Transcript LecCh6Annotationx

Genome Annotation
• Now that you’ve assembled your genome,
what is next?
• GENOME ANNOTATION
• What is that?
• Why is it important?
• How do you do it?
Challenges to Genome
Annotation?
• Finding genes involves computational
methods as well as experimental validation
• Computational methods are often inadequate,
and often generate erroneous ‘gene’ (false
positive) sequences which:
– Are missing exons
– Have incorrect exons
– Over predict genes
– Where the 5’ and 3’ UTR are missing
What kinds of things are we looking to
annotate?
•
•
•
•
•
•
CDS - coding sequences
mRNA
Alternative RNA
Promoter and Poly-A Signal
Pseudogenes
ncRNA
Pseudogenes
• Could be as high as 20-30% of all Genomic sequence
predictions could be pseudogene
• Non-functional copy of a gene
– Processed pseudogene
•
•
•
•
Retro-transposon derived
No 5’ promoters
No introns
Often includes polyA tail
– Non-processed pseudogene
• Gene duplication derived
– Both include events that make the gene non-funtional
• Frameshift
• Stop codons
• We assume pseudogenes have no function, but we really
don’t know!
Noncoding RNA (ncRNA)
• ncRNA represent 98% of all transcripts in a mammalian cell
• ncRNA have not been taken into account in gene counts
• cDNA
• ORF computational prediction
• Comparative genomics looking at ORF
• ncRNA can be:
– Structural
– Catalytic
– Regulatory
• tRNA – transfer RNA: involved in translation
• rRNA – ribosomal RNA: structural component of ribosome, where
translation takes place
• snoRNA – small nucleolar RNA: functional/catalytic in RNA
maturation
• Antisense RNA: gene regulation/silencing?
Covariance model searches are
extremely compute intensive. A small
model (like tRNA) can search a sequence
database at a rate of around 300
bases/sec. The compute time scales
roughly to the 4th power of the length
of the RNA, so larger models quickly
become infeasible without significant
compute resources.
BLAST
• Seeks high-scoring segment pairs (HSP)
– pair of sequences that can be aligned without gaps
– when aligned, have maximal aggregate score
(score cannot be improved by extension or trimming)
– score must be above score threshold S
• Public Search engines
– WWW search form
http://www.ncbi.nlm.nih.gov/BLAST
– Unix command line
blastall -p progname -d db -i query > outfile
Which Matrix?
• Triple-PAM strategy (Altschul, 1991)
– PAM 40
Short alignments, highly similar
• tblastn against ESTs
– PAM 120
– PAM 250
Longer, weaker local alignments
• Looking in the twilight zone
• BLOSUM (Henikoff, 1993)
– BLOSUM 90
Short alignments, highly similar
– BLOSUM 62
Most effective in detecting known
members of a protein family
• Standard on NCBI server – works in most cases
– BLOSUM 30
Longer, weaker local alignments
Protein coding genes in prokaryotes,
and simple eukaryotes
• Use ORF finder
http://www.ncbi.nlm.nih.gov/gorf/orfig.cgi
• Simple ATG/Stop
• Simple link to FASTA formatted files and
BLAST.
• Problems:
– In frame Methionine
– Small protein
• Solution: comparative genomics
Figure 11 from: Methods in comparative genomics: genome correspondence, gene identification and regulatory motif
discovery. Kellis M, Patterson N, Birren B, Berger B, Lander ES. J Comput Biol. 2004;11(2-3):319-55.
Saccharomyces
Saccharomyces
Saccharomyces
Saccharomyces
cerevisiae.
