Practical Tools For Innovation
O'Reilly Bioinformatics Technology
January 28-31, 2002 -- Tucson, AZ
AI Techniques for Microarray Data Analysis
Ben Goertzel, CEO, Cognitive Bioscience
Date: Tuesday, January 29
Location: Canyon I
Microarray technology (including gene chips, spotted microarrays, and other related devices) provides unprecedented powerful insights into gene expression; and it can soon be expected to do the same for protein expression. However, the practical analysis of the data derived from microarray experiments is fraught with difficulties, stemming from the large volume and high noise level of the data sets, and the subtlety of the underlying data patterns. Standard statistical techniques, such as clustering and principal components analysis, are widely used but give highly uneven results. However, more advanced data analysis methods can yield greater insights, being less easily misled by noise, and more able to draw diverse background knowledge into the data analysis process.
In this talk we will discuss three particular examples of data analysis techniques from the AI field that can be fruitfully applied to the analysis of microarray data:
Supervised learning techniques for data categorization. Well-known methods like SVM’s and decision trees can be used, for instance, to learn rules distinguishing cancerous from noncancerous cells. Appropriate construction of “feature vectors” makes a large difference in performance.
Clustering using a priori background knowledge. Though popular, clustering is generally a poor data analysis method, but invocation and appropriate merging of knowledge from relevant databases can significantly improve its performance.
Probabilistic rule inference from microarray time series data. While exact dynamical rules cannot be inferred from noisy and incomplete data, probabilistic rules approximately governing the dynamics can be.
oreilly.com Home |
Conferences Home |
Bioinformatics Conference Home
© 2001, O'Reilly Media, Inc.