Quantification of cross hybridization on oligonucleotide - - PowerPoint PPT Presentation

Quantification of cross hybridization on

• ligonucleotide microarrays

Li Zhang

• Dept. of Biostatistics, UT MDACC, Houston, TX 77030

DNA/RNA duplex on oligonucleotide microarrays

The probe is a 25-mer DNA oligo:

ATCAGCATACGAGAGAATGATGGAT ||||||||||||||||||||||||| AAUAGUCGUAUGCUCUCUUACUACCUAGC

cRNA fragment in solution expressed from a targeted gene

ATCAGCATACGACAGAATGATGGAT

Modes of binding on probes

• 1. Gene-specific binding:

(Mismatches=0)

• 2. Cross hybridization:

(I) Non-specific binding (Mismatches>5) (II) Binding of related sequences (0<Mismatches<5 )

Gene-specific binding energy: Non-specific binding energy: Binding energy = f(distance, interacting partners)

Binding energies

) , (

1 +

∑

= Ε

i i i

b b ε ω

) , ( * * *

1 +

∑

= Ε

i i i

b b ε ω

Thermodynamic model of binding on a probe

∑

− =

2 ,

) ln (ln

ij

• bs

ij

I I T

B e N e N I

ij ij

E E j ij

+ + + + =

*

1 * 1

Minimization of T

• Energy parameters
• B, N*, Nj
• N*, B are the same on a microarray;
• Nj is the same in a probe set.

Probe Signal: Fitness: Constraints:

Weight factors reflect dynamic properties

• f binding on the probes

Non-specific binding(PM & MM) Gene-specific binding (PM) Gene-specific binding (MM)

Stacking energy of base-pairs

Fitting the model

ln (signal) Probe index

B e N e N I

ij ij

E E j ij

+ + + + =

*

1 * 1

The baseline of non-specific binding

Non-specific binding energy

B e N e N I

ij ij

E E j ij

+ + + + =

*

1 * 1

• 3
• 2
• 1

1 2 3

Middle 3 bases of PM probe

< ln(PM/MM) > E*(PM)-E*(MM)

A C G T

The effect of mismatch depends on the nearest-neighbors

A A C G T B e N e N I

ij ij

E E j ij

+ + + + =

*

1 * 1

Effect of cross hybridization on model fitting

Latin-square tests with ‘spiked-in’ genes

1 2 3 4 5 6 7 8 9 10 11 12 13 14 1 0.25 0.5 1 2 4 8 16 32 64 128 256 512 1024 2 0.25 0.5 1 2 4 8 16 32 64 128 256 512 1024 3 0.5 1 2 4 8 16 32 64 128 256 512 1024 0.25 4 1 2 4 8 16 32 64 128 256 512 1024 0.25 0.5 5 2 4 8 16 32 64 128 256 512 1024 0.25 0.5 1 6 4 8 16 32 64 128 256 512 1024 0.25 0.5 1 2 7 8 16 32 64 128 256 512 1024 0.25 0.5 1 2 4 8 16 32 64 128 256 512 1024 0.25 0.5 1 2 4 8 9 32 64 128 256 512 1024 0.25 0.5 1 2 4 8 16 10 64 128 256 512 1024 0.25 0.5 1 2 4 8 16 32 11 128 256 512 1024 0.25 0.5 1 2 4 8 16 32 64 12 256 512 1024 0.25 0.5 1 2 4 8 16 32 64 128 13 512 1024 0.25 0.5 1 2 4 8 16 32 64 128 256 14 1024 0.25 0.5 1 2 4 8 16 32 64 128 256 512

Gene Sample

Data source: Affymetrix Inc.

Accuracy of estimated gene expression levels

Reproducibility of estimated gene expression levels

Correlation of expression levels between batches of experiments

Spotting cross hybridizing probes

Cross hyb probes: unknown EST Cross hyb source: salivary alpha-amylase Cross hyb probes: interleukin-8 receptor type B Cross hyb source: angiotensinogen serine (or cysteine) proteinase inhibitor

The missing 12th gene?

Is the missing gene an alternative splicing variant?

Gene name: rTS beta protein

Conclusions

• Probe signals can be decomposed into two

modes: gene specific binding and non-specific binding.

• Sequence dependence of probe signals can be

determined by a thermodynamic model.

• For the given data set, the amount of cross

hybridization and its sources on the probes can be determined.

Acknowledgements

Ken D Aldape Ken Hess Keith A. Baggerly James Mitchell Norris Clift Lianchun Xiao Kevin R. Coombes

Website for downloading the program

http://bioinformatics.mdanderson.org

Perfect Match

Clustering crosshyb effects

Plot of residues

• 3
• 2
• 1

1 2 3

• 4
• 2

2 4 ln(MMfitted) - ln (MM) ln(PMfitted) - ln (PM)

The misfits happen in the same probe pairs

Basic questions of microarrays

• How can we determine gene expression

levels from probe signals?

• How does probe binding affinity depend on

the probe’s sequence?

• How much binding on a probe is due to non-

specific binding?

• Why sometimes a mismatch probe signal is

stronger than the corresponding perfect match probe signal?

Why a good physical model for microarrays is important

• Recognize erratic probe signals
• Eliminate inefficient probes
• Extract accurate and reliable gene

expression levels

Protocol of a microarray experiment

Oligonucleotide microarray technology

Affymetrix array ~ Affinity matrix

Basic features:

• Probe set -- Multiple

probes for a gene target

• Probe pair -- Perfect

match vs. mismatch

Effects of alternative splicing

Even probes:

2 4 6 8 10 12 14 16

Odd probes:

1 3 5 7 9 11 13 15

1st half probes:

1 2 3 4 5 6 7 8

2nd half probes:

9 10 11 12 13 14 15 16

Fitting the model with sub- divided probe sets

Alternative splicing: DNA: ------------------------ mRNA1: ------------------------ mRNA2: ----------- mRNA3: ------------

• (

Mechanism of non-specific binding

• n the probes
• A. Non-specific binding energy is much higher

than gene specific energy (E* - E = 5 kBT)

• B. Source of non-specific binding is much

higher than source of gene-specific binding.

• C. Non-specific binding is very loose, flexible,

and contains many mismatches.

• D. Non-specific binding depends on stacking

energy, which in turn depends on the probe sequence.

Energetic aspect of probe design

1 2 3 4 5 6 7 8 9

• 5
• 4
• 3
• 2
• 1

1 2 3 4 5

Gene specific binding energy ln (observed signal) Background

Total signal Gene specific signal

Boltzmann Distribution

E1 E2 N1 N2

T k E E

B

e N N

2 1

2 1 − −

=

E

e N N N

∆

+ = + 1 1

2 1 1

Binding affinity:

Binding on microarrays

• Some probes always give

strong signals, even when the targeted gene is absent.

• Some probes always give

weak signals, even when the targeted gene is abundant. Probe’s response to gene expression:

Level of expression Probe Signal