1. EXTENDING GENE ANNOTATION WITH GENE EXPRESSION
Chris Stoeckert, Penn Center for Bioinformatics

2. Extending Gene Annotation with Gene Expression Patterns
Genomic Sequence
ESTs
Clones, tags, cDNA, oligos
BLAST
arrays, SAGE,
differential display
Genes
What is it?
What does it do? Sequence similarity
How does it do it?
Where/When does it do it? Gene expression
What happens when
it doesn’t do it right? } }

3. Data Flow for Annotation by Gene Expression
ESTs
Clones, tags, cDNA, oligos
arrays, SAGE,
differential display
BLAST
Database of
Transcribed Sequences
RNA Abundance Database
Pattern Generator and Analysis
What is the gene? Where is it expressed?
Look for co-regulation. Look for networks.
Sequence Analysis

4. RAD Schema Enhances Annotation by Providing Better Understanding of Data
Data verification
consistency within an experiment
Reproducibility
consistency between experiments
Comparison between platforms
can track data for same mRNA
Integration with other resources
DOTS (gene info), Anatomy (sample info)

5. RAD Schema: Three sets of Tables Provide Flexibility
Experiments
Data
Platforms
Anat_rel
Others...
SpotResult
ExperimentCondition
StanfordSpotFamily
SpotFamilyResult
Experiment
GenomeSystems
SpotFamily
ExperimentyResult
SynteniSpotFamily

6. RAD:DOTS Interface RNA Abundance Database
Database of Transcribed Sequences
Anatomy
Cellular role
Clones/PCR
filter/array
Experiment
probe
other parameters
hyb/wash conditions
Results
signal/ %
background adjustments
SWISS-PROT
neighbors
protein
EST
clusters
RNA
Genomic
sequence
DNA
Regulatory elements

7. RAD Web Interface
choose through DOTS

8. RAD Web Interface
example of retrieving by cell role or library

9. Expression Pattern Algorithm
Input:
files with identifier and value (e.g., IMAGE clone ID,
percent of total signal) for each experiment.
tolerated variance between replicates.
Dynamic range (e.g., top 15% of signals are meaningful)
dependencies between experiments
Output:
expression patterns based on ratios between experiments
(list of bins)
variance between replicates
genes above background
distribution of ratios for use in statistical analysis

10. Expression Pattern Algorithm
1. Determine minimum useful value for each group of replicate
experiments based on specified dynamic range. Raise all values
below the minimum useful value to equal this value.
2. Determine the ratio (cutratio) that contains specified percentage
of ratios between replicates. Default = 2.
3. Take ratios between average values for each group of replicates.
Use the median value if group or use specified reference group in
denominator.
4. Bin the ratios. Use powers of the cutratio and the range of ratios
to generate the cut-off ratios for each bin. Generate a second set of
bins offset from the first to capture ratios which straddle the first set
of bins.

11. Statistical Analysis of Patterns
Models the ratios as a multinomial experiment
Null hypothesis is independence between genes, given
a model describing dependencies between experiments.
(independent, reference, conditional, 1st order Markov)
The number of expected patterns are calculated based on the
distribution of ratios and the experiment model. The likelihood
of the observed number of patterns can be calculated (or simulated)
using the number of expected patterns.
Simulators throw weighted die to generate patterns for each gene to
obtain the number of genes in a specified pattern. A score is
generated using this expected number and the actual numbers of
genes in a pattern.

12. Sample output of pattern program

13. Extended Annotation from Expression Patterns
RAD:
Pattern:
DOTS:
GenBank:
(GAIA)
TESS:
Extract EST data in experiment groups.
Find ESTs with same pattern.
Map ESTs to transcribed sequences.
Get promoters.
(genomic sequence upstream of transcribed sequences)
Look for shared transciption factor binding sites.

14. Extended Annotation from Expression Patterns
Comparison of array data stored in RAD from:
HEL, HEL+hemin, CD34, erythroblasts
Significant cluster of clones down-regulated in erythroblasts:
10/24 clones coded for ribosomal proteins according to DOTS.
Obtained promoters for 4 of these ribosomal proteins from GenBank.
Used TESS to find shared transcription factor binding sites.
All contain sites for PU.1 which is antagonistic to red cell growth.
Result is subset of ribosomal proteins which are co-regulated
and potential mechanism for co-regulation.

15. Summary genomic sequence ESTs candidate genes DOTS: integrated info
RAD: facilitate analysis
Pattern: co-regulation
candidate function
candidate genes with related role

16. Acknowledgements