HypWeb: Hypothesis Web System

HypWeb: Hypothesis Web System


HypWeb is a system for developing hypotheses within the CNP (Consortium for Neuropsychiatric Phenomics). HypWeb currently allows specification of hypotheses involving the LA2K Database. Specifically, it provides three main subsystems:

ViVA Viewer -- the easy-to-use Visualizer for LA2K


ViVA Viewer is a HypWeb system that permits visualization of hypotheses. With it, researchers can make conjectures about "effects" -- hypotheses asserting that some experimental measure yields different values on different groups of subjects in the LA2K/LA5C data. ViVA Viewer is a service that uses a battery of data exploration tools to provide intuition about rough hypotheses about this data. The input form looks like this:
http://datamining.cs.ucla.edu/hypweb/figures/LA2K_Viewer_Interface.png
As this page indicates, an input query to HypWeb is a hypothesis space -- a rough hypothesis that includes three things: From these three things, HypWeb generates a hypothesis web -- a web site that integrates relevant information about the space of hypotheses. HypWeb generates automated data visualizations and links the results with relevant literature and published findings; the resulting web site is a kind of review or report, but it can be developed over time.

Sample ViVA Viewer Session

Suppose we want to study the effect of alcohol abuse on response time and accuracy of LA2K subjects. With HypWeb we can define a hypothesis space for these subjects that we can then explore.

For this space of hypotheses, we might want to focus on the AGE of subjects. We can select AGE groups with the LA2K SUBJECT GROUPS menu:
http://datamining.cs.ucla.edu/hypweb/figures/LA2K_Viewer_Group_Specification.jpg

This screen shot shows that we have selected the AGE groups:
(20 or less, 21-25, 26-30, 31-35, 36-40, 41-45, 46-50, 51-55, 56-60, 61 or more).
By selecting it, we are requesting the 10 subpopulations indicated in this list to be used in our visualizations.

With these groups, we then specify which table in the database we are interested in. The ALL EPRIME MEAN ACC is a compendium of all Accuracy results from interactive EPRIME tasks:
http://datamining.cs.ucla.edu/hypweb/figures/LA2K_Viewer_Effect_Specification.png
This table tells us all key accuracy results from a large battery of tasks, and so it is offers compact summary of a great deal of data.
Finally we can select exploratory data analysis schemes of interest from the HYPOTHESIS EXPLORATION METHODS menu:
http://datamining.cs.ucla.edu/hypweb/figures/DefaultHypothesisExplorationMethods.png
The methods checked here are the default ones, but we can select any we like. The selections (Group Sizes, Histograms, Scatterplots, Correlation Ellipses) will be used to generate results.

The page generated by compiling this information starts with a summary of our specification of the space of hypotheses here.
http://datamining.cs.ucla.edu/hypweb/figures/LA2K_Viewer_Result_Description.png


By clicking on any image you can get PDF for it, and for related images that are not shown. For example, histograms showing results for each of the six groups can be obtained by clicking on the histogram result image:
http://datamining.cs.ucla.edu/hypweb/figures/LA2K_Viewer_Result_Statistics.jpg

All results are color-coded by group, and the interface allows you to select colors if you want that. Notice that there are only 7 groups shown, although we selected 10 earlier; it turns out that, with the Healthy Control subjects that have complete files, 3 of these groups have no members.
We can then explore the various method outputs by clicking on an image to obtain all the PDF output:
http://datamining.cs.ucla.edu/hypweb/figures/LA2K_Viewer_Result_Screen.jpg


Click here to use the Hypothesis Space Viewer

How it Works

ViVA Viewer first extracts the desired table from the database, and joins it with Group information. The result of this join indicates which group each row in the table is in. It then executes a script in R that runs all the requested analysis/visualization methods.

Some caveats:

ViVA Explorer -- the advanced Hypothesis Explorer for LA2K


ViVA Explorer is another HypWeb system that permits analysis of hypotheses. With it, researchers can make conjectures about "effects" -- hypotheses asserting that some experimental measure yields different values on different groups of subjects. ViVA Explorer is a service that uses a battery of data exploration tools to provide intuition about rough hypotheses about this data. The input form looks like this:
http://datamining.cs.ucla.edu/hypweb/figures/LA2K_Explorer_Start_Page.png
As this page indicates, an input query is a hypothesis space -- a rough hypothesis that includes three things: From these three things, HypWeb generates a hypothesis web -- a web site that integrates relevant information about the space of hypotheses. HypWeb generates automated data visualizations and links the results with relevant literature and published findings; the resulting web site is a kind of review or report, but it can be developed over time.

