1. Genomics and Personalized Care in Health Systems Lecture 7 Gene Finding (Part 2) Ab initio and Evidence-Based Gene Finding
Leming Zhou, PhD
School of Health and Rehabilitation Sciences
Department of Health Information Management

2. Annotation Goals of this Class
Create an evidence based set of gene annotations for the dot chromosome and a similarly sized “control” region of a long chromosome arm.
This gene annotation set includes all putative isoforms, coding sequence only.

3. Ab initio Gene Prediction
Gene prediction using only the genomic DNA sequence
Search for “signals” of protein coding regions
Typically uses a probabilistic model
Hidden Markov Model (HMM)
“Putative models”- external evidence needed to verify predictions (e.g. mRNA, ESTs)
We will use multiple gene finders (e.g. Genscan, Nscan, SNAP) for our Drosophila annotations

4. Performance of Gene Finders
Most gene finders can predict prokaryotic genes accurately (>98%)
However, gene-finders do a poor job of predicting genes in eukaryotes
Not as much is known about the general properties of eukaryotic genes
Splice site recognition, different isoforms

5. Common Problems Common problems with gene finders Other challenges
Fusing neighboring genes
Splitting a single gene
Missing exons or entire genes
Over predict exons or genes
Other challenges
Nested genes
Noncanonical splice sites
Pseudogenes
Different isoforms of same gene

6. How to Improve Predictions?
Improve algorithms
Identify conserved sequences from multiple sequence alignments
Consensus model based on results from multiple gene finders
Computational predictions that incorporate biological evidence (e.g. ESTs, cDNAs)
Manual annotation
Collect all the available evidence from multiple biological and computational sources to create the best gene models

7. Generate Consensus Gene Model
Each gene prediction program has different strengths and weaknesses.
Create consensus gene set by combining results from multiple gene finders to generate the best reference set
For the 12 Drosophila species that have been sequenced, the reference gene set are available from FlyBase

8. Manual Annotation Evidence for genes: Goals for this class
EST or other expression data
Conservation to known genes
Computational predictions
Goals for this class
Use web resources to collect evidence
Practice collecting and analyzing data

9. Web Databases and Tools
For this class we will use the following sites:
UCSC Genome Browser for Drosophila:
FlyBase:
NCBI:
GEP Gene Finder:
GEP Model Checker:

10. Phylogenetic Tree for Drosophila

11. UCSC Genome Browser
GEP version, parts of genomes, GEP data, used for annotation of Drosophila species
Strengths
Genome Browser: graphical views of genomic regions
BLAT: BLAST-Like Alignment Tool
Table Browser: access underlying data used to generate the graphical genome browser
Weaknesses
Constraints on ability to display data
Uses
View and organize available computational results
Follow along to view a section of drosophila genome

12. Adjusting Display Options
Adjust following tracks to “pack”
Under “Gene and Gene Prediction Tracks”
FlyBase Genes, RefSeq Genes, N-SCAN, CONTRAST
Adjust following tracks to “dense”
Under “mRNA and EST Tracks”
D. melanogaster mRNAs, Spliced ESTs
Under “Comparative Genomics”
Conservation
Adjust following tracks to “full”
Under “Mapping and Sequencing Tracks”
Base Position
Under “Variation and Repeats”
RepeatMasker

13. FlyBase http://www.flybase.org Strengths Weaknesses Uses
Lots of ancillary data for each gene
Curation of literature for each gene
Weaknesses
Difficult to find data if you don’t already know where to look
Uses
Species-specific BLAST searches
Genome browsers and data sets for all the sequenced Drosophila species

14. Basic Strategy for Annotation
Use Ab Initio prediction to focus attention on genomic features (areas) of interest
Add as much other evidence as you can to refine the gene model and support your conclusion
What other evidence is there?
Basic gene structure
Motif information
BLAST homologies: nr, protein, est
Other species or other proteins

15. Eukaryotic Genome Annotation
BLAST homology: nr, protein, EST
Homology to known proteins argues against false positive
Consider length, percent identity when examining alignments
Without good EST evidence you can never be sure; make your best guess and be able to defend it
Other species or other proteins
For any similarity hit, look for even better hits elsewhere in the genome
If you are convinced you have a gene and it is a member of a multi-gene family, be sure to pick the right ortholog

