#######  #### ###  ##
         ###   ###  ##
 ####### ###   ###  ##
 ##      ###    #####
 ##   #####      ###            

Project Fragsifier: STR Sequence Fragment Classifier

Fragsifier is a forensic short tandem repeat (STR) sequence extraction tool for massively parallel sequencing data.

Original article: Forensic STR allele extraction using a machine learning paradigm

Fragsifier uses a machine learning approach to classify potential STRs in addition to the classic flanking sequence alignment technique.

Algorithm overview

To find STRs on a read, the tool performs the following procedures:

• Detect all repeat stretches present on the read
• Create a list of STR candidates containing all possible STR sequences (combinations of repeat stretches)
• Use k-mer/n-gram based sequence classifiers to predict the locus of each candidate
• Perform flanking sequence alignment on each STR candidate to locate the STR boundaries and generate an alignment score
• Candidate STRs with congruent sequence classification and flanking sequence alignment results are returned as a legitimate STR

Note: Fragsifier is designed to analyze repeat stretches so will skip SNP sequences.

Installation

Fragsifier was developed in Python 3.6. It is recommened that the program be installed in a Conda virtual environment.

Fragsifier is now available on PyPI and can be installed via pip:

pip install fragsifier

Installing using the pip command will also install the required packages automatically.

Alternatively, Fragsifier can be installed from the download source folder by running:

pip install .	# run in setup.py directory

Fragsifier can also be run directly from the source folder by running:

pip install requirements.txt	# install requirements first
python fragsifier.py		# type --help for a list of adjustable parameters

Usage

Fragsifier works out of the box with the pretrained models as described in the paper. The tool readily extracts STR loci described in STRait Razor v2s.

Sequence model retraining

The sequence model can be retrained using custom data following the below procedures:

• Delete all the files in the models folder
• Add training sequence files to the models folder:
  • A FSV_reference_sequences_examples.csv file containing example sequence from each STR locus
  • A FSV_training_negative_examples.csv file negative (non-STR, noise) sequences
  • Examples of above files can be found in the models folder (prior to being deleted)
• Make sure to update the locus.config file in the routines folder with new STR loci information
• Machine learning prediction models will be rebuilt automatically when the tool is next run. A 'building models' message will be displayed.

Tip: The models can be retrained in the source folder when downloaded, and then installed to the system using pip install . afterwards.

Release notes

v1.0.1

• Reorganized code
• Code can now be run as a script in the local folder or be installed as a package
• Installation via pip now installs dependencies
• Fragsifier now available as a package on PyPI

Alexander YY Liu | yliu575@aucklanduni.ac.nz