Russell Greiner: Application Pull
Projects (Russ Greiner)

Human Metabolome Project

David Wishart, ...

The mandate of the Human Metabolome Project is to identify, quantify, catalogue and store all metabolites that can potentially be found in human tissues and biofluids at concentrations greater than one micromolar.

We have some preliminary results on chemo-informatics: learning to predict properties of chemicals (here, solubility of metabolites).

Proteome Analyst

Duane Szafron, Paul Lu,  
(David Wishart, Roman Eisner, Brent Poulin, Alona Fyshe, Luca Pireddu, ... )

The Proteome Analyst is a web server designed to predict protein properties, such as general function and subcellular localization, in a high-throughput fashion.

Brain Tumour Analysis Project

Al Murtha, Jörg Sander, Dana Cobzas   Alireza Farhangfar
(Mark Schmidt, Marianne Morris, Alden Flatt, ChiHoon Lee, Ilya Levner)

Oncologists often use MRI scans to identify the tumour region within a brain. Unfortunately, these scans typically reveal only a portion of the tumour. This means the obvious treatment --- irridating all-&-only the visible tumour region --- is not effective, as the "occult" tumour will continue to thrive, to the patient's detriment. Many physicians will therefore irridate a 2cm margin around the visible region; unfortunately, it is not clear that this enlarged region contains only tumour, nor that it contains all of the tumour.

We are therefore seeking a way to predict the location of the occult tumour cells. Given the assumption that these regions that are occult today will become visible later, we are therefore attempting to learn how the tumour will grow, as a function of tumour's location and other properties (size, type). As training data, we have access to the data (MRI scans, etc) from 650 previous patients, where each patient visited between 1 and 11 times, often over years.

This is a joint project with members of the Cross Cancer Institute.

PolyomX: Patient-Specific Treatment for Cancer Patients

The PolyomX Project represents a systematic approach to link the knowledge gained from the human genome project to healthcare. It was established to take advantage of Alberta's unique, province-wide cancer registry. This registry, established in 1975, allows institutes such as the Cross Cancer Institute (CCI) to centralize the diagnosis, treatment and follow-up of all of the province's cancer patients. This centralized cancer registry also allows the CCI researchers access to thousands of anonymous tumor and blood samples as well as anonymized patient profiles. The result is a one-of-a-kind tumor bank that allows large-scale population studies to be performed at a genome-wide level.

These samples are then analyzed using the most recent advances in genomics, proteomics, metabolomics and bioinformatics, towards developing a broad-spectrum molecular analysis of human cancers and their correlations to certain clinical outcomes. We anticipate that this wealth of information will lead to major advances in the fight against cancer, eg, helping us to understand why some patients benefit from a given drug while others suffer adverse reactions.

To date, we have results on...

  • predicting which women are likely to develop breast cancer
  • which men will have long-term bleeding problems after applying radiation treatment to address prostate cancer

We plan to pursue similar studies with the other types of cancers already being tumor-banked here: (breast), lung, ovarian, gastro intestinal/colorectal, lymphoma, prostate and certain leukemia. We will also extend these analysis to include other patient information, including gene-expression data (from microarrays) and metabonomic information (eg, based on NMR-based urinalysis).

This is a joint project with members of the Cross Cancer Institute.

Completed Projects

Adaptive User Interfaces

Benjamin Korvemaker

If I watch what you type for a while, can I predict what you will type next? (This is in the context of commands given to the linux operating system.)

WebIC: An Effective "Complete-Web" Recommender System

Tingshao Zhu, Gerald Häubl, Bob Price, Kevin Jewell

Many web recommendation systems direct users to webpages, from a single website, that other similar users have visited. By contrast, our WebIC web recommendation system is designed to locate "information content (IC) pages" --- pages the current user needs to see to complete her task --- from essentially anywhere on the web. WebIC first extracts the "browsing properties" of each word encountered in the user's current click-stream --- eg, how often each word appears in the title of a page in this sequence, or in the "anchor" of a link that was followed, etc. We have used the data collected from a set of annotated web logs acquired in a user study, to produce user- and site-independent rules that identify which words are likely to appear in a page is an IC-page for a specific session --- eg, of the form:

    "any word with the properties (1) whenever it appears in an anchor, the user tends to follow that anchor, and (2) whenever it appears in the title of a page, the user seldom "backs out" from that page,
    tends to be a word that appears in the IC-page".
Notice this rule deals only with how the word appears in the current browsing session, which mean a word can be an IC-word for one session but not for another, even for the same user. Our empirical results show that the learned classifier works effectively --- that is, it can accurately identify which words will appear in the IC-pages.

We are beginning two new studies --- one to try out our system in the context of a paticular set of e-commerse tasks, and the other, both to collect a larger set of data of how a wider selection of users browse the web for a larger set of contexts, and also to explore various ways to use these IC-words to determine which pages best satisfy the user's information need (based on the context of her browsing behavior). We will also attempt to find user-specific rules, to determine if different users have significantly different mappings from browsing properties to "IC-ness".