Molecular Predictor Repository? (not gene set repository)

I have a simple question.  Say that I have the results from a gene expression analysis done in my laboratory or pulled from a public repository.  Say the sample has something to do with cancer (or I think that it might).  Say I read about so called 'signatures' that have been found to be associated with key phenotypes related to cancer.  (Here is a list of 13 signatures like this).

How do I now test to see which, if any, of these signatures are showing up in my sample?

I have my input, (e.g. the Affy CEL file from my experiment), how do I get the output that indicates that my sample shows an active wound response, suggests poor outcomes in breast cancer patients, looks like lung-specific metastasis, etc. etc.

This should be relatively easy, no?  I've got data about human gene expression, these people have made useful predictive models that take human gene expression as input.  Where is the website?

Some people have directed me to useful resources like GeneSigDB that provide curated repositories of "gene signatures".  However, these "signatures" are just sets of genes, they are not predictive models.  If all that we needed were gene sets, no one would ever need to train a random forest classifier or a support vector machine on the data associated with those gene sets.  Sets of phenotypically related genes are great, but I need the full predictive model.

The only system that I know of that seems to have the capacity to answer my question (had the model builders used it) is the Synapse platform.  For example, if you are good at R, you should be able to use Synapse to execute any of the models submitted to the recent breast cancer prognosis challenge.  This is a great step forward for the community (though it recapitulates pretty much everything from the more generic world of scientific workflow systems like Taverna).

But still.. a) comparatively very few published predictive models are in Synapse and b) should I really have to know R to answer that question?

respond!

Community Intelligence session at ISMB 2012

Looking for something to do in Long Beach, California on the afternoon of Tuesday, July 17, 2012 ?  Stop by the special session on Community Intelligence in Bioinformatics at this year's conference on Intelligent Systems in Molecular Biology (ISMB)!

We have four distinguished speakers lined up:

• Alex Bateman will discuss the use of Wikipedia in RFAM, PFAM and the rest of science.
• Alex Pico will tell us about the continued expansion of WikiPathways.
• Andrew Su will be on tap to discuss the latest and greatest breakthroughs in the Gene Wiki.
• Firas Khatib will tell us how his team managed to translate protein folding into a game (called Foldit) and that game into an expanding series of scientific discoveries.  

With nearly 7 billion people on the planet and a rapidly increasing number of them connected to the Web, it is time to figure out how to work together effectively, both to solve the problems that we are collectively causing and to solve the problems whose solutions will extend and enrich all of our lives.  Each of these talks will try to help answer the question of how we can translate the incredible scale, connectivity, and creativity of the world's population into scientific progress.  Hope to see you there.

with a flower in my hair

Assuming the fog clears enough to land, I'm heading to San Francisco today to attend the 2011 AMIA Summit on Translational Bioinformatics.   I'll be presenting Tuesday morning about mining structured gene annotations from the text of the Gene Wiki.  Supporters and hecklers would be welcome!

bio2rdf 2 biogps

Last Friday I had the pleasure of having lunch with with Andrew Su of the Genomics Institute of the Novartis Foundation.  Among other things, he introduced me to one of his projects called BioGPS.  BioGPS  is an interesting, kind of minimalist approach to gene-based data integration.  Essentially, it allows you to register gene-related 'plugins' that other users can assemble like an iGoogle home page.  Each plugin amounts to an html-producing url that contains  one of a variety of gene ids as a parameter.  So you might have a plugin for ncbi gene, another for kegg, etc. and they are all displayed together using a very smooth, interactive iframe canvas.

This is clearly useful to many people (they get about 150,000 pageviews/month) but its flexibility is limited by the way it currently accesses information - simply by gathering HTML from existing web pages.  Since many pages have overlapping content there is inevitably (screen)wasteful duplication in the aggregate view.  As others have said before, a little bit of semantic web could go a long way to improving this resource - and, because of the way the system is built and the way SPARQL endpoints work, its very easy to do it.

