MDMS-A Web Tool to Manage

1
MDMS-A Web Tool to Manage Analyze Gene
Expression Microarray Data

• Sachin Mathur

2
Overview

• Steps in analysis of Gene Expression Microarray
  Data
• Preprocessing
• Filtering
• Statistical Analysis
• Machine Learning Data Mining (Clustering)
• Functional Analysis
• Data Analysis features in MDMS
• Workflow in MDMS
• Analysis of Early Lung Development dataset using
  MDMS
• MDMS Demo

3
Steps in Microarray Data Analysis
Analysis of Data Deriving Knowledgebase from
Datum and mining Information from the
knowledgebase
4
Steps in Microarray Data Analysis

• Image Quantification
• Check for artifacts, Segmentation
• Extraction of expression values of genes
• Preprocessing
• Background Correction
• Normalization
• Summarization
• MAS5, RMA, GC-RMA, DChip

www.swegene.org/SWEGENE_microarray_eng.php?Id18
5
Steps in Microarray Data Analysis

• Filtering
• About 10-50 of the genome is not expressed in a
  given tissue
• Aim is to isolate the genes that are expressed
• Also helps in more accuracy in statistical
  significance tests
• Specific Non-specific filtering
• Filter of Presence/Absence calls
• Filter on expression signal, Variability in gene
  expression

6
Steps in Microarray Data Analysis

• Statistical Analysis
• Many genes will be expressed to perform many
  routine tasks in the cell
• Aim is to isolate genes responsible for
  phenotypic variation
• Interesting Vs Random
• Variant significance tests T-Test, ANOVA
• Multiple Testing Correction

7
Steps in Microarray Data Analysis

• Machine Learning Approaches Data Mining
• Small changes in gene expressions can
  collectively regulate an important pathway, which
  by themselves may not be statistically
  significant
• Limitations with fewer replicates and fitting
  approximate models on data during statistical
  analysis
• Aim is to find significant patterns in the data
  set.
• Periodic, Time-lagged, cyclic
• Machine Learning approaches mine data for
  information data mining using computational and
  statistical techniques (Eg Clustering)

8
Functional Analysis

• Functional Analysis
• Given a statistically significant pattern or list
  significant of genes, how significant is it
  biologically?
• Aim is to find genes that are responsible for the
  phenotypic condition
• Extracting annotations and finding functionally
  similar genes.
• Gene Ontology
• Gene set enrichment, relating genes to known
  pathways

http//cardioserve.nantes.inserm.fr/ptf-puce/image
s/camembert_go.gif
9
Data Analysis Features in MDMS

• All data analysis features in MDMS are
  implemented through Bioconductor Package
  (http//www.bioconductor.org)
• Covers many aspects of data analysis for
  Gene-Expression, SNP, Custom made arrays
• Many different tests for quality control,
  preprocessing, filtering, statistical analysis,
  machine learning and functional analysis
• Large user community, helpful mailing lists, used
  by many labs in many countries
• Tutorials are available on the website and
  hands-on training is also available.
• Better than all available packages in terms of
  coverage of data analysis aspects.
• Open Source

10
Data Analysis Features in MDMS

• MDMS supports Affymetrix Gene Expression arrays
• No Image Quantification (usually done at
  microarray facility)
• Quality Control
• 3/5 bias
• Detection calls
• Background signals
• Correlation coefficients between arrays

11
MDMS - Preprocessing

• Preprocessing
• MAS5 Default Affymetrix normalization
• RMA Robust Multichip Analysis
• GC-RMA, DChip (Li-Wong)
• MAS5 and RMA are highly recommended
• Available literature shows significant advantages
  of RMA over MAS5

12
MDMS - Filtering

• Filtering
• Expression value cut-off
• Eg. All genes gt 200
• Detection calls
• Eg. All genes that are detected as Present
• Fold Change
• Eg. All genes that have gt 2 fold or less than -2
  fold
• Inter-Quartile Range (1st 3rd quartiles)
• For genes that show higher variability
• All analysis is done on a log 2 scale

13
MDMS Statistical Analysis

• Significance Tests
• LIMMA (Linear Models of Microarrays)
• SAM (Significance Analysis of Microarrays)
• EBAM (E-Bayes Analysis of Microarrays)
• Correction for Multiple Testing
• FDR, Bonferroni, Holms correction
• Machine Learning
• Clustering
• Hierarchical Clustering, K-Means, Self Organizing
  Maps.

