Micro Array Data Analysis 06

8/3/2019 Micro Array Data Analysis 06

1/55

Microarray Data Analysis

The Bioinformatics side of the bench


2/55

The anatomy of your data files

from Affymetrix array analysis

.DAT= image file (107 pixels)

.CEL= measured cell intensities

.CDF= cell descriptions files (identifyprobe sets and probe set pairs)

.CHP= calculated probe set data

.RPT= report generated from .CHP


3/55

Quality Control (QC) of the

chipvisual inspection

Look at the .DAT file or the .CHP fileimage

Scratches? Spots?

Corners and outside bordercheckerboard appearance (B2 oligo)

Positive hybridization control

Used by software to place grid over image

Array name is written out in oligos!


4/55


5/55

Chip defects


6/55

Internal controls

B. subtilisgenes (added poly-A tails)

Assessment of quality of sample preparation

Also as hybridization controls

Hybridization controls (bioB, bioC, bioD, cre)

E. coli and P1 bacteriophage biotin-labeled cRNAs

Spiked into the hybridization cocktail

Assess hybridization efficiency

Actin and GAPDH assess RNA sample/assay quality

Compare signal values from 3 end to signal valuesfrom 5 end

ratio generally should not exceed 3

Percent genes present (%P)

Replicate samples - similar %P values


7/55

1. Experimental Design

2. Image Analysisscan to intensity measures (raw

data)

3. Normalizationclean data

4. More low level analysis-fold change, ANOVA,

data filtering

5. Data mining-how to interpret > 6000 measures

Databases

Software

Techniques-clustering, pattern recognition etc.

Comparing to prior studies, across platforms?

6. Validation

Microarray Data Process/Outline


8/55

Experimental Design

A good microarray design has 4 elements

1. A clearly defined biological question or hypothesis

2. Treatment, perturbation and observation of biologicalmaterials should minimize systematic bias

3. Simple and statistically sound arrangement that minimizescost and gains maximal information

4. Compliance with MIAME (minimal information about

microarray experiment)

The goal of statistics is to find signals in a sea of noise

The goal of exp. design is to reduce the noise so signals canbe found with as small a sample size as possible


9/55

Observational Study vs.

Designed Experiment

Observational study-

Investigator is a passive observer whomeasures variables of interest, but does not

attempt to influence the responses

Designed Experiment-

Investigator intervenes in natural course ofevents

What type is our DMSO exp?


10/55

Experimental Replicates

Why? In any exp. system there is a certain amount of

noiseso even 2 identical processes yield slightlydifferent results

Sources?

In order to understand how much variation there isit is necessary to repeat an exp a # of independenttimes

Replicates allow us to use statistical tests toascertain if the differences we see are real


11/55


12/55

Technical vs. Biological Replicates

As we progress from the starting material to the scanned

image we are moving from a system dominated by biologicaleffects through one dominated by chemistry and physics noise

Within Affy platform the dominant variation is usually of abiological nature thus best strategy is to produce replicates ashigh up the experimental tree as possible


13/55

Low level data analysis / pre-processing

Varying biological or cellularcomposition among sampletypes.

Differences in samplepreparation, labeling orhybridization

Non specific cross-hybridization of target toprobes.

Lead to systemic differencesbetween individual arrays

Raw Data Quality Control

Scaling

Normalization andfiltering.


14/55

Image Analysis - Raw Data


15/55

From probe level signals to gene abundance

estimates

The job of the expression summary algorithm is

to take a set of Perfect Match (PM) and Mis-Match (MM) probes, and use these to generatea single value representing the estimatedamount of transcript in solution, as measuredby that probeset.

To do this, .DAT files containing array images are firstprocessed to produce a .CEL file, which containsmeasured intensities for each probe on the array.

It is the .CEL files that are analyzed by the expressioncalling algorithm.
http://bioinformatics.picr.man.ac.uk/mbcf/example_ma.jsphttp://bioinformatics.picr.man.ac.uk/mbcf/example_ma.jsp


16/55

MAS 5.0 output files

For each transcript (gene) on the chip:

signal intensity

a present or absent call (presence call) p-value (significance value) for making that

call

Each gene associated with GenBank

accession number (NCBI database)


17/55

How are transcripts determined to be

present or absent?

Probe pair (PM vs. MM) intensities

generate a detection p-value

assign Present, Absent, or Marginalcall for transcript

Every probe pair in a probe SET has

a potential vote for presence call


18/55

PM and MM Probes

The purpose of each MM probe is to provide a directmeasure of background and stray-signal (perhaps dueto cross-hybridization) for its perfect-match partner. Inmost situations the signal from each probe-pair is simplythe difference PM - MM.

For some probe-pairs, however, the MM signal isgreater than the PM value; we have an apparently

impossible measure of background.


19/55

Thank goodness for software!!!

MAS 5.0 does these calculations for you .CHP file

Basic analysis in MAS 5.0, but it wonthandle replicates

Import MAS 5.0 (.CHP) data into othersoftware, Genesifter, GCOS, SpotFire,

and many others


20/55

Signal Intensity Following these calculations, the MAS 5.0

algorithm now has a measure of thesignal for each probe in a probeset.

Other algortihms, ex RMA, GCRMA,dCHIP, PLIER and others have beendeveloped by academic teams to improvethe precision and accuracy of this

calculation In our Exp we will use RMA and GCRMA


21/55

How do we want to analyze

this data?

