Information-theoretic approach to create a parsimonious Gene Ontology. The purpose of doing this is to prepare a small-enough network for Visible Neural Network (VNN).

Please refer to notebook/ for a step-by-step walk through.

• 1_Calculate_Entropy_for_Terms.ipynb demostrates how to calculate entropy for terms.
• 2_Visualize_Entropy.ipynb show what terms are of low entropy.
• 3_Pruning_Low_Entropy.ipynb removes child terms when parent terms are of low entropy. And examine the structure of the pruned ontology.
• 4_Pruned_ontology_quick_viz*.ipynb visualizes the pruned ontology.
• 5_Get_Differential_Terms.ipynb tries to make sure that the pruned ontology is still capable of distinguishing the difference between neuron and microglia.
• 6_Visualize_differential_term.ipynb shows the "differential terms" between different cell types.
• 7_Output_to_DCell.ipynb outputs ddot.Ontology the pruned ontology to DCell format. [See here for an example].

For those who are preparing the ontology for neural network, notebook 1,3,7 are the true pipeline. But taking at a look at the network statistic in notebook 3,5,6 might help you choose a better entropy threshold.

Unfortunately this repository needs two seperate conda envrionment. Most notebooks can work in DDOT environment. see ddot.yml Notebook 2 needs [GOATTOOLS]. see ontoPrune.yml

(accessible to Ideker Lab) /cellar/users/hsher/ontoPrune/data/prune_go* are pruned Gene Ontology at different entropy thershold. Load by using ddot.Ontology.read_pickle().

/cellar/users/hsher/ontoPrune/data/*.topo contain ready-to-use DCell Topology file. BP, MF, CC refers to Biological Process, Molecular Function, Cellular Component. 080 means entropy threshold 0.8. -drop means dropping the genes directly connected to the root (BP/MF/CC)