idoerg/InformationAccretion

Compute Information Accretion of ontology terms from a dataset

InformationAccretion

Compute Information Accretion (Information Content) of ontology terms from a dataset.

Information accretion ($ia$), introduced in [Clark and Radivojac, 2013], is a measure of how much information is added to an ontology annotation by node $v$ given that its parents $\mathcal{Pa}(v)$ are already annotated. Specifically,

$$ia(v) = \log_2 \frac{1}{\mathrm{Pr}(v|\mathcal{Pa}(v))} = \log_2 \frac{\mathrm{Pr}(\mathcal{Pa}(v))}{\mathrm{Pr}(\mathcal{Pa}(v)|v) \mathrm{Pr}(v)}$$

To calculate this probability, we use the observed annotations in a dataset as the empirical distribution on which to estimate the probabilities. Thus, $\mathrm{Pr}(v|\mathcal{Pa}(v))$ is computed as the number of examples (proteins) annotated with the term corresponding to node $v$ and the terms corresponding to nodes $\mathcal{Pa}(v)$ divided by the number of examples annotated with the terms corresponding to nodes $\mathcal{Pa}(v)$. Note that the annotation of the term corresponding to node $v$ necessarily implies the annotation of terms $\mathcal{Pa}(v)$ to guarantee a consistent subgraph (valid annotations). This means that $\mathrm{Pr}(\mathcal{Pa}(v)|v)=1$ and the denominator can be computed simply as the count of examples annotated with the term corresponding to node $v$.

Counts are regularized by introducing an artificial record (protein) annotated with every terms and adding it to the counts. That is, we add 1 to the counts for all terms to avoid dividing by 0.

In this code, we take the Gene Ontology graph in .obo format and a set of annotations and compute $ia$ for every term in the ontology.

If you are not sure that the terms have already been propagater of the structure of the graph has changed since that propagation, you may wish to re-propagate the term and/or check that they are already propagated before computing $ia$.

[Clark and Radivojac, 2013] Clark WT, Radivojac P. Information-theoretic evaluation of predicted ontological annotations. Bioinformatics (2013) 29(13): i53-i61.

Usage

If you have a dataset of GO annotations, eg terms.tsv, and the corresponding GO graph structure, eg go-basic.obo, compute IA with the following call
python ia.py --annot terms.tsv --graph go-basic.obo
This operation takes about 2 minutes for a dataset of 5M terms

If your annotated terms are not already propagated, enable the --prop flag:
python ia.py --annot terms.tsv --graph go-basic.obo --prop
This operation may add about 4 minutes to the runtime for a dataset of 5M terms

Disclaimer

This is an independent implementation of the methods described in [Clark and Radivojac, 2013]

Please cite the orignal work [Clark and Radivojac, 2013] if using this measure and this repo if using this code.