Is a Graphical Search Engine for Biology Ontologies Worth Developing?

[-Job-]

For my CSE Design Lab the teacher asked us to make a component for a search engine which uses a database populated from ontologies related to the biology field and then allows the user to view this information graphically in the form of a graph with ndoes and edges. The teacher and some graduates students had been working on this for a while, so we weren't supposed to redo what they had, but i didn't think the way they did things was the best, so me and my group did our version. I'm curious if anybody thinks such a tool is usefull or worthwhile and if they'd ever consider using it. If so i can develop it further instead of abandoning it at the end of the semester.
Any comments welcome, the page is at:
http://www.bloo.us/cse269
(choose display results as graph to see the graph, the text results are kind of plain for now. You can type things like, "mandible", "bone", "thrombosis", "blood vessel" etc. It also allows you to create your own graph, use the right-click menu)

 

[Moonbear]

I don't understand what anyone is supposed to do with this? I first tried some search terms, and nothing happened. Then I tried one of your examples and typed in "mandible" and got a tree of related terms, but it doesn't do anything more than that. If that's all it does, it seems pretty limited in usefulness, sort of like a thesaurus for biological terms, but not much else.

 

[Kalouste]

The engine only contains info about Fruit fly and mouse, so maybe that's why some terms you've typed (Moonbear) are not in the database.

I typed tail and I got the "tree of related terms" Moonbear was talking about. Right-click on term to see properties. The idea of making this in Flash language is really cool and user-friendly. I guess some images could help, though.

Ps: Why use 3 dialogue boxes and not only one instead? Could use logical notation for search.

Cheers!

 

[Moonbear]

The engine only contains info about Fruit fly and mouse, so maybe that's why some terms you've typed (Moonbear) are not in the database.

I typed tail and I got the "tree of related terms" Moonbear was talking about. Right-click on term to see properties. The idea of making this in Flash language is really cool and user-friendly. I guess some images could help, though.

Ps: Why use 3 dialogue boxes and not only one instead? Could use logical notation for search.

Cheers!

Okay, I can't right click because I left my mouse at the office: control clicking on a mac works. But that just gives me more related terms without anything of interest about them. Like I said, it's like a thesaurus. I used your example of tail. When I expanded the information, it gave me "part of caudal vertebra." Well, that's wrong. Caudal vertebrae are part of the tail, not the other way around. And wouldn't anyone going to the trouble of using a search engine know that? He asked for an honest opinion if anyone would use it to know if he should pursue it after the class is over. Unless there are a lot of features that will be added on that aren't there yet, I don't see any point in telling him to waste his time after he's done with this as a class assignment to keep working on it. I look at it and think, "so what?" It doesn't provide definitions (maybe useful for a young student, though not user-friendly enough for them I don't think) or links out to journal articles, or genbank data. So far, it only seems to use very simple anatomy for the terms in its database.

Maybe Job can explain who he thinks would use it and how he thinks it will benefit them. Maybe I'm just thinking about the wrong user group for this, or it isn't developed enough to appreciate what it's supposed to do.

 

[-Job-]

It's not really about the text results moonbear, it's about the alternative graphical display. In the future the nodes (terms) and edges (relations) you create will be stored in the database, and you'll be able to retrieve them easily. I'm also doing pathfinding, so you'll be able to type in two terms and i'll tell you how they're related (might be useful for genes and proteins). Right now it is a thesaurus because that's the data we scanned in.
You're very right about the three input boxes Kalouste, that's only there temporarily, and i like the idea of the images.
Who i perceived this tool to be good for is for researchers who need to easily organize lots of data, about how proteins and genes interact. There's plenty of gene ontologies we can read in. We also have a http://www.bloo.us/cse269/standalone/G6Graph.zip" version. Thanks for the feedback.

I appreciate your honesty Moonbear, trust me, you're not offending me.

 

[Moonbear]

It's not really about the text results moonbear, it's about the alternative graphical display. In the future the nodes (terms) and edges (relations) you create will be stored in the database, and you'll be able to retrieve them easily. I'm also doing pathfinding, so you'll be able to type in two terms and i'll tell you how they're related (might be useful for genes and proteins). Right now it is a thesaurus because that's the data we scanned in.
You're very right about the three input boxes Kalouste, that's only there temporarily, and i like the idea of the images.
Who i perceived this tool to be good for is for researchers who need to easily organize lots of data, about how proteins and genes interact. There's plenty of gene ontologies we can read in. We also have a http://www.bloo.us/cse269/standalone/G6Graph.zip" version. Thanks for the feedback.
I appreciate your honesty Moonbear, trust me, you're not offending me.

Oh, I see. So, it's not the particular database you're using that you think is of interest, but the idea of a way of displaying related data from any database? So, for example, you'd take something like the data being compiled at this website: http://www.nursa.org/template.cfm?threadId=117 and instead of having to follow link after link to get more detailed information or find coregulators known to interact with a given receptor, you'd get it displayed as a tree so you could immediately see how everything interacts, and then click on an item in the tree to get more information if you're interested? That might have more use for that sort of data set. Though, are there limits to the complexity it can handle? When I first visited that NURSA website, they used to have a list of all related molecules that interact with a given cofactor, but as more was known, and the complexity increased, they've changed the format. Now they have more functional information. I think displaying it all in a tree may be limited by people willing and able to sit down and find all the related information and get it into a database in the first place. I could envision taking Genbank data, displaying homologous genes in other species, variants of genes in the same species, protein products of those genes, and then interacting proteins with those protein products, any of which you can click and get the same information for that protein and it's gene, with links available to the articles initially citing these relationships, that might be a useful approach.

The downside is that putting it in a database gives the appearance that relationships are known, when often, they may still be hypothetical.

So, thanks for the clarification of the purpose. I think it makes a bit more sense now.

Is there any relationship of objects in the tree to the lengths of the lines connecting them?

 

[-Job-]

Yes, you got the idea right. We are working on a per user thing, where each user is able to keep their data (with the option of making it public available). So when displaying data you can choose what relation types you want to display and whether you want to display only your data, the public data or both.
We are trying to integrate this data with pubmed, so that the relations have associated the articles at pubmed that use them.
There's no relation between the length of the edge and the relation. You can actually drag the nodes around. You can also right click and create your own nodes and edges. The teacher actually mentioned the same thing you said about some of the relations being only hypothetical, we're supposed to indicate that in the graph (by using a different color for example). I also asked the teacher who would be willing to sit down and put the data in, and he said the authors would be interested in that because they want their research material to be cited or used in any way. I don't know if that's the case or not to tell you the truth.

I forgot to mention that we also want to allow the user to be able to display an article at pubmed as a graph with nodes and edges. Hopefully once this data accumulates, the pathfinding option will be an valuable thing (how does protein X interact with some gene Y?). Pathfinding is very expensive to perform (computation-wise), but i have a neat algorithm that can do it pretty fast.