So, the idea is that you could take a sparql endpoint (that yields html as an option), write a query with a gene as a parameter, capture the url that contains the query and then you have a very specific kind of plugin that only shows precisely what information you want.  By assembling a collection of these you could produce a view on the gene information space that was very precisely tailored to individual needs.

I made a simple example of this pattern with the plugin "OMIM disorders where gene is linked to pathogenesis" which you can see in their plugin library.

It hits this endpoint
http://atlas.bio2rdf.org/sparql

with this query
 PREFIX omim: PREFIX rdfs: select distinct ?OMIM_disorder where { ?s omim:PATHOGENESIS ?o . ?o bif:contains "VEGF" . ?s rdf:type omim:GeneticDisorder . ?s rdfs:label ?OMIM_disorder }

where the VEGF would be replaced by the gene that you were researching.

Here is an example BioGPS view composed of four plugins.  My bio2rdf-sparql example is there on the top right.

To really do this properly, I think you would want to build a little helper application that would help users  compose the queries and would allow for some basic formatting options for presenting the results of these SPARQLing BioGPS plugins.

publisher removal

In my last post, I mentioned offhand that I could not access a PDF (about an ontology for autonomic license management) without paying a $29 fee to Springer.  Though the post was not a direct request for help running around this paywall, I have now received the pdf from 5 different people - several of whom I have never met before.

Clearly, the (micro) community that read that post believe that research articles should be shared in an open-access fashion and that it is both wrong for publishers to charge access fees and right to sabotage the publishers via peer-to-peer exchange of such articles.

I'm wondering if this micro-community (that is you) would feel any differently if the fees paid for such articles went directly to the researchers that produced them rather than to an apparently irrelevant publisher ?  

?

(Also, I wonder about the Radiohead style "tip jar" approach.  This would allow you to read the article first and then contribute a payment afterwords if you felt that the research in the article was worthy of supporting.)

CWA at the YMCA

Somewhere high in the air between New York and Minneapolis, my first stop on my way home, I feel compelled to explain a few things to myself.  Why on Earth have I just spent the last several nights living in the YMCA in Flushing, New York?  Why did I decide to go on my first self-funded professional excursion at a time when I have no income and very little savings?  What did I hope to get and what did the trip deliver? 

The inspiration for this minor adventure was the inaugural meeting of the Concept Web Alliance (CWA) at the New York Hall of Science.  The mission statement of the CWA (written partly at this meeting) is as follows:

“To enable an open, collaborative environment to jointly address the challenges associated with high volume scholarly and professional data production, storage, interoperability, and analyses for knowledge discovery”

The idea is to form an alliance of like-minded researchers and science publishers interested in sharing knowledge in a computationally accessible fashion (i.e. not plain text and such that information from multiple sources can easily be integrated and interacted with).  The basic building block envisioned for these efforts is the ‘triple’ – a Concept-Relation-Concept structure.  (The word ‘triple’ and the interesting new verb ‘triplification’ - meaning to convert some non-triple-structure like text into a set of triples - were almost certainly the most commonly uttered words in the presentations at the meeting.) 

For those familiar with semantic Web standards such as RDF (a generic triple-based language for representing and sharing information) and OWL (a set of languages for representing knowledge in the form of ontologies) it is perhaps most interesting to consider what is not present in a CWA triple and what was not discussed at all in the public portions of the meeting.  The following words never came up ‘description logic’, ‘axiom’, ‘class’, ‘reality’.

The intended materialization of the Concept Web vision - at the moment - thus seems to be an open collection of informal (non logic-based) concept representations, identified by URIs retrievable on the Web, that can be linked together to form semantic networks.  This graph-structure could be queried (e.g. using SPARQL) for the ‘facts’ that it would contain where each such fact would be linked to extensive information about where it came from (who (or what algorithm) suggested it, when, and with what confidence).  Interestingly, this is very similar in its flexibility, lack of built-in reasoning, and its strong notion of provenance tracking to the Freebase model. 

