Analysis and Management of Microarray Data

1. Analysis and Management of Microarray Data Dr G. P. S. Raghava

2. Major Applications  Identification of differentially expressed genes in diseased tissues (in presence of drug)  Classification of differentially expressed (genes) or clustering/ grouping of genes having similar behaviour in different conditions  Use expression profile of known disease to diagnosis and classify of unknown genes

3. Management of Microarray Data  Magnitude of Data – Experiments  50 000 genes in human  320 cell types  2000 compunds  3 times points  2 concentrations  2 replicates – Data Volume  4*1011 data-points  1015= 1 petaB of Data

4. Gene expression database – a conceptual view Samples Sample annotations Gene expression matrix Genes Gene annotations Gene expression levels

5. Management of Microarray Data Major Issues  Large volume of microarray data in last few years – Storage and efficient access – Comparison and integration of data  Problem of data access and exchange – Data scattered around Internet – Supplementary material of publications – Difficult for user to access relivent data  Problems with existing databases – Diverse purpose – Developed for specific purpose

6. Management of Microarray Data  Specific Database – Platform (eg.Stanford MA Database; SMD) – Organism (Yeast MA global viewer) – Project (Life cycle database of Drosophila)  Problem with Supplement and MA databases – Lack of direct access – Quality not checked – No standard format – Incomplete data

7.  Comprehensive database server to manage massive amount of Microarray Data – Biomaterial Information – Raw Data & Images – Web Tools (normalization; data viewing; analysis)  Run on local servers allows full management and permission to add and view data  Minimum Information about Microarray Experiment (MIAME)  BASE http://bioinformatics1.uams.edu:8081:/

8. Public Databases Gene Expression data is an essential aspect of annotating the genome Publication and data exchange for microarray experiments Data mining/Meta-studies Common data format - XML MIAME (Minimal Information About a Microarray Experiment)

9. GEO at the NCB ? I

10. Microarray Data Mining Challenges too few records (samples), usually < 100 too many columns (genes), usually > 1,000 Too many columns likely to lead to False positives for exploration, a large set of all relevant genes is desired for diagnostics or identification of

11. Analysis of Microarray Data  Analysis of images  Preprocessing of gene expression data  Normalization of data – Subtraction of Background Noise – Global/local Normalization – House keeping genes (or same gene) – Expression in ratio (test/references) in log  Differential Gene expression – Repeats and calculate significance (t-test) – Significance of fold used statistical method  Clustering – Supervised/Unsupervised (Hierarchical, K-means, SOM)  Prediction or Supervised Machine Learnning (SVM)

12. Low Level Analysis or Preprocessing of gene expression data Scale Transformation Normalization and Scaling Replicate Handling Missing value Handling Flat pattern filtering Pattern standardization

13. Normalization Techniques  Global normalization – Divide channel value by means  Control spots – Common spots in both channels – House keeping genes – Ratio of intensity of same gene in two channel is used for correction  Iterative linear regression  Parametric nonlinear nomalization – log(CY3/CY5) vs log(CY5)) – Fitted log ratio – observed log ratio  General Non Linear Normalization – LOESS – curve between log(R/G) vs log(sqrt(R.G))

14. Classification Task: assign objects to classes (groups) on the basis of measurements made on the objects Unsupervised: classes unknown, want to discover them from the data (cluster analysis) Supervised: classes are predefined, want to use a (training or learning) set of labeled objects to form a classifier for classification of future observations

15. Cluster analysis Used to find groups of objects when not already known “Unsupervised learning” Associated with each object is a set of measurements (the feature vector) Aim is to identify groups of similar objects on the basis of the observed measurements

16. Unsupervised Learnning  Hierarchical clustering: merging two branches at the time until all vari-ables(genes) are in one tree. [it does not answer the question of “howmany gene clusters there are”?]  K-mean clustering: assuming there are K clusters. [what if this assumption is incorrect?]  Self Organizing Maps (SOM) – Split all genes into similar sub-groups – Finds its own groups (machine learning)  Principle Component – every gene is a dimension (vector), find a single dimension that best represents the differences in the data  Model-based clustering: the number of clusters is determined dynamically [could be one of the most promising methods]

17. Average linkage hierarchical clustering, melanoma only unclustered ‘cluster’

19. Supervised Analysis Fisher’s linear discriminant analysis Quadratic discriminant analysis Logistic regression (a linear discriminant analysis) Neural networks Support vector machine

20. Example: Tumor Classification  Reliable and precise classification essential for successful cancer treatment  Current methods for classifying human malignancies rely on a variety of morphological, clinical and molecular variables  Uncertainties in diagnosis remain; likely that existing classes are heterogeneous  Characterize molecular variations among tumors by monitoring gene expression (microarray)  Hope: that microarrays will lead to more reliable tumor classification (and therefore more appropriate treatments and better outcomes)

21. Higher Level Microarray data analysis  Clustering and pattern detection  Data mining and visualization  Controls and normalization of results  Statistical validatation  Linkage between gene expression data and gene sequence/function/metabolic pathways databases  Discovery of common sequences in co-regulated genes  Meta-studies using data from multiple experiments

22. Thanks