Mauro Delorenzi

1
The ISREC / NCCR Bioinformatics Core Facility
(BCF)DAFL OPEN HOUSE December 10, 2003
Mauro Delorenzi
2
Overview

• What is the BCF ?, (towards a Lemanic
  Centre for Data Analysis ?)
• About spotted cDNA Microarrays
• Detection of differential expression by different
  Microarray Platforms (spotted cDNA, Agilent and
  Affymyetrix) and MPSS (case study,
  preliminary analysis)
• Tools Development for Analysis

3
BCF What is it ?

• ISREC-based, supported by the NCCR for molecular
  oncology, member group of the SIB
• Crated by the NCCR molecular oncology to assist
  its DAF (which is now absorbed into the DAFL) and
  its microarray users in their biomedical research
• A group devoted to the bioinformatics and
  statistical aspects of gene expression research,
  in particular to the analysis of data generated
  with the microarray technology

4
BCF and DAFL

• Shared operations with the DAFL
• array testing, annotation
• normalization and quality control services
• collaboration in future developments of the
  technology (platform evaluation, optimisation) ?
• collaboration in statistical data analysis of
  research projects with the DAFL bioinformatics
  team and with biostatistics dept at EPFL

DAFL as "nucleation center" ?
5
BCF Our Main Components

• 1. Technical Support
• advice in experimental design and data analysis
• production, control, development of spotted
  arrays
• processing of microarray data, data quality
• 2. Collaborations
• - statistical data analysis of research
  projects
• 3. Education
• practical training through classes / workshops
• 4. Research Development
• development / testing tools methods

• gt we get requests for assistance and
  collaborations
• gt we get requests of young postdocs that ask for
  jobs
• Started an "Open House Concept"
• post someone in the BCF
• pay for service / collaboration

6
Spotted cDNA arrays

• Data processing service through web interface
• set up for NCCR, now shared with DAFL
• reachable through links from our webpage
• http//www.isrec.isb-sib.ch/BCF/index.html
• MAIN AIMS
• - Proper Normalization
• Diagnostic plots (quality control, error analysis
  )
• Updated Information on the spotted clones / genes
  

7
Spotted cDNA arrays
Human 10k Array 8x4 subarrays
8
Diagnostic plot spatial-effect-visualization
diagnosis / feedback available for
inhomogeneous hybridization weak
signals saturation spot detection
problems overall performance can be useful also
for Agilent slides
problem alleviated but not eliminated by stronger
correction
9
Experience with spotted cDNA arrays

• after different improvements we reached the stage
  where peak performance is excellent
• reproducibility is high ...
• ... when all steps worked well,
• ... which is not always the case
• robustness is relatively low,
• genomic coverage of ours arrays is relatively
  poor
• quality control is important (web interface
  provides kind of a minimal standard and an
  immediatae feedback)
• the BCF/DAF/DAFL has set up a system of controls
• failures can happen in the production and in the
  RNA isolations and hybridisations

10
Study Design

• Platforms
• Affymetrix GeneChips short oligo arrays
• Agilent long oligo arrays
• in-house spotted cDNA arrays
• MPSS (massively parallel signature sequencing),
• collaboration with the Ludwig Institute
• Basic Design
• replicate measurements for two mRNAs (human
  placenta and testis)
• dye swap for two-color systems (Agilent, cDNA)
• 2 to 3 millions tags sequenced for MPSS

11
Study Design II

• Experimental Method
• as recommended by "specialists"
• Affymetrix Biozentrum Basel
• Agilent Institut Goustav Roussy (Paris)
• Spotted cDNA arrays Otto Hagenbuechle's
• (DAF, now DAFL) crew
• MPSS Lynx (California), Victor Jongeneel's crew
• Data Handling
• as recommended by "specialists" see above
• but RMA quantile normalization for Affymetrix

12
Platform Comparison Study

• Purpose
• to assess accuracy and reproducibility of
  different gene expression platforms
• to compare features of different measurement
  types
• to understand the system (important for
  normalization and downstream analysis)

13
Comparison principle

• crossplatform matching is done through the Tromer
  database of transcripts, conserving only genes we
  classify as reliably mapped between platforms
• we have not yet looked at probe(set)s that could
  not be well mapped to known transcripts
• "peak technical performance" this is a case
  study, not a systematic study
• comparison based on M (log ratio) and A (log
  intensity) values
• accuracy cannot be assessed, as true M values are
  not known

14
DiffVAvrg plots testing reproducibilty
ydiff in M xavrg Int (A)
Affy U133A
Agilent h1A
DAF(L) h10k
15
correlations
first quartile (25 less frequent RNAs)
fourth quartile (25 most frequent RNAs)
16
Agreement top up 200 (placenta)
M range Affy 1.66 - 7.94 Agil 1.48 -
6.17 NCCR 1.83 - 7.12
17
Agreement top down 200 (testis)
M range Affy -8.27 - -1.65 Agil -6.07 -
-1.47 NCCR -6.18 - -1.79
18
Preliminary Conclusions I

• The three microarray platforms compared performed
  very similarly in terms of which genes are
  detected as differentially expressed ...
• ... and also similarly in terms of the
  distributions of M values and the deviation
  between replicated measurements
• ... so similarly that it is hard to find real
  intrinsic differences between the three platforms
  ....
• ... at least in this case study, which has strong
  effects at all signal intensities (at strongest
  at average-high intensities)
• Sensitivity (LOD), detection of DE at low signal
  intensity is likely to be similar too
• One would need to know real M values to see who
  is right when they diagree, use MPSS?, plan qPCR

19
Preliminary Conclusions II

• The Affy-RMA M values are better
  variance-stabilized, but reproducibility is good
  for all platforms except for weak signals in
  Agilent (likely due to bgr sbtr)
• The Affy-RMA M values are more strongly
  "compressed" towards zero at low intensity, if
  this is a good strategy cannot be said on the
  basis of our results, it reduces false positive
  calls but might make DE at low intensity
  undetectable (but is it detectable at all?)

20
Preliminary Conclusions III

• Microarray vs MPSS
• M values, quantitative comparison
• the disagreement is considerable ...
• ... so large that it is hard to reconcile the
  values (by using confidence intervals)
• M values, qualitative comparison
• there is a good degree of agreement
• approximately the same to all three microarray
  platforms

21
Interpretation tool Isrec Ontologizer Io
Thierry Sengstag Coll. with Pascal Anderle
Selection of hierarchical level Classification
of probe sets Classification of
UniGenes Classification of RefSeqs Flagging of
ambiguous results
22
Analysis Tool Can we detect joint action of
two genes?
Postdoc Asa Wirapati

• Look at each possible pair, try to separate
  tissue types by a straight line. Two aims
• good discrimination (with a small number of
  genes)
• identify genes that are discriminative in
  combination, but not taken singularly
• Several dozens pairs are statistically
  significant (w.r.t. randomly permuted labels),
• they could represent cases of bona-fide
  biologically significant "joint action" (Godard
  2003)

23
END
questions