Transcription factor motif biases calculator

Motivation

---

• Comparison of gene programs between cell types is necessary for making decisions on Transcription Factors (TFs) useful for conversions.
• Design and selection of gene sets for this propuse can be automated considering heuristics, such as top-N up-regulated genes.
• Comparison of gene groups should consider between different cell lineages and redundancy between gene sets.

Solution

---

This Python workflow:

1. Calculates motif biases (log2FC and Z-scores (up-coming)) between two cell types of interest, or a celltype versus a group of celltypes, using top-N genes and TF motifs for TF-gene associations.
2. Summarizes values as a table, for downstream analyses

Workflow steps

1. Expression values are obtained an normalized into Z-scores from a expression resource (e.g. TabulaMuris)
2. Z-scores also preparing sets of N genes (e.g. 1000) for each cell type.
3. Using pre-annotated motifs (from CIS-BP) cell type pairs are compared based on this metric.
4. For each comparison, a log2FC is reported.

Installation and running (typical time: less than 5 minutes)

1. Clone repository.

git clone gimme_motif_bias.git
cd gimme_motif_bias

#### Create an enviroment suitable to run this (gmb can be replaced by any name you want).
conda create --name gmb python=3.6 --file requirements.txt
source activate gmb

2. Motif hits CIS-BP (mouse genome) (~1GB).
   • download the following file and uncompress it in input Motif hits mm10 (Dropbox)

Dependencies (they must be installed before running).

(If creating an envinroment with requirements.txt, then already good to go).

• Python 3 https://www.python.org/
• Data Science packages for Python: pandas numpy
• MyGene (for ENSEMBL IDs conversion steps).
  • website
  • Conda Installation
  • Pip

Execution examples

python gimme_motif_bias.py --help # print help and exit
python gimme_motif_bias.py --listont # list all available ontologies and finish
python gimme_motif_bias.py --listmotifs ASCL1 # list all motifs related to ASCL1
# 1 versus 1
python gimme_motif_bias.py -a neuron -b hepatocyte --motifid M08474_1.94d 
# force rewriting
python gimme_motif_bias.py -a neuron -b hepatocyte --motifid M08474_1.94d --overwrite
# 1 versus many
python gimme_motif_bias.py -a neuron -b shortlist1 --motifid M08474_1.94d --overwrite

Output

• A TSV table with the respective effect sizes, p-values, in long format see output/motif_ensemblid.tsv.gz
• Excel table in similar format (if --xlsx is given).

Multiple runs

• If running sequencially for several pairs, run with option --overwrite to update the current table
• Values for repeated queries will unless cleaning the output directory. This is designed to save CPU time in the long run.

Running time

• Around 1-2 minutes for one pair.
• 10 minutes for full execution between cell type versus all one TFs (one CPU, default parameters, verification against other genomes and RNA secondary structure assessment).
• Adding more cell types increases quadratically running time quadratically.
• Increasing the number of motifs increases linearly the running time.

Misc

• Custom cell types and gene sets can be added manually in input/genes_by_ont as text files. Mind the format for labels to be recognized.
• You can add any custom set of hits of your interest in motif_hits_cisbp_build_1.94d_mm10, following the motif hit format.

Feedback, errors, or further questions

• Report in Issues.
• e-mail to Ignacio Ibarra (ignacio.ibarra@helmholtz-muenchen.de).