Sample ViVA Explorer Session

Suppose we want to study the effect of alcohol abuse on response time and accuracy of LA2K subjects. With HypWeb we can define a hypothesis space for these subjects that we can then explore.

For this space of hypotheses, we might want to focus on gender, and also on differences between Hispanic and non-Hispanic subjects. The SCID diagnoses in LA2K provide 3 classifications regarding Alcohol (No Diagnosis, Alcohol Abuse, Alcohol Dependence), so altogether we are considering 4 different groups of subjects: We can specify these groups with the LA2K GROUP DEFINITIONS menu:
http://datamining.cs.ucla.edu/hypweb/figures/LA2K_Explorer_Group_Specification.jpg

Notice that the checked boxes specify the six groups we want: the first column specifies Group 1 (Male, Hispanic, non-Alcohol) and the last column specifies Group 4 (Feale, non-Hispanic, non-Alcohol).

This interface is extremely flexible, and permits very general definition of groups. The i-th column can be checked with whatever features we want for the i-th group. To help keep things straight, the color of each column identifies a group.

With these group definitions, we then specify which effect variables we are interested in. These variables are indicators (field names) in the database, and they can be chosen from the HYPOTHESIZED EFFECTS OVER THESE GROUPS menu:
http://datamining.cs.ucla.edu/hypweb/figures/LA2K_Explorer_Effect_Specification.png

Finally we can select exploratory data analysis schemes of interest from the HYPOTHESIS EXPLORATION METHODS menu:
http://datamining.cs.ucla.edu/hypweb/figures/DefaultHypothesisExplorationMethods.png

The selections (Histograms, Scatterplots, Parallel Coordinates, Correlation Heatmap, Correlation Ellipses, and Principal Components Analysis) are used in generating results.

The page generated by compiling this information starts with a summary of our specification of the space of hypotheses here.
http://datamining.cs.ucla.edu/hypweb/figures/LA2K_Explorer_Result_Description.png

Things are pretty self-explanatory after this.

By clicking on any image you can get PDF for it, and for related images that are not shown. For example, histograms showing results for each of the six groups can be obtained by clicking on the histogram result image:
http://datamining.cs.ucla.edu/hypweb/figures/LA2K_Explorer_Result_Statistics.jpg

All results are color-coded by group, and the interface allows you to select colors if you want that. ViVA Explorer offers you many ways to explore data.
http://datamining.cs.ucla.edu/hypweb/figures/LA2K_Explorer_Result_Screen.jpg


Click here to use the Hypothesis Space Explorer

Exploration Performance

ViVA Explorer first extracts the relevant information from the database, based on the specifications for groups and effect variables. It then executes a script in R that performs all the requested tests.

Some caveats:

HypSpace -- the HypWeb Hypothesis Space manager


HypSpace is a new system within HypWeb for defining, editing, and exploring Hypothesis Spaces. These spaces are rough descriptions of hypotheses. (such as Male non-Hispanic subjects or ADHD Male Hispanic subjects) involving LA2K task/test measures -- can use HypSpace to manage information about different measure values among these groups.

Currently HypSpace can both store this information and perform automated data exploration, generating visual presentations of this data. These presentations are linked a browsable report with references to relevant literature. This report is actually a web site -- a hypothesis web -- that then subsequently be a hub for collaborative hypothesis development.
An example of results from querying the database of is shown in the figure below, The query requested all CNP hypotheses involving `impulsivity', and they are displayed both as a table and as a graph. Where the graph allows visualization, a tabular view is also important for conveying detail:

HypSpace is a hypothesis space manager. It provides the ability to store, retrieve and analyze hypotheses as objects. Essentially, it provides the ability to view hypotheses from a high-level, collaborative or administrative perspective, and this is a novel capability for large-scale interdisciplinary science.

Click here to use the Hypothesis Query system


Acknowledgements




Output information here is provided "as is", and with no representation or warranty expressed or implied by any parties.



To print in color (even to PDF), your browser must be set properly:
FirefoxPage Setup → Print Background (Colors & Images)
SafariPrint; then, in print options: Safari → Print backgrounds
IETools → Internet Options → Advanced → Printing → Print Background Colors.


NIH_logos

Consortium for Neuropsychiatric Phenomics

UCLA Semel Institute, Room C8-849    760 Westwood Plaza, CA 90095
Phone: 310-825-9474    Fax: 310-825-2850    E-mail: rbilder@mednet.ucla.edu



Copyright © 2007–2009 Consortium for Neuropsychiatric Phenomics    All Rights Reserved.