Predicting protein stability. This repo is a refactoring of @ccorbi's main pulse pipeline for my needs.

From paper here: Semi-supervised Learning Predicts Approximately One Third of the Alternative Splicing Isoforms as Functional Proteins.

Change preprocess/preprocess_settings.json and features/features_settings.json to your desired settings before use.

• python2.7
• preprocessing requires cufflinks and samtools
• BLAST is required in the preprocessing step after alternative splice site extraction
• last step of the preprocessing step requires the biopython package
• disorderome scoring step of feature extraction step requires iupred
• domain scoring requires [pfam] (http://www.vcru.wisc.edu/simonlab/bioinformatics/programs/install/pfamscan.htm)
• perl is required for some scripts in feature extraction and in last step
• perl needs the following modules (download with cpan <module name>:
  • Getopt::Long
  • LWP::UserAgent
  • HTTP::Request::Common qw(POST), qw(GET)
  • HTTP::Headers
  • XML::XPath
  • XML::XPath::XMLParser
  • JSON
• sable is required in the last step of feature extraction

Clone the repo, setup:

git clone https://github.com/wonjunetai/pulse.git
cd pulse
python setup.py # set up all directories for pipeline output

Need to import info_canonical_v3.ddbb, refGenome/, uniprot_sprot.fasta, and param_files/.

Use readlink -e . when you're in root, and export that variable to PULSE_PATH using:

export PULSE_PATH=<result from readlink -e .>

First, insert all of your .bam files into input/cell_lines, and then run:

python app.py

Preprocessing

1. Cufflinks generates transcriptome from sequence alignment data (.bam file).
2. Find exon-skipping alternative splicing events from transcriptome.
3. Identify anchors to alternative splicing events using BLAST.
4. Filter out alternative splicing events without a good anchor.
5. Generate an index file to locations of alternative splcing events.

Feature Extraction

1. Transmembrane scoring.
2. Post transcriptional modification scoring.
3. Eukaryotic linear motif scoring.
4. Disorderome scoring.
5. Domain scoring using PFAM.
6. Sable scoring.
7. Mutation scoring.
8. Conservation/network features generation.

Machine Learning Step

Five iterations of random forest classifier - takes the average of all iterations to generate final votes from ensemble and importance of each feature.

• Need to download the uniprot using API and insert into pulse/input. After the uniprot_sprot file is downloaded into input/, need to run:

formatdb -i ./input/uniprot_sprot.fasta -p T            # creates a hashed version of uniprot_sprot.fasta database