While some of you who like to work with reasoners and OWL or who think that it is better to talk about ‘universals’ and ‘particulars’ than it is to talk about ‘concepts’ may find this lack of formality a little disappointing, I am growing more and more enthusiastic about it because fits the publish-then-filter nature of the Web perfectly.  We see again and again that once information is out there on the Web, its value increases tremendously.  (In fact, many very smart people seem to think that when there is enough text and other unstructured data online that is all we will really need to solve most of our information problems.) By providing a very low barrier for entry and then focusing computer science efforts on handling the noise, complexity, and the conflicts that will inevitably arise (the filter part), I think this triple-publishing approach has great potential to push research in a productive direction.  In particular, I think it will push people to spend more time working on other, more flexible modes of inference that don’t just die when a logical conflict is detected - all that squishy probability stuff that the semantic Web has managed to ignore for so long and that happens to be the stuff that makes almost all interesting AI-like technology work now.  Furthermore, the triple-focus absolutely does not stop groups that participate in the CWA from making use of approaches grounded in formal logics in their own development. 

While there may or may not be good reasons to take one philosophical stance over another when creating knowledge bases, the fact will always remain that there will be conflicts of opinion about this.  When dealing with a small group, it may be possible to convince or force acceptance of a particular world view, but it is not IMHO going to be possible to enforce something as arguable as the philosophy of the representation of the nature of being on the scale of the Web.  By focusing on the smallest possible units, the triples, and leaving the more precise formalizations and the philosophy out of the vision as much as possible, the CWA might make it possible for a diverse, interoperable ecology of knowledge bases to emerge and co-exist.  Ideally, those who wish to make use of, for example - description logic reasoning, should be able to benefit from the pool of URIs in the Concept Web if for nothing other than for the many multi-lingual labels and textual definitions that will be associated with each of them. 

It is still very early days for the CWA – probably far too early to really speculate too far about the consequences of its basic technological approach as even this approach is still very much up for debate.  Still..  I’m not sure exactly how to express this, but the meeting smelled good.  There were enough capable, powerful, enthusiastic people together in that room that seemed to have enough of a shared vision that I think it is very likely that something of that vision is likely to come to life.

So, was it worth it?  I think that it was in the end.  I got an early look at something that might provide solutions to many of the problems that I’ve spent the past several years of my life thinking about (the social construction of a biosemantic web).  I got to reconnect with old friends and make some new ones.  I had a chance to see New York for the first time (the scale of which blew my mind).  And, last but not least, it just might be the last such academic event I get to take part in.  Depending on the choices I make and the dictations of the wheels of fate I may be in the process of losing the privilege of working in the ivory tower.  If it was indeed my goodbye to the community of scholars, it was a good one.

So yes, it was a worthwhile trip and it remains an exciting time to be thinking about the Web - concept or otherwise.  (and the YMCA wasn’t really so bad in the end ;).

You can follow - and perhaps influence - the evolution of the Concept Web on their blog.

Dissertation now online

The full text of my dissertation, "Strategies for amassing, characterizing, and applying third-party metadata in bioinformatics", is now available via UBC's information repository.  It is "manuscript-based" so each of the chapters except the introduction and the conclusion can be read and understood independently.  (So there is really no reason for anyone ever to try to read the whole thing in its entirety.)  

Here is the abstract:

Bioinformatics resources on the Web are proliferating rapidly. For biomedical researchers, the vital data they contain is often difficult to locate and to integrate. The semantic Web initiative is an emerging collection of standards for sharing and integrating distributed information resources via the World Wide Web. In particular, these standards define languages for the provision of the metadata that facilitates both discovery and integration of distributed resources. This metadata takes the form of ontologies used to annotate information resources on the Web. Bioinformatics researchers are now considering how to apply these standards to enable a new generation of applications that will provide more effective ways to make use of increasingly diverse and distributed biological information. While the basic standards appear ready, the path to achieving the potential they entail is muddy. How are we to create all of the needed ontologies? How are we to use them to annotate increasingly large bodies of information? How are we to judge the quality of these ontologies and these proliferating annotations? As new metadata generating systems emerge on the Web, how are we to compare these to previous systems? The research conducted for this dissertation seeks new answers to these questions. Specifically, it investigates strategies for amassing, characterizing, and applying metadata (the substance of the semantic Web) in the context of bioinformatics. The strategies for amassing metadata orient around the design of systems that motivate and guide the actions of many individual, third-party contributors in the formation of collective metadata resources. The strategies for characterizing metadata focus on the derivation of fully automated protocols for evaluating and comparing ontologies and related metadata structures. New applications demonstrate how distributed information sources can be dynamically integrated to facilitate both information visualization and analysis. Brought together, these different lines of research converge towards the genesis of systems that will allow the biomedical research community to both create and maintain a semantic Web for the life sciences and to make use of the new capabilities for knowledge sharing and discovery that it will enable.

About the title..  I ended up using the generic "metadata" rather than something more specific because I needed a way to concisely capture things that range from Del.icio.us tags to classes in the Foundational Model of Anatomy.  "Metadata" seemed to do the job, but it remains a little vague and therefore unsatisfying.  Similarly, "third-party metadata" is broader than necessary.  I don't touch institutionally generated third-party metadata - just what I guess you would probably call "socially generated metadata".

Enjoy.

Tick Tock

I've just returned from watching my second PhD defense in as many weeks. The first one was my younger sister's, and this one was a good friend of mine who started at UBC at the same time as me. I'm reminded of the time in my mid twenties when friends started getting married and disappearing, one by one. The obvious response is to wonder, "when will I be next?". Well it looks like it will be some time yet for me; however, I think I had a little bit of a conceptual breakthrough the other day that may help straighten my path a bit.

I've spent a bit more than half of the past year working with a professor from the library and information sciences on ways to generate useful comparisons of different "types" of information organization system. Throughout this collaboration we've struggled to both communicate with one another and write collaboratively because of the vastly different training and work practices of the social sciences versus the biological sciences.  The realization I had was that his concern is completely to do with information systems at the level of the Type - social tagging, semantic web, subject indexing, and so on, while my concern should really be with information systems at the level of the Instance - Connotea, the Gene Ontology, MEDLINE..   I think this is really the mark that distinguishes bioinformatics from information science,  at its heart it is a completely applied science.  I hope that keeping that up front in my mind will help me keep on track going forward.

For this particular project, now 4 months overdue, I'm now planning to change from building a general purpose framework for defining axes of similarity of semantic indexing systems to conducting a straighforward comparison of Pubmed and Connotea.  This comparison will be specifically designed only to identify ways that the data from Connotea might be used to enhance user's experiences with Pubmed.  I think this will make for more definable goals and easier justifications for the work though I will be using some of the more generic code and concepts we came up with along the way.  

In any case, well done Obi and Erin!

The main problem with LSIDs

As one of the authors of a paper that espoused LSIDs as a great idea and complained about their lack of adoption by the community, I feel obliged to come out and say that I've come to the conclusion that we should not be using them at all (for the time being).

The fundamental problems with LSIDs are the first two symbols in the acronym. If they were simply The Identifier System and thus were not constrained from the beginning to operate in the comparatively small world of the life sciences, the proposal might have stood a chance of working. I think they were really just not ambitious enough. It really is a fantastic bit of work, this paper describing the need for and the implementation of the spec is one of the best works on data integration that I have come across. But, the fact remains that, with a few exceptions, no one has adopted the spec because (IMHO) developers prefer to work with the most widely accepted standards (e.g. those from the W3C) so they don't end up having to re-implement everything when new standards replace the old ones. (There is also the issue of the need for registries but I think that is secondary).

So, until we get to the point where Sun provides a built-in Java class called LSID with methods like getMetaData and getData that I can call with the same ease as the HTTP get method, I think I'll wait with everyone else.