Stanford University Boolean Analysis of Large Gene-expression Datasets Debashis Sahoo PhD Candidate, Electrical Engineering Joint work with David Dill,

Stanford University

Boolean Analysis of Large Gene-expression Datasets

Debashis SahooPhD Candidate, Electrical Engineering

Joint work with David Dill, Andrew Gentles, Rob Tibshirani, Sylvia Plevritis

Stanford University

Outline

Standard microarray work flowData collection and preprocessingBoolean analysisBiological insightsConclusion and future work

Stanford University

Microarray Work Flow

mRNA Hybridization Scanning

Image processingNormalizationData analysis

Stanford University

Data Collection

There are thousands of microarray freely available

GEOArrayExpressSMDCelsius

Stanford University

Preprocessing

Get original RAW CEL files for one platform together.

Typical number of CEL files : 2,000-11,000

Use RMA to normalize the CEL filesNeed a memory efficient algorithmGenerates expression values for each probeset

Stanford University

Existing Methods

Correlation analysisConditional probabilityMutual information

Stanford University

Boolean Analysis

Get RAW Data Normalize

Determine thresholds

Discover Boolean relationshipsNew Biology

Stanford University

Example

Stanford University

Determine threshold

Sort the gene expressionsUse StepMiner to determine the threshold

Stanford University

Determine threshold

Its hard to determine a threshold for this gene.StepMiner usually puts a threshold in the middle for this case.

Stanford University

Discover Boolean Relationships

Analyze scatter plots between two genes.Divide the space into four different regions using the thresholds (quadrants).Determine sparse quadrants.Determine the Boolean relationships.

WNT5A high PAX5 low

0

1 3

2

Stanford University

Statistical Tests

Compute the expected number of points under the independence model

Compute maximum likelihood estimate of the error rate

statistic =(expected – observed)

expected√

a00

(a00+ a01)

a00

(a00+ a10)+( )1

2error rate =

a00

a01 a11

a10

Stanford University

Boolean Relationships

Tightly co-regulated genes forms two sparse quadrants.There are six possible Boolean relationships

EquivalentOppositeA low B lowA low B highA high B lowA high B high

Stanford University

Boolean Relationships

Equivalent

Opposite

PTPRC low CD19 low XIST high RPS4Y1 low

COL3A1 high COL1A1 highFAM60A low NUAK1 high

Symmetric Asymmetric

Stanford University

Boolean Implication Network

Directed graphNodes:

For each gene AA highA low

Edges:A high to B low

A high B low

A high

B low

A low

B high

C high

C low

Stanford University

New Biology

This slide is under construction!!

Stanford University

Biological Insights

Gender Organ Tissue

Development Differentiation Co-expression

Stanford University

Example Application

ImmunologyB Cell differentiationGoal:

Discover genes that mark unique B Cell precursors

Stanford University

Differentiation Tree

Hematopoietic stem cell differentiation is a tree

Root: HSCLeaf

LymphocytesB Cell, T Cell, NK cell, Dendritic cell

ErythrocytesGranulocytes: Basophil, Neutrophil, EosinophilMonocytes: Dendritic cellThrombocytes

Stanford University

KIT high A high B low B220 low CD19 low

KIT

A

B

B22

0C

D19

A high

B low

Stanford University

Conclusion

Boolean analysisDirectly visible on the scatter plot.Enables discovery of asymmetric relationship.Follow biology.Potential application to Immunology

Future workCancer progression New biology

Stanford University

Acknowledgements

The Felsher Lab:Natalie WuCathy ShachafDean Felsher

Funding: ICBP Program (NIH grant: 5U56CA112973-02)

Leonore A HerzenbergJames Brooks Joe LipsickGavin SherlockHoward ChangStuart Kim

Stanford University

The END