Mbirn: FBIRN:BrainMap Discussion

Summary of the FBIRN/BrainMap initial meeting, March 17 and 18 2005, at UCI. (i.e., Cryptic Notes for internal use only)

What was discussed, in roughly this order:

1) Short description of FBIRN and what it is trying to do; the idea of the mediated query of multiple databases; each DB has to be registered with the Mediator; starting with a single DB design.

Have to have one DB schema that doesn’t change with each experiment in order to be maximally efficient. Extensible is the keyword (e.g., we can put the optical imaging data from cancer without changing the schema). Currently this changes as we debug it, find out what is working and what doesn’t.

In order to have a DB this general, the queries become incredibly complicated and touch a lot of tables—this is a performance issue, esp. querying multiple DB around the country and having to compile the information that comes back. The first version of the Mediator was incredibly slow; the second version will be hopefully faster.

That’s databases; then there’s actually storing the data and the idea of the SRB (not a file system; a collection management system). And the metadata attached to the files on the SRB; we’ve standardized on a minimum metadata set at the various levels that have to be included with data as it is uploaded.

QA phantom portal description and explanation of grid/remote application launching—so anyone can do analyses without using their own desktop CPU/space, etc.

2) Point of our interactions: 1) Searching Brainmap as well as ours; 2) using their interface to describe/input experiments. Peter F. at the Human Brain Project: discussions there indicated that these databases have to be interoperable, and we agree.

3) XML issues: to describe images and metadata so that it is understandable to the DB and without the DB. Working with NIfTI to develop a descriptive schema of the subject, their measures, their data, the conditions, timing, etc.—everything that is needed to use the data (and then some).

BrainMap DB—special issue of meta-analysis method using ALE (activation likelihood estimator/tion) based on coordinates (usually max activation, but can be center of mass of thresholded clusters). Can highlight unexpectedly consistent clusters in unlikely regions, or pull up something of the circuitry.

--Talairach/MNI space—they can/will transform the coordinates upon upload to a consistent Talairach space (so they can interact with the TD). David’s SPM toolbox: Could use that to package up the results for submission?

--could interact with our derived data; we could pull thresholded maps that would be comparable to what is in their database.

XML detailed discussion… The need to store all the info including stimuli details (where on the screen was the cross? What color was it and what color was the background?) and imaging acq details and formats etc. All that metadata goes into the XML wrapper.

E.g., activation type is a statistics type with extra information. Includes the pre-processing and processing steps that tell you how you got to the activation map, all the output from SPM (effective smoothness, p value corrections etc.), and the option for thresholded clusters and their characteristics (extents, p values, units), etc. and the option for labeling it anatomically.

Peter K. has one of our SIRP datasets and was working on a description of in Brain Submit. He’s also working on a description of condition timing for BrainMap which it doesn’t already have (in context of NIfTI).

4) After lunch: Discussion of BrainMap’s Search and View; finding papers, plotting results (turning everything into Talairach space)…

5) And BrainMap Submit.

Paradigm class and domain are based on Angie’s expertise. Entries are done by hand by Angie and a research assist. and a class of grads on meta-analysis who have to enter N papers that semester.

Condition Name and Experiment Name is not queryable. Currently chosen by how the authors refer to it.

Using Stimulus, Response, and Instruction to classify experiments (have dropped Rate and some other details). This is not sufficient for what we need (unless we only want to be able to do meta-analysis, rather than combining data); but there is a trade-off with the time needed to describe the experiment!

External variables—blood pressure, breathing rate, EEG, startle, --and behavioral data (only behavioral data usually relevant for SPMs).

Note that “experiment” is actually a *contrast*. Not the usual way of thinking about things!

Behavioral Domain: what you are studying (emotion, language, etc.) Paradigm Class: how we are studying it (visual discriminations, word generation, etc.) There is no current ontology or hierarchy of concepts! Just lists of what is in the papers in their database—Angie currently cleaning it up.

6) Here’s what we need: A domain expert on each subfield (attention, memory, language, vision, interoception, etc.) to talk about these areas; and an abstract ontologist to help layout the form, determine the distinct levels of description, etc.

7) Worked through UMLS as an option –it’s a tool for standardized concepts but may be too limited for what we’re doing? And Arrowsmith—useful for the meta-analysis searches but not building a taxonomy?

8) On Friday 3/18 The fBIRN XML schema and the BrainMap ENT schema were laid out in detail, and a one to one mapping of the fields from one to the other was developed. Certain fields are missing in each and a plan was developed to incorporate the missing info.

Action Plans:

• 1) JT: talk to Carol Bean about including Angie as the link to Brain Map in the Ontology Task Force/subcommittee of the Task Force/whatever
• 2) JT: Monday afternoon see if there’s a group to review the Behav. Domains, Paradigm Classes, and Context lists as a way of describing Experimental Contrasts. (Brain Submit)
• 3) A’s: Using an updated FBIRN XML schema, they will try to export an ENT into the FBIRN XML format, starting with SIRP.