PTM age

Estimating age of PTMs based on a list of sites mapped to Ensembl Compara organisms. For the moment, the pipeline works only in phosphorylation.

Dependencies

• Python (v2 & v3)
• Biopython (v2 & v3)
• phyml (available in (home/linux)brew)
• newick-utils (available in (home/linux)brew)
• phytools (R package)
• ape (R package)
• biomaRt (R package)
• RMySQL (R package)
• ghostscript
• wget

Setup

The pipeline takes as input 2 files:

• data/species_list.txt a list of species used to generate the phylogenetic trees.
• data/all_sites_ptmdb.tab a list of PTMs in some of these species.

Additionally the next variables need to be updated in the Makefile.

COMPARARELEASE ?= 86 # Compara release
SPECIESTAG ?= species_n10 #tag to distinguish the organisms list
ANCTHRESHOLD ?= 0.51 # threshold for ancestral reconstruction
SAMPLESIZE ?= 4000 #number of positives and negatives: Total 2x

Running the pipeline

The pipeline consists in a few steps that are paralellized using LSF's bsub. After some of the steps a number of jobs will be launched and you should not continue to the next step until all of them are finished.

1- Download all Ensembl data, filter the set of organisms of interest and regenerate alignments and trees

make prepare

The next few steps calculate the probabilities of all STY to be phosphorylated according to a SVM trained using netphorest. The default pipeline trains the models using the available data in each organism but it only calculates the probabilities based on the human model. Some of the steps require some time to run or implemented in the LSF system, therefore you will have to wait and check that all jobs are finished before runing the next instruction. If you are ok using 0.5 as a probability for all phosphoacceptor residues you can jump to step X (WARNING: call to make ancestral not yet implemented to run without probabilities).

2- Run Netphorest in every organism

make netphorest_proteomes
#in LSF "bsub -n 10 lsmake netphorest_proteomes""

3- Prepare feature files, training set and test set to train the SVM. The total number of positives and negatives is contained in the variable SAMPLESIZE.

make prepareAllSVM

4- Train and test the SVM based on the previously generated files

make trainSVM

5- Prepare full dataset of STYs to classify

make prepareFullSVM

6- Classify the full set of sites using the human ST model

make classifyFull

7- Convert the predictions into probabilities

make getProbabilities

If everything worked fine the resulting probabilities based on the SVM can be found in temp/allSTY_probabilities.tab.

8- Now we will run the ancestral reconstruction. The next step will retrieve stats at 2 different windows (0 and +/-3 residues).

make ancestral

This step produces some stats on the different modified alignment columns.

wc -l results/stats* 

9- Collects ancestral states

make collectStates 

10- Compiles all results in summary file

make collectOrigins 

The resulting summary files will be accessible in the results directory.

cat results/all_origins_species_n10_w0_0.51.tab | grep "Homo sapiens" | cut -f9,9 | sort | uniq -c

Authorship

This project was developed by Romain Studer. David Ochoa just prepared organized the scripts in a common project and wrote the Makefile to automate the process.