paradoxus,
mikatae,
bayanus
Ab initio gene identification
• Goals
– Identify coding exons
– Seek gene structure information
– Get a protein sequence for further analysis
• Relevance
– Characterization of anonymous DNA genomic
sequences
– Works on all DNA sequences
Gene-Finding Strategies
Genomic Sequence
Content-Based
Site-Based
Comparative
Bulk properties of
sequence:
• Open reading frames
• Codon usage
• Repeat periodicity
• Compositional
complexity
Absolute properties of
sequence:
• Consensus sequences
• Donor and acceptor
splice sites
• Transcription factor
binding sites
• Polyadenylation
signals
• “Right” ATG start
• Stop codons
out-of-context
Inferences based
on sequence homology:
• Protein sequence
with similarity to
translated product
of query
• Modular structure of
proteins usually
precludes finding
complete gene
Lecture 4.2
14
Gene-Finding Methods
Genomic Sequence
Rule-Based
Cutoff method:
• Criteria applied sequentially
to identify possible exons
• Rank or eliminate candidates
from consideration based on
pre-determined cutoff at
each step
Neural Network
Composite method:
• Criteria applied in parallel
• Training sets used to optimize
performance
• Weight scores in order of
importance
Evaluation Statistics
TP
FP
TN
FN
TP
Actual
Predicted
Sensitivity
Fraction of actual coding regions that are correctly
predicted as coding
Specificity
Fraction of the prediction that is actually correct
Correlation Combined measure of sensitivity and specificity,
Coefficient
ranging from –1 (always wrong)
to +1 (always right)
FN
TN
The Process
• Compute the prediction
• Confirm with biological sequences (also with computational
tools)
• Integrate all of this
• Annotate genome (often via a GUI: Graphical User Interface)
• Validate
• Re-annotate/Update
• Check it twice
• Submit to GenBank
Lots of Software:
•
•
•
•
•
•
•
•
•
•
EnsEMBL (EBI)
Sequin (NCBI)
PseudoCAP (SFU)
GMOD (CSHL)
Pegasys (UBiC)
Apollo (EBI/Berkeley)
GeneMark (Georgia Institute of Tech)
GeneScan (MIT)
GenomeThreader (University of Hamberg)
HMMgene (Technical University of Denmark)
GenBank Features
-10_signal
-35_signal
3'clip
3'UTR
5'clip
5'UTR
attenuator
CAAT_signal
CDS
conflict
C_region
D-loop
D_segment
enhancer
exon
GC_signal
gene
iDNA
intron
J_segment
LTR
mat_peptide
misc_binding
misc_difference
misc_feature
misc_recomb
misc_RNA
misc_signal
misc_structure
modified_base
mRNA
N_region
old_sequence
polyA_signal
polyA_site
precursor_RNA
primer_bind
prim_transcript
promoter
protein_bind
RBS
repeat_region
repeat_unit
rep_origin
rRNA
satellite
scRNA
sig_peptide
snoRNA
snRNA
S_region
stem_loop
STS
TATA_signal
terminator
transit_peptide
tRNA
unsure
variation
V_region
V_segment
GenBank Features: the important ones
-10_signal
-35_signal
3'clip
3'UTR
5'clip
5'UTR
attenuator
CAAT_signal
CDS
conflict
C_region
D-loop
D_segment
enhancer
exon
GC_signal
gene
iDNA
intron
J_segment
LTR
mat_peptide
misc_binding
misc_difference
misc_feature
misc_recomb
misc_RNA
misc_signal
misc_structure
modified_base
mRNA
N_region
old_sequence
polyA_signal
polyA_site
precursor_RNA
primer_bind
prim_transcript
promoter
protein_bind
RBS
repeat_region
repeat_unit
rep_origin
rRNA
satellite
scRNA
sig_peptide
snoRNA
snRNA
S_region
stem_loop
STS
TATA_signal
terminator
transit_peptide
tRNA
unsure
variation
V_region
V_segment
Gene Prediction Caveats
• Predictions are of protein coding regions
– Do not detect non-coding areas (5’ and 3’ UTR)
– Non-coding RNA genes are missed
• Predictions are for “typical” genes
–
–
–
–
–
Must predict a beginning and an end
Partial or multiple genes are often missed
Training sets may be biased
Methods are sensitive to G+C content
Weighting of factors may be inordinately biased
Genome annotation problems:
•
•
•
•
•
•
•
•
•
•
•
Assembling the genome
Analysis & interpretation
Lack of consistency from gene to gene
Lack of consistency from person to person
Lack of controlled vocabulary
Parts we don’t know
Bacteria vs mammals
Graphical user interface
Gene expression/molecular interactions
Dimensions
Updates and maintenance
The ideal annotation of “MyGene”
All clones
All SNPs
Promoter(s)
MyGene
All mRNAs
All proteins
All structures
• All protein modifications
• Ontologies
• Interactions (complexes,
pathways, networks)
•Expression (where and
when, and how much)
•Evolutionary relationships
Some Concluding remarks
• Trust but verify
• Beware of gene prediction tools!
• Always use more than one gene prediction
tool and more than one genome when
possible.
• Active area of bioinformatics research, so be
mindful of the new literature in this .