1 of 29

1
Evaluating Genes and Transcripts(Genebuild)
2
Outline

• Ensembl gene set
• EST genes ab initio predictions
• Manual curation (Vega / Havana)
• Ensembl / Havana merged gene set
• CCDS project

3
Biological Evidence
All Ensembl gene predictions are based on
experimental evidence

• UniProt/Swiss-Prot
• A manually curated database and therefore of
  highest accuracy
• NCBI RefSeq
• A partially manually curated database
• UniProt/TrEMBL
• Automatically annotated translations of EMBL
  coding sequence (CDS) features
• EMBL / GenBank / DDBJ
• Primary nucleotide sequence repository

4
The Ensembl Genebuild
Genome assembly

Experimental evidence
Ensembl Genes

Computer programs
5
The Ensembl Genebuild
A new release of Ensembl doesnt contain a new
genebuild for each species! New genebuilds are
only done if there is

• a new genome assembly
• a lot of new supporting evidence

6
Genome Assemblies

• Genome assemblies are not created by
• Ensembl, but provided by other institutes /
• consortia
• NCBI human, mouse
• Rat Genome Sequencing Consortium rat
• Sanger zebrafish
• Broad Institute mammals
• Baylor College cow
• Washington University chicken
• etc. etc.

7
The Ensembl Genebuild

• Targeted build
• Align species-specific proteins to the genome to
  create transcripts
• Similarity build
• Align proteins from closely related species to
  locate additional transcripts
• Add UTRs using mRNA evidence
• Eliminate redundant transcripts and create genes

8
Special cases

• Pseudogenes
• Non-coding RNA genes
• sequences from RFAM and miRBase dbs and Infernal
• hand-checked set
• Ig Segment Genes (Immunoglobulin and T-cell
  receptor segments)
• sequences from IMGT db and Exonerate

9
Classification of Transcripts

• Ensembl Transcripts and Proteins are mapped to
  UniProt/Swiss-Prot, NCBI RefSeq and
  UniProt/TrEMBL entries
• Genes that map to species-specific protein/mRNA
  records are classified as known
• Genes that do not map to species-specific
  protein/mRNA records are classified as novel

10
Names and Descriptions

• Transcript names are inferred from mapped
  transcripts and proteins
• Swiss-Prot gt RefSeq gt TrEMBL ID
• Novel transcripts have only Ensembl identifiers
• Genes are assigned the official gene symbol if
  available
• HGNC (HUGO) symbol for human genes
• Species-specific nomenclature committees (MGI,
  ZFIN etc.)
• Otherwise Swiss-Prot gt RefSeq gt TrEMBL ID
• Gene description is inferred from mapped database
  entries, the source is always given

11
Projections

• Name and description of gene is sometimes
  projected from its human or mouse orthologue
• Done for
• human to most other mammals
• mouse to rat
• human to fish
• Gene classified as known_by_projection

12

• Find the supporting evidence for the transcript
  of the Ensembl GALP (Galanin-like peptide
  precursor) gene of human.
• On how many pieces of evidence was this
  transcript build?
• Why do two pieces of evidence not support the
  first exon of the transcript?
• From what source did Ensembl get the name for the
  gene? And from where did it get the description?

13
More Supporting evidence
ContigView
14
Configuring the Genebuild

• Genebuild configured for each species
• Data availibility
• Targeted build most useful in human, mouse
• Similarity build most useful in C. intestinalis,
  mosquito
• Structural issues
• Zebrafish
• Many duplications
• Genome from different haplotypes
• Mosquito
• Many single-exon genes
• Genes within genes

15
Low Coverage Genomes

• Low coverage genomes (2x) come in lots of
  scaffolds classic genebuild will result in
  many partial and fragmented genes
• Whole Genome Alignment (WGA) to an annotated
  reference genome this method reduces
  fragmentation by piecing together scaffolds into
  gene-scaffolds that contain complete gene(s)

16
Low Coverage Genomes
reference assembly
NNNNNN
gene-scaffold
17
EST Gene Set

• ESTs (Expressed Sequence Tags) are single reads,
  high chance of sequencing mistakes
• EST libraries are regularly contaminated with
  genomic DNA
• Generally 400 bp, so unlikely to cover a whole
  gene
• THEREFORE
• EST gene predictions are less reliable and thus
  kept separate from the core Ensembl Gene Set

18
EST Gene Set
ContigView
EST
EST gene
19
Ab initio Predictions

• Predict translatable transcript structures solely
  on the basis of genome sequence
• No validation with biological expression
  information
• GENSCAN for vertebrate genomes
• SNAP better for invertebrates
• NB Both programs are over-predicting transcript
  structures

20
Ab initio Predictions
ContigView
GENSCAN prediction
21
Automatic vs Manual Annotation

• Automatic Annotation
• Quick
• Use unfinished sequence or shotgun assembly
• Consistent annotation

• Manual Annotation
• Slow
• Need finished sequence
• Flexible, can deal with inconsistencies
• Most rules have exceptions
• Consult publications as well as databases

22
Annotation that Causes Problems for Ensembl

• Multiple variants
• UTRs
• Pseudogenes
• Non-coding genes (ncRNAs)
• Overlapping genes, anti-sense genes
• Gene duplication events

23
Manually Curated Gene Sets

• Whole genome
• FlyBase fruitfly
• WormBase C. elegans
• SGD yeast
• Part of the genome
• Vega human, zebrafish, mouse, dog

24
Vega Genome Browser
http//vega.sanger.ac.uk
25
Vega Transcripts
Vega Havana transcripts annotated by the Havana
team at Sanger Vega External transcripts
annotated by other Vega teams
26
Ensembl / Havana Merge
Full-length protein-coding transcripts annotated
by the Sanger Havana team (part of Vega) are
merged with the human and mouse Ensembl
transcript sets

• Transcripts
• Ensembl/Havana gold
• Ensembl red / black
• Havana blue
• Genes
• Ensembl/Havana gold
• Ensembl red / black
• Havana blue

27

• Find the Ensembl Epc1 (enhancer of polycomb
  homolog 1) gene of mouse.
• How many transcripts has Ensembl annotated for
  this gene? And Havana?
• About how many transcripts does the Ensembl and
  Havana annotation agree?

28
CCDS(Consensus Coding Sequences)

• Collaboration between NCBI, UCSC, Ensembl and
  Havana to produce a set of stable, reliable,
  complete (ATG-gtstop) CDS structures for human and
  mouse
• Long term aim is to get to a single gene set for
  human and mouse
• The genebuild pipeline has been modified to
  retain these blessed CDSs (stored in a database
  for incorporation in the build)

29
Q

A
Q U E S T I O N S A N S W E R S