Exploratory Analysis of Multimodal Data

1
Exploratory Analysis ofMulti-modal Data

• Jyoti Shankar
• Division of Epidemiology

2
What is Multi-modal Data?
Classical Categorical Data
Newer data from Microarrays


3
What exactly is a Microarray?
4
What exactly is a Microarray?
5
What kind of data is it?

• In Classical Clinical Studies
• ?1000 patients ------- 100 data points
• In a Microarray
• ? 100 patients -------1000s of data points
• HIGH-DIMENSIONAL DATA
• ? More measurements than there are samples

6
The vant Veer Breast-Cancer Dataset

• Study Hypothesis
• Gene Expression signatures can predict clinical
  outcomes of breast cancer patients much better
  than the classical categorical clinical
  predictors in this case, Lymph Node Status
  Histological Classification.

7
The vant Veer Breast-Cancer Dataset
8
The vant Veer Breast-Cancer Dataset
9
The vant Veer Breast-Cancer Dataset

• Women under 55 years of age, with a LN ve breast
  cancer that has a poor prognosis signature, have
  a 28 fold OR (CI 7 - 107), (P 10-8) to
  develop distant metastasis within 5 years
  compared to those with good prognosis signatures.
  
• (OR 15 (CI 4 56) in validation set.

10
Microarray Conclusions

• Current Guidelines 90 of LN ve young breast
  cancer patients are candidates for adjuvant
  treatment.
• 70 - 80 would NOT have developed distant
  metastases. Risk of side-effects.
• Tailor adjuvant treatment, reduce costs and
  morbidity.

11
Clinical Results from the Breast Cancer Dataset
Categorical Data
Area under the ROC Curve
12
Best Approach Combining Multimodal Data?

• Why waste information?
• Use newer and more intuitive means of exploratory
  analysis.
• Example The reef SOM ? a metaphoric display of
  biomedical multi-modal data for an intuitive
  exploratory analysis.

13
The SOM System of Exploratory Analysis
14
The reef SOM System of Exploratory Analysis
15
The Fish Glyphs
16
The Fish Glyphs
17
What the data looks like
18
Final Model
19
Advantages

• The reef SOM is multi-modal itself!
• Intuitive, simple to train, entertaining!
• 2 modalities the geology modus the fauna
  modus ( additional modi as reqd)
• Easily interpreted!
• Speedy detection of structural pattern.

20
Thanks for your attention!

• Fusing Biomedical Multi-modal Data for
  Exploratory Data Analysis Christian Martin,
  Harmen grosse Deters, Tim Wilhelm Nattkemper
  ICANN 2006, Part II, LNCS 4132, 798-807, 2006
• http//www.techfak.unibielefeld.de/ags/ani/projec
  ts/somreef/
• REEFSOM - A Metaphoric Data Display for
  Exploratory Data Mining Harmen grosse Deters,
  Wiebke Timm, Tim Wilhelm Nattkemper Brains, Minds
  and Media, 2, Apr / 2006Journal intern ID bmm305
• van 't Veer LJ, Dai H, van de Vijver MJ, He YD,
  Hart AA, Mao M, Peterse HL, van der Kooy K,
  Marton MJ, Witteveen AT, Schreiber GJ, Kerkhoven
  RM, Roberts C, Linsley PS, Bernards R, Friend SH.
  Gene expression profiling predicts clinical
  outcome of breast cancer. Nature. 2002 Jan
  31415(6871)530-6
• Eden P, Ritz C, Rose C, Ferno M, Peterson C.
  "Good Old" clinical markers have similar power in
  breast cancer prognosis as microarray gene
  expression profilers. Eur J Cancer. 2004
  Aug40(12)1837-41.
• Quackenbush J. Microarray analysis and tumor
  classification. N Engl J Med. 2006 Jun
  8354(23)2463-72. Review