14
MDMS-Functional Analysis

• Functional Analysis through GOAPhAR
• Gene Annotation
• Protein Annotation
• Biological Pathways
• Gene Ontology Annotation
• Protein Interaction Evidence
• All gene lists generated using the data analysis
  options can be saved in the database for future
  use. These can be also downloaded as text files.

15
MDMS-WORKFLOW
Microarray Core
USER
Data Repository
Software Rat2302, Hg133U
MDMS Database
Preprocessing
Filtering
Statistical Analysis Machine Learning
GOAPhAR
Annotation
16
Data Analysis Example

• Data set specifications (GSE3541)
• The aim of the study is to find genes involved in
  early lung development.
• Mechanical Stress was applied to fetal type II
  endothelial cells taken from 19 day old rat
  embryos
• Data set Processing
• Data was preprocessed by MAS5
• Expression gt 200, Invariant change between pairs
  of control experiment samples gt 50 (75
  filtered)
• SAM statistical method was used to find
  significant genes (92 genes, 63 up and 29
  down-regulated)
• 34 up-regulated genes were selected for further
  analysis

17
Biological Significance of Clusterings

• K-Means was applied to 34 genes, with K2, 3, 4,
  .,29
• Random clusterings were generated for K
  2,3,4,29 to compare the statistical clusterings
  to random
• Biological significance scores were calculated
  for all clusterings.
• A z-score and P-value was calculated for each K
  value

18
Probeset Cluster Gene Symbol Gene Title
1377064_at 0 Dusp6 dual specificity phosphatase 6
1386908_at 0 Glrx1 glutaredoxin 1 (thioltransferase)
1367811_at 1 Phgdh 3-phosphoglycerate dehydrogenase
1368489_at 1 Fosl1 fos-like antigen 1
1368789_at 1 Acpp acid phosphatase, prostate
1375213_at 1 Pck2_predicted phosphoenolpyruvate carboxykinase 2 (mitochondrial)
1387925_at 1 Asns asparagine synthetase
1368990_at 2 Cyp1b1 cytochrome P450, family 1, subfamily b, polypeptide 1
1370690_at 2 Hspa9a_predicted heat shock 70kDa protein 9A
1375025_at 2 Gm963_predicted Gene model 963
1376134_at 2 RGD1307789 similar to hypothetical protein MGC3207
1387408_at 2 Siah2 seven in absentia 2
1367741_at 3 Herpud1 homocysteine-inducible, endoplasmic reticulum stress-inducible, ubiquitin-like domain member 1
1368391_at 3 Slc7a1 solute carrier family 7 (cationic amino acid transporter, y system), member 1
1372665_at 3 Psat1 phosphoserine aminotransferase 1
1375964_at 3 Psph phosphoserine phosphatase
1368203_at 4 Scnn1a sodium channel, nonvoltage-gated 1 alpha
1369868_at 4 Iag2 implantation-associated protein
1371900_at 4 Cugbp1 CUG triplet repeat, RNA binding protein 1
1373412_at 4 Nt5c3_predicted 5'-nucleotidase, cytosolic III (predicted)
1387088_at 4 Gal galanin
1389725_at 5 Tm7sf2 transmembrane 7 superfamily member 2
1367795_at 6 Ifrd1 interferon-related developmental regulator 1
1368582_at 6 Slc7a3 solute carrier family 7 (cationic amino acid transporter, y system), member 3
1369772_at 6 Slc6a9 solute carrier family 6 (neurotransmitter transporter, glycine), member 9
1370080_at 6 Hmox1 heme oxygenase (decycling) 1
1372042_at 6 Cmtm3_predicted CKLF-like MARVEL transmembrane domain containing 3
1377112_at 6 Cda_predicted cytidine deaminase
1398771_at 6 Slc3a2 solute carrier family 3 (activators of dibasic and neutral amino acid transport), member 2
1374034_at 7 Cars_predicted cysteinyl-tRNA synthetase
1374221_at 7 Slc29a3 Solute carrier family 29 (nucleoside transporters), member 3
1374324_at 7 --- Transcribed locus
1372601_at 8 Atf5 activating transcription factor 5
1386888_at 8 Eif4ebp1 eukaryotic translation initiation factor 4E binding protein 1
19
Biological Significance of Clusterings

• The study found that genes related to amino acid
  synthesis, amino acid transport and sodium ion
  transport contributed to lung development.
• 1 gene for sodium ion transport
• 4 genes for amino acid transport were found in 2
  clusters
• 4 genes for amino acid synthesis were found in 2
  clusters

20
MDMS

• Demonstration - Using MDMS to analyze data

21
MDMS

• Questions, comments, suggestions