Data Mining of GeneChips基因芯片的数据挖掘.ppt

《Data Mining of GeneChips基因芯片的数据挖掘.ppt》由会员分享，可在线阅读，更多相关《Data Mining of GeneChips基因芯片的数据挖掘.ppt（21页珍藏版）》请在汇文网上搜索。

1、Data Mining of GeneChips,Outline,Background Molecular Biology 101 GeneChip Technology Synthesis, Sample Preparation, Scanning Data Flow and Management Multidrug Resistance Preliminary Data Analysis Affymetrix Microarray Suite 5.0 Scoring and Composite Analysis Statistical Algorithms (Absolute/Compar

2、ison Analysis) Data Mining Tool 2.0 (Preview) SOM Algorithm Correlation Coefficient Clustering,Molecular Biology 101,A - T G - C Traditional methods result in “one gene at one time” analysis The need for higher throughput has led to the development of two (competing) technologies,cDNA Spotted Array,

3、cDNA Microarrays utilize various technologies (spotting, piezoelectric, drop-touch) to create arrays of entire genomes using 500-5000 bp oligos representing entire genes Microarrays are scanned using laser microscopy with fluorescent labeled DNA,GeneChip Technology,Utilizes photolithography and comb

4、inatorial chemistry to synthesize microarrays consisting of 25mer oligos representing smaller regions of various genes Probe Cell (PM or MM) Probe Pair (PM + MM) Probe Set ( 18 Pairs) Typical experiment involves isolation of mRNA, synthesis of cDNA, fragmentation, labeling , and hybridization Scan r

5、esults in a raw data file (*.CHP),Data Management,Nat Genet 1999 Jan;21(1 Suppl):51-5 Gene expression informatics-its all in your mine. Bassett DE Jr, Eisen MB, Boguski MS,Use in Functional Genomics,Antimicrobial resistance in bacteria Multi-drug pumps (MDRs) Facilitate extrusion of amphipathic cati

6、ons (toxic) TolC (E. coli), NorA (S. aureus),Evolutionary paradigm: Why do plants continue to produce amphipathic compounds if they are extruded? Perhaps they produce MDR inhibitors that when combined with these normally ineffective antimicrobials provide a strong synergistic effect,Methods,E. coli

7、TolC mutants were challenged with these compounds for several hours and then assayed for gene expression,S. aureus (WT + NorA Mutant) susceptibility to various plant derived compounds were tested in the presence of 5-methoxyhydnocarpin (MDR inhibitor) and potential antimicrobials isolated,Lewis et a

8、l. (2000). Proc. Natl. Acad. Sci. USA 97:1433-1437.,Preliminary Data Analysis,Average probe cell intensity is calculated based upon the 75th percentile of 36 pixels An Absolute Call is based on a decision matrix employing the Positive Fraction, Pos/Neg Ratio, and Log Average Ratio Background is calc

9、ulated from 1 of 16 sectors where the lowest 2% of probe cells are averaged and subtracted from the all probes within that sector Noise (Q) is calculated from the cells used in the background calculation,Comparison Analysis,Global Normalization Multiplying the average intensity of the experimental f

10、ile by a NF resulting in the same average intensity as the baseline Global Scaling Multiplying BOTH data sets by SFs resulting in average intensities equal to a target intensity set by the user Fold Change,Results,TolC + Coumestrol,Cluster Analysis,Cluster analysis helps identify gene expression pat

11、terns in large data sets and groups with similar expression profiles Affymetrix DMT offers 2 methods Self Organizing Map (SOM) (Tamayo et. al., PNAS 1999) Centroid/ K-means Analysis (Golub et. al., Science 1999) Correlation Coefficient Clustering Other methods Hierarchical Agglomeration (Eisen et. a

12、l., PNAS 1998) Super-Paramagnetic (Getz et. al., Physica A 2000),SOM Clustering,Iterative process based on a number of genes (points) in k experiments (dimensions) Initially a grid of centroids is placed onto the k-dimensional space The number of centroids, each representing a cluster, is determined

13、 by the number of rows or columns set by the user (i.e. 3x2=6) The algorithm then adjusts the centroid position towards clusters of points (two-dimensional) Each iteration moves the centroid closer to the target point(s),N= the node being updated P= the data point being considered fi(N) = position o

14、f N at iteration i Np= the target node = distance N moves toward P,SOM Clustering,SOM Clustering,Ref. Gaddy Getz, Weizmann Institute, Israel,SOM Clustering,Ref. Gaddy Getz, Weizmann Institute, Israel,SOM Clustering,Ref. Gaddy Getz, Weizmann Institute, Israel,SOM Clustering,Ref. Gaddy Getz, Weizmann

15、Institute, Israel,Correlation Coefficient,(X,Y)m= mean Avg Diff for probe sets across all analyses (X,Y)i= mean Avg Diff for probe set from analysis i can range from 1 to +1,Example of Clustering,Nat Genet 1999 Jan;21(1 Suppl):51-5 Gene expression informatics-its all in your mine. Bassett DE Jr, Eis

16、en MB, Boguski MS,Conclusions,SOM Clustering Faster than most other clustering algorithms (based on distance) Could potentially break larger clusters Results depend on initial centroid location Correlation Coefficient Fast but will only determine clusters with a positive relationship Choose your method based on your needs,