Isoforms

Manifest

• APCanalysis - Ismael's sandbox
• archive - old stuff no longer used
• bin - programs
• data - output of geniso runs mostly
• gong - Gong's sandbox
• ian - Ian's sandbox
• lib - libraries (just isoform.py right now)
• models - PWMs and the worm.splicemodel

Programs

• cmpiso - compares collections of isoforms in GFF

• geniso - generates isoforms and their probabilities (old version)

• modelbuilder - creates the various model files

• optiso - optimizes model parameters with a genetic algorithm

• run_geniso - parellelizes geniso on the smallgenes dataset

• run_optiso - parellelizes optiso

• nmd-ish.py - re-scores isoforms that are NMD targets

• isorandom - counting isoforms in random sequences

Data

Data collection is described in smallgenes.

• results_optiso2.csv - model weights for each gene
• APCisos.base.tar.gz - APC results without optiso and without nmd
• APCisos.optiso.tar.gz - APC results with optiso only
• APCisos.nmd.tar.gz - APC results with nmd only
• APCisos.optiso.nmd.tar.gz - APC results with optiso and with nmd

Models

See the models directory for standard models and modelbuilder for how to build the models.

Trivia

There are 19.938 billion RNASeq_splice records in WormBase. As a rough estimate of intron frequency, divide intron counts by 20 billion.

Generating APC isoforms

The APC algorithm generates all possible combinatons of isoforms given an input gene sequence in fasta format, scores each isoform, and returns a gff of predicted isoforms.

Other components of the algorithm include optiso. optiso uses a genetic algorithm to optimize the weights for each part of the APC model (kmer, pwm, length).

The smallgenes dataset is used as input. git clone this repo in the directory containing your other github repos, such that:

Code/
├──smallgenes/
├──isoforms/

git clone https://github.com/KorfLab/smallgenes.git

Next unpack the smallgenes dataset in isoforms/

ln -s ../smallgenes/genomes/c.elegans/smallgenes.tar.gz
tar -xvf smallgenes.tar.gz

Create the APC model file with modelbuilder. This requires the OpenTURNS library to be installed. Can be installed using conda:

conda install -y openturns

Next use modelbuilder to create worm.splicemodel, which contains the models used as input for geniso3. Individual model files are also created and can be viewed in models/

./modelbuilder smallgenes/ worm models/ > worm.splicemodel
mv worm.splicemodel models/

Next run optiso as the weights are a component of geniso3. Note that run_optiso by default uses 1 CPU core. This will take a while, I recommend doing this on a remote cluster. Same for run_geniso. This program creates results_optiso2.csv.

chmod +x run_optiso2
./run_optiso2 smallgenes/ models/worm.splicemodel --cpu 15

run_geniso can be used with or without weights (optiso), and with or without NMD detection. By default, one CPU core is used. Make sure the output file names describe the chosen parameters. Without weights and without NMD:

./run_geniso smallgenes/ models/worm.splicemodel --outdir APCisos.base/ --outname APC.base --cpu 15

With weights and without NMD:

./run_geniso smallgenes/ models/worm.splicemodel --weights results_optiso2.csv --outdir APCisos.optiso/ --outname APC.optiso --cpu 15

Without weights and with NMD:

./run_geniso smallgenes/ models/worm.splicemodel --outdir APCisos.nmd/ --outname APC.nmd --nmd --cpu 15

With weights and with NMD:

./run_geniso smallgenes/ models/worm.splicemodel --weights results_optiso2.csv --outdir APCisos.optiso.nmd/ --outname APC.optiso.nmd --nmd --cpu 15

The output of run_geniso is a directory with gff files containing the APC isoforms for each gene in the smallgenes dataset.