Corticosteroids are drugs used to treat pathologies requiring antiinflammatory intervention' A major

1
Selecting Maximally Informative Genes to Enable
Temporal Expression Profiling Analysis E. Yang,
C.M.Roth and I.P. Androulakis Rutgers University

• Analysis
• Six essential motifs of gene expression
  containing 400 genes were identified. The
  selection process exhibits great precision as
  indicated by the fact that, essentially, the same
  subset of liver specific genes is identified
  whether 8,799 or 217 probes are used
• Projection of informative genes temporal
  trajectories on two principal vectors captures
  qualitatively the effect of a bolus drug
  injection
• In order to assess the extent to which the motifs
  contain co-regulated genes as well, the
  probabilities of observing common TF regulating
  the corresponding genes were evaluated
• Preliminary indications seem promising despite
  the limited number of available TFs
  (http//genometrafac.cchm.org)

• Temporal Analysis of Expression
• Relationships among co-expressed genes provide
  evidence for their involvement in biological
  functions
• As gene expression experiments are being
  augmented by considering the entirety of the
  regulatory networks affecting the transcription
  process two critical questions have to be
  addressed in a systematic and comprehensive
  manner
• Establish the complete set of patterns of gene
  expression and their non-intuitive relations
• Establish the link between genes exhibiting
  correlation in terms of expression patterns and
  the corresponding mechanisms affecting
  transcription
• We propose an unbiased estimator of temporal
  expression characteristics and a framework for
  assessing critical regulatory links among genes
  sharing common expression profiles

Corticosteroids are drugs used to treat
pathologies requiring anti-inflammatory
intervention. A major initiating mechanism
involves ligand-glucocorticoid receptor binding
followed by receptor mediated transcriptional
modulation of gene expression.
Expression Motifs
16 hr
17.5 hr
15 hr
1
3
2
1 2 3
6.25 hr
6.25 hr
6.25 hr
4
6
5
3 hr
2.25 hr
5.25 hr
Almon et al. (J. Pharm. Exp. Therap., 2003)
15 hr
15 hr
15 hr
It is unlikely that to date all glucocorticoid
regulated genes have been identified in any
target tissue
Time to recovery Time to min/max
Expression profiling can be used to distinguish
temporal patterns of changes in gene expression
in response to a drug in vivo. These patterns can
be used to identify groups of genes regulated by
common mechanisms. The rationalization of the
assignment of genes to expression motifs in their
regulatory and functional context remains an open
issue.
4 5 6

• Characterizing the Transcriptional State of the
  System
• Clustering algorithms identify groups of profiles
  whose aggregation (class assignment) minimizes a
  distance metric
• We consider the expression motifs as a new
  basis to expand the evolution of the
  Transcriptional State (TS) of the system
• The TS is defined as the temporal deviation of
  the c.d.f. of the expression values relative to
  the control state quantified via a
  Kolmogorov-Smirnov metric

• Transcriptional Profiling of Cotricosteroid
  Effects on Rat Liver
• Almon et al. (Funt. Integr. Genomics, 2003)
  undertook a major animal study to assess the
  impact of a single 50 mg/kg dose of
  methylprednisolone sodium succinate on ADX rats
• 2-4 rats we sacrificed at the following times
  after MPL administration 0.25, 0.5, 0.75, 1, 2,
  4, 5.5, 6, 7, 8, 12, 18, 30 48, and 72h. Four
  animals were designated as controls, i.e., zero
  time samples
• Isolated RNA from the livers of the individual
  animals were used to prepare targets which were
  hybridized to 47 individuals RG_U34A Affymetirx
  Genechips containing 8,799 rat specific probes
• In order to facilitate the 217 corticosteroid
  regulated probe sets were selected for further
  study
• An ad hoc procedure was used to cluster the 217
  expression profiles to 6 classes and subsequently
  Almon et al. (J. Pharm. Exp. Ther., 2003)
  pharmacogenomic models were proposed to quantify
  the mechanism of gene expression of each class

Piece-wise Aggregate Approximation PAA
The minimum set of informative motifs is
identified based on its ability to best capture
the temporal deviation of the TS from the control
state

• Conclusions and Future Work
• We demonstrated a novel unsupervised approach for
  extracting critical motifs of gene expression in
  microarray experiments
• Preliminary results indicate that dominant motifs
  of co-expressed genes are efficiently identified
  and their common regulatory structure explored
• Future work alternative PPH, combinatorial
  exploration of motif combinations, extension of
  TF database, validation nd quantification of
  regulatory interactions in terms of available
  gene expression measurements

• Selection of Informative Motifs
• Simple iterative optimization procedure
  determines the minimum number of motifs that
  maximize the deviation from the control state

Proximity Preserving Hashing
Points in an n-dimensional space are mapped to
adjacent points in a lower (1-d) space Having a
symbolic (discrete) representation enables the
assignment of distinct identifiers to each
profile Sequences of symbols hashing to similar
buckets (motifs) originate from similar
sequences Represent each transcriptional profile
by the corresponding hash value assigned to the
surrogate time series
Liver specific genes
All available genes
The authors would like to acknowledge Prof. R.
Almon (SUNY Buffalo) for insightful guidance and
J. Vitolo (CBE, Rutgers) for stimulating
discussions in the early stages of this work
Control state Informative subset
(Almon et al. (J. Pharm. Exp. Therap., 2003)