These tests are likely to take several minutes because they include metamorphic tests (cf. Giannoulatou et al. 2014) which involve running spruceup on simulated data with known properties. You can speed things up by increasing the number of cores in the metamorphic test configuration file found under tests/metamorphic_tests/simulation.conf. If you want just a quick check of basic functions, run unit tests without the time-consuming metamorphic tests. To do this, from the top spruceup directory run:

python -m unittest tests.test_simple

Issues and development

If you encounter bugs or problems running the code, create a new issue on GitHub. Everyone is encouraged to contribute.