A work in progress. BioPericles is designed to use supervised Machine Learning techniques to classify samples into clusters. It currently extracts features using SNP analysis although other features could be used.

In the future it will hopefully be able to identfy which features are the most descriminating between clusters. This could be useful for creating diagnostic tests.

python setup.py develop
pip install -r requirements.txt

cluster-sequences

Takes a multifasta of sequences and a metadata file mapping sequences to clusters and splits the file into one file per cluster.

Requires: biopython

export data=biopericles/tests/data
cluster-sequences -o ${data} ${data}/clusters_spreadsheet.csv ${data}/multifasta.aln

get-cluster-consensus

Takes a list of multifastas, one per cluster. For each multifasta it calculates the bases which are consistent throughout the cluster and marks those that are not with an 'N'.

Requires: biopython

export data=biopericles/tests/data
get-cluster-consensus -o ${data}/cluster-consensus.aln ${data}/multifasta.aln

calculate-ancestral-sequences

Takes an aligned multifasta, builds a phylogenetic tree from the sequences (using RAxML) and then calculates what the sequences for the internal nodes might look like (using FASTML). Outputs a fasta file and newick tree to specified files or defaults to the current directory.

Requires: biopython, PyVCF, RAxML, FASTML

export data=biopericles/tests/data
calculate-ancestral-sequences ${data}/animals.mfa

get-snp-features

Takes an aligned multifasta and runs snp-sites to find SNPs. These SNPs are used to output a CSV with a row per sample and columns for each SNP position. Positions which match the reference (the first sequence in the multifasta) have the value 0 while differences are marked with 1.

You can also pass a VCF file as input by setting the --input_format option to vcf. This bipasses any call to snp-sites. VCFs can have a couple of formats. The first is that output by snp-sites which has INFO set to AB, genotype fields set to either . if this sample matches the 'reference' or the alternative base otherwise. See file_with_SNPs.aln.vcf for an example. Alternatively you can provide a VCF with a Genotype field like file_with_SNPs_in_GT_format.aln.vcf.

For convenience, if you provide the script with a multifasta, you can ask to save the SNPs calculated by snp-sites in VCF format using the --vcf-output option. This file is output in the Genotype style rather than as produced by snp-sites. Users can then filter / edit this file before recalculating different features.

Requires: biopython, snp-sites, PyVCF

export data=biopericles/tests/data
get-snp-features ${data}/file_with_SNPs.aln
# or alternatively
get-snp-features --input_format vcf ${data}/file_with_SNPs_in_GT_format.aln.vcf

train-classifier

Takes a CSV of features and a CSV with sample to label mappings and trains a classifier. The features should be supplied with one row per sample and a column per feature with 1 or 0 in each cell.

Although designed to work with the output from get-snp-features there is no reason you cannot add or remove features by altering the columns.

It is possible to pass in a second feature file using the --cluster-features option. This file is used to get a list of features which should be used for classifier training. This is useful if you only want to use SNPs which are consistent within clusters.

The classifier is output as a Python Pickle although this could be changed. Pickles of unknown origin should be considered dangerous; the load method on SampleClassifier expects to be passed an MD5 or SHA1 before loading a classifier from disc. These values are output at the end of classifier training.

Requires: numpy, scikit-learn, scipy

export data=biopericles/tests/data
train-classifier ${data}/fruit_example_features.csv ${data}/fruit_example_metadata.csv fruit_classifier.pkl

The scripts are pretty reusable, prototypes are less polished but might be a useful starting point for someone trying to use BioPericles on their own data. They're not very well tested so use them with care.

annotate_vcf.py

Takes a VCF, GFF and a Fasta file. It uses these to output a VCF with anotations taken from the GFF file to identify synonymous, nonsynonymous and intergenic variants by looking for CDS features. It uses a format similar to Variant Effect Predictor but provides less detail.

Requires: GenomeTools

install_dependencies.sh

Installing the dependencies for BioPericles can be a bit of a pain. A script is included in the prototypes directory to help install dependencies in a clean system on which you have root access. NB This script manipulates your PATH and LD_LIBRARY_PATH environment variables. It has only been tested in an Ubuntu Docker Container.

install_dependencies.sh works a bit like a Makefile. As long as you run it from the same location it will not re-download or build the dependencies and will just re-update your environment variables if required. This is useful for Travis CI because it allows us to cache the build folder to speed up subsequent test runs.

cd BioPericles/prototype
sudo ./install_dependencies.sh
sudo pip install -r ../requirements.txt

• Use the classifier to classify things
• Make it clearer what the test set is or possible to specify what it should be
• Use the classifier to identify a subset of features which are the most discriminative
• Consider how the classifier could be tuned (either manually or automatically)
• Make it easier to score the classifier using a separate test set