Automated cisRegulatory Annotation of genomes

1
Automated cis-Regulatory Annotation of genomes

• Saurabh Sinha
• Dept. of Computer Science,
• UIUC.

2
Automated genome annotation

• Routine steps today
• Gene prediction
• Orthology maps
• Gene functions

3
Genes are not the whole story
Genetic regulatory network controlling the
development of the body plan of the sea urchin
embryo Davidson et al., Science,
295(5560)1669-1678.
4
Gene Regulation

• Biological processes, including development, are
  coordinated by spatio-temporal interactions among
  genes
• Some genes (transcription factors) regulate the
  expression of other genes
• GENE REGULATORY NETWORK (GRN)
• GRN is a key substrate of evolution
• morphological diversity arises out of evolution
  tinkering with GRN

5
Annotating cis-Regulatory elements

• Goal is to unravel the GRN for some biological
  process, in some species
• Each edge of GRN is of the form transcription
  factor X regulates gene Y
• Many other possible forms, this is the most
  well-studied
• Molecular implementation of this edge binding
  site for the transcription factor, located near
  the gene
• Sub-goal Find these binding sites in the genome

6
Preview

• Different bioinformatics methods to annotate
  genomic footprints of cis-regulatory interactions
• Different methods differ in terms of
• Input Data e.g., single species or multiple
  species known motifs or unknown motifs etc.
• Precise goal Find actual binding sites Find
  clusters of binding sites Find target genes
  etc.

7
(1) Annotation of transcription factor binding
sites (TFBS)

• Output Predicted binding sites (10 bp long) of
  a given TF
• Input Prior characterization of binding site
  affinity of a transcription factor motifs
• From high throughput techniques (e.g., Chromatin
  Immunoprecipation, Bacterial 1 Hybrid, etc.)
• Input multiple, closely related genomes

8
Examples

• Harbison et al. (Nature 2004) on yeast
• gt 3000 binding sites for over 100 TFs
• Stark et al. (Genome Res. 2007) on Drosophila
• gt 46000 regulatory interactions

Binding site prediction
9
(2) Genome-wide prediction of motifs

• If the binding specificity of a TF is not known
  from experiments, it can be computationally
  predicted
• Output A set of motifs that probably represent
  TF binding affinities
• Input multiple, closely related genomes. Some
  methods attempt to find motifs from single
  species alone.

10
Examples

• Kellis et al. (Nature 2003) on yeast
• 72 motifs identified.
• Xie et al. (Nature 2005) on human
• 174 motifs identified
• Out of how many ? Perhaps 2500, but not all have
  distinct motifs
• Down et al. (PLoS CB, 2006) on Drosophila
• 120 motifs identified (out of 700)

11
(3) Prediction of clusters of TFBS

• In many cases, binding sites do not work alone
• Clusters of binding sites, of the same or
  different TFs, together mediate a particular
  expression pattern
• Such clusters are called cis-regulatory modules
  (CRMs). Typically, CRM prediction is more
  accurate than individual TFBS prediction
• Output Annotation of CRMs (1000 bp long)
  involved in a certain biological process. Expect
  1-3 per gene.
• Input Set of motifs involved in a biological
  process
• Input (Optional) multiple, closely related
  genomes

12
Examples

• Blanchette et al. (Genome Res. 2006)
• human vs. mouse comparison. gt 100000 modules
  predicted.
• motifs taken from large database (TRANSFAC)
• Noyes et al. (Unpublished, 2007)
• multiple Drosophila species comparison
• motifs taken from high throughput assay (B1H)
• Based on our earlier work (ISMB 2003, ISMB 2006)
• Smith et al. (Mol. Sys. Biol. 2007)
• multiple mammalian species
• motifs taken from large databases
• tissue-specific gene expression data

13
(4) Prediction of TF-gene interactions

• Individual TFBS prediction may suffer from high
  false positive rate
• May be more practical to only predict TF X
  targets Gene Y
• Output Pairs of (TF, Gene) regulatory
  interactions
• Input Motifs of known TFs
• Input (Optional) multiple, closely related
  genomes

14
Examples

• Sinha et al. (PNAS 2006)
• Honeybee genome
• Motifs from another insect (fruitfly)
• Sinha et al. (Genome Res. 2007)
• Human genome
• Motifs from TRANSFAC
• Penacchhio et al. (Genome Res. 2007)
• Human genome
• Multiple mammalian species analyzed

15
(5) Annotation of miRNA targets

• Transcription factor binding to DNA is not the
  only mode of gene regulation
• MicroRNAs binding to 3 UTRs of mRNA is another
  major mode of gene regulation
• Output (miRNA, Gene) interactions
• Input miRNA sequence
• Input multiple, closely related genomes

16
Examples

• Krek et al. (Nature Genetics 2005.)
• human genome
• Lewis et al. (Cell 2005)
• human genome
• Grun et al. (PLoS CB, 2005)
• Drosophila genome

17
(6) CRM Prediction without known motifs

• More on the research side
• Output CRMs in a genome
• Input Set of functionally related CRMs in the
  same genome
• Current estimates (unpublished data)
• 50 specificity in Drosophila
• Joint work with Gene Robinson, preliminary work
  in ISMB 2007.

18
(7) CRM Prediction across evolutionary gaps

• Output CRMs in a new genome
• Input Orthologous CRMs in a different genome
• Assumption Alignment not available as a guide
• E.g., Knowledge of CRMs in fruitfly, export
  this knowledge to the honeybee or wasp genome
• Ongoing work joint work with Gene Robinson.

19
Summary

• Already achievable Genome-wide prediction of
• binding sites
• binding site affinities (motifs)
• Cis-regulatory modules (CRMs)
• TF-gene interactions
• miRNA targets
• Issue Accuracy not very high but still very
  useful !
• One direction for the future of automatic
  annotation
• Exporting annotation from one species to
  another, without the aid of alignments