Detection and Compensation of CrossHybridization in DNA Microarray Data

1
Detection and Compensation of Cross-Hybridization
in DNA Microarray Data

• Joint work with
• Quaid Morris(1), Tim Hughes(2)
• and Brendan Frey(1)

Jim Huang (1),

• Probabilistic and Statistical Inference Group,
  University of Toronto
• (2) Banting Best Department of Medical
  Research, University of Toronto

2
Description and Applications of DNA Microarrays

• Microarrays consist of a 2-D array of probes,
  each with a short DNA sequence attached. These
  sequences are called oligonucleotide sequences.
• The output of each probe is approximately
  proportional to the amount of DNA that binds to
  the probe from a given tissue the data for each
  probe is an N-dimensional expression profile
  vector, where N is the number of tissues used on
  the array.
• DNA microarrays can be used to measure the level
  of gene expression across these N tissues.

3
Hybridization and cross-hybridization
DNA from tissue sample

• The process of 2 complementary DNA strands
  binding is called hybridization
• Ideally, an oligonucleotide probe will only bind
  to the DNA sequence for which it was designed and
  to which it is complementary
• However, many DNA sequences are similar to one
  another and can bind to other probes on the
  array
• This phenomenon is called cross-hybridization

ATCTAGAAT
TCGAT CCTA
AGCTAGGAT
TCGAT CCTA
Hybridization
Cross-hybridization
Oligonucleotide Probe
4
The trouble with cross-hybridization

• With cross-hybridization, each probe will signal
  the presence of multiple sequences other than
  that it was designed for
• This skews the observed data from the expected
  data.

Observed expression profile vector
(cross-hybridized)
Expected expression profile vector (no
hybridization)
5
Detecting cross-hybridization (1)

• To test for whether cross-hybridization is
  impacting the gene expression data, we perform a
  BLAST sequence match on all oligonucleotide probe
  sequences used on the microarray
• Many probes will be matched with sequences for
  which it wasnt specifically designed.

6
Detecting cross-hybridization (2)

• We compute the Pearson correlation coefficient ?
  between matched probe sequence expression
  profiles and between the profiles of
  randomly-paired probes
• Approximately 33 of the BLAST-matched probes
  have ? gt 0.95, whereas only 2 of
  randomly-matched probes have ? gt0.95
• This difference in the 2 distributions indicates
  that cross-hybridization indeed has a significant
  impact on the observed gene expression data.

7
Compensating for cross-hybridization

• We model the observed, cross-hybridized
  expression profile vector x as a matrix product
  of a hybridization matrix ? and an unobserved
  expression profile vector z in which there is no
  cross-hybridization.
• The elements ?ij of the ? matrix are set as
  parameterized functions of the Gibbs free energy
  ?Gij between probes i and j.
• To compensate for cross-hybridization, we use a
  generalized Expectation-Maximization algorithm in
  which we solve for z and ? iteratively.