On 2013 Nov 09, Jonathan Moore commented:

Very nice job to the team who made this. The prospects for integration of automated learning and literature-based curation of ontologies are great, and it looks like a useful tool for discovery - I'd like to try it in discovery mode with some other species with 'thin' ontology annotations.

One thing I wondered about is the need for the hierarchical tree imposed at step one, and its effects on the final result. I wonder whether the tree-like requirement leads to the possibility of falling into a local optimum final graph, ending up with more of the initial tree-like structure than the data really suggest, rather than finding the optimum final graph from the true shape of the input data. The computational requirements for learning an arbitrary graph at step one seem daunting though.

Maybe this could be part of the reason for the better match to GO for cellular components, which maybe tend to be more inherently hierarchical in comparison to molecular functions or biological processes.

On 2013 Nov 09, Jonathan Moore commented:

Very nice job to the team who made this. The prospects for integration of automated learning and literature-based curation of ontologies are great, and it looks like a useful tool for discovery - I'd like to try it in discovery mode with some other species with 'thin' ontology annotations.

One thing I wondered about is the need for the hierarchical tree imposed at step one, and its effects on the final result. I wonder whether the tree-like requirement leads to the possibility of falling into a local optimum final graph, ending up with more of the initial tree-like structure than the data really suggest, rather than finding the optimum final graph from the true shape of the input data. The computational requirements for learning an arbitrary graph at step one seem daunting though.

Maybe this could be part of the reason for the better match to GO for cellular components, which maybe tend to be more inherently hierarchical in comparison to molecular functions or biological processes.