Pairwise analysis is most appropriate Control vs. DMSO

List of genes that are upregulated ordownregulated

Determine fold up or down cutoffs What is significant?

1.5 fold up/down? 2 fold up/down? 10 fold up/down?


22/55

Normalization - clean data

Normalizing data allows

comparisons ACROSS differentchips

Intensity of fluorescent markers mightbe different from one batch to the other

Normalization allows us to compare

those chips without altering theinterpretation of changes in GENEEXPRESSION


23/55

Why Normalize Data?

The experimental goal is to identify biological variation(expression changes between samples)

Technical variation can hide the real data

Unavoidable systematic bias should be recognized andcorrected

Normalization is necessary to effectively make comparisonsbetween chips-and sometimes within a single chip.

There are different methods of normalization the

assumptions of where variation exist will determine thenormalization techniques used.

Always look at data before and after normalization

Spike in controls can help show which method may be best


24/55

Caveat

There is NO standard way toanalyze microarray data

Still figuring out how to get the bestanswers from microarrayexperiments

Best to combine knowledge ofbiology, statistics, and computers toget answers


25/55

MAS 5.0 GCRMARMA

RMA

GCRMAMAS 5.0

Venn Diagrams


26/55

Data processing is completednow what?

Fold change, ANOVA, Data filtering


27/55


28/55


29/55


30/55


31/55


32/55


33/55


34/55


35/55

Where are we now?

Ran analysis, output is a GENELIST

List indicates what genes are up ordown regulated

p values for t-test

Graphs of signal levels

Absolute numbers not as important here asthe trends you see

Now what????


36/55

What is the first set of genes on our chipsthat will be filtered out?


37/55

Follow the links

Click on a gene

Find links to other databases

Follow links to discover what theprotein does

Now the fun part begins.


38/55

Back to Biology

Do the changes you see in geneexpression make senseBIOLOGICALLY?

If they dont make sense, can you

hypothesize as to why those genesmight be changing?

Leads to many, many moreexperiments


39/55

A Common Language for Annotation of

Genes from

Yeast, Flies and Mice

The Gene Ontologies

and Plants and Worms

and Humans

and anything else!

Gene Ontolog


40/55

Gene Ontology

Objectives

GO represents concepts used to classifyspecific parts of our biological knowledge: Biological Process

Molecular Function

Cellular Component

GO develops a common language applicableto any organism

GO terms can be used to annotate geneproducts from any species, allowingcomparison of information across species


41/55

Sriniga Srinivasan, Chief Ontologist, Yahoo!

The ontology. Dividing human

knowledge into a clean set of categoriesis a lot like trying to figure out where tofind that suspenseful black comedy atyour corner video store. Questionsinevitably come up, like are Movies partof Art or Entertainment? (Yahoo! liststhem under the latter.) -Wired

Magazine, May 1996


42/55

Molecular Function= elemental activity/task

the tasks performed by individual gene products; examples arecarbohydrate bindingand ATPase activity

Biological Process= biological goal orobjective broad biological goals, such as mitosisor purine metabolism, that are

accomplished by ordered assemblies of molecular functions

Cellular Component= location or complex subcellular structures, locations, and macromolecular complexes;

examples include nucleus, telomere, and RNA polymerase IIholoenzyme

The 3 Gene Ontologies


43/55

Function (what) Process (why)

Drive nail (into wood) Carpentry

Drive stake (into soil) Gardening

Smash roach Pest Control

Clowns juggling object Entertainment

Example:

Gene Product = hammer

Bi l i l E l


44/55

Biological Examples

Molecular FunctionBiological Process Cellular Component


45/55

Validation

Not enough to just do microarrays

Usually validate microarray results

via some other technique

rt-PCR

TaqMan

Northern analysis

Protein level analysis

No technique is perfect


46/55

Dynamic Nature of Yeast Genome


47/55

Dynamic Nature of Yeast Genome

eORF= essential

kORF= known

hORF= homologyidentified

shORF= short

tORF= transposonidentified

qORF= questionable

dORF= disabled

First published sequence claimed 6274 genes a # thathas been revised many times, why?


48/55

The Affy detection oligonucleotide sequences are frozen at the timeof synthesis, how does this impact downstream data analysis?

6603

4373

1410

820


49/55

term: MAPKKK cascade (mating sensuSaccharomyces)

goid: GO:0007244

definition: MAPKKK cascade involved in transductionof mating pheromone signal, as described inSaccharomyces

definition_reference: PMID:9561267

Terms, Definitions, IDs


50/55

SGD


51/55


52/55

SGD public microarray data sets available


53/55

for public query


54/55

Homework

1. Go to http://www.yeastgenome.org/and find 3 candidate genes ofknown f(x) and one of undefined f(x) that you might predict to bealtered by DMSO treatment

2. What GO biological processes and molecular mechanisms areassociated with your candidate genes?

3. Where, subcellularly does the protein reside in the cell?

4. What other proteins are known or inferred to interact with yours? Howwas this interaction determined? Is this a genetic or physicalinteraction?

5. Find the expression of at least one of your known genes in anotherpublic ally deposited microarray data set?

1. Name of data set and how you found it?

2. What is the largest Fold change observed for this gene in the public study?

6. Now that you are microarray technology experts can you give me 3reasons why the observed transcript level difference may not beconfirmed through a second technology like RTQPCR?

Suggested Reading
http://www.yeastgenome.org/http://www.yeastgenome.org/


55/55

Suggested Reading

Micro Array Data Analysis 06

Documents