16. Drosophila Sequence Annotation

17. GEP projects http://gep.wustl.edu D. erecta
Close to D. melanogaster
easier to annotate
D. grimshaw, D. virilis and D. mojavensis
Further from D. melanogaster
More difficult to annotate
Will use example in D. virilis but basic technique is the same

18. The Drosophila genomes
Average gene size will be smaller than mammals
Very low density of pseudogenes
Almost all genes will have the same basic structure as D. melanogaster orthologs; mapping exon by exon works well for most genes
Genes only rarely move to different chromosomal element

19. Goals
Identify genes
For each gene identify and precisely map (accurate to the base pair) all exons
Do this for ALL isoforms

20. Evidence Based Annotation
Human curated analysis
Much better outcome than standard ab initio gene finders
Goal: Collect, analyze and synthesize all evidence available and come up with most likely gene model
Evidence for Gene Models
Basic Biology
Expression Data
Conservation
Computational

21. Basic Biology Known biological properties
Coding regions start with a methionine
Coding regions end with a stop codon
Genes are only on one strand of DNA
Exons appear in order along the DNA
Introns start with GT (or rarely GC)
Introns end with a AG
CCTAGAGTACCA….CAGATAGCTAGGAG

22. Expression: EST, mRNA, Other
Protein coding genes must be expressed
Positive result very helpful
Negative result less informative
Did not find message because looking in wrong tissue or developmental stage
Transcription rate so low, messages remain undetected
Drosophilids: only 20,000 EST from each species; only helpful in rare cases

23. Conservation Coding sequences evolve slowly
Similarity to known genes suggests new genes
D. melanogaster very well annotated
12 other drosophila genomes now available
This will be your most important source of evidence

24. Computational Evidence
Assumption: there are recognizable signals in the DNA sequence that the cell uses; it should be possible to detect these algorithmically
Many programs designed to detect these signals
These programs do work to a certain extent, the information they provide is better than nothing; high error rates

25. How to Proceed You will need to investigate all features of interest:
Ab initio gene finder results
Watch out for ends - fused or split genes
Regions of high similarity with D. melanogaster proteins and EST’s, identified by BLAST but not overlapping with 1 above
Overlapping genes usually on opposite strand
Be vigilant for partial genes at fosmid ends
Regions with high similarity to known proteins (i.e. BLAST to nr) but not found by 1 or 2 above

26. Practical Example in D. virilis
The following example will give a general outline of how we recommend you proceed, works in many cases
Goal: come up with the best gene model that integrates all the evidence in a manner that maximizes likely outcomes and minimizes contradictions

27. Basic Annotation Procedure
For each feature (exon in this case) of interest:
Identify the likely ortholog in D. mel.
Use D. mel. database to find gene model of ortholog and identify protein seq for each exon
Use BLASTX to locate exons; search one by one, find conservation, note position and frame
Based on locations, frames of conservation, as well as other evidence create gene model; identify the exact base location (start and stop) of each CDS (coding exon) for each isoform
Confirm your model using Gene checker and genome browser.

28. Basic Procedure (graphically)
contig
feature
BLASTX of predicted gene to D. melanogaster proteins suggests this region orthologous to Dm gene with 1 isoform and 5 exons:
BLASTX of each exon to locate region of similarity and which translated frame encodes the similar amino acids: 1 3 3 2 1
Frame
alignments

29. Basic Procedure (graphically)
Zoom in on ends of exons and find first met, matching intron Doner (GT) and Acceptor (AG) sites and final stop codon, making sure to keep the frame intact 1 3
Met GT AG GT
Once these have been identified, write down the exact location of the first base and last base of each exon. Use these numbers to check your gene model and if correct generate and save GFF file
1121
1187
1402
1591
1754
1939
2122
2434
2601
2789

30. Example Annotation Open 4 Tabs gander.wustl.edu Flybase.org
Genome browser
D. virilis - Mar chr10
Flybase.org
Tools  blast
The Gene Record Finder
Info from most recent D. mel genome
The search term is case sensitive (Fts is different from fts)
Blast page BLASTX  click the checkbox:

31. Example Annotation from Drosophila virilis
Settings are: Insect; D. virilis; Mar. 2005; chr10 (chr10 is a fosmid from 2005)
Click submit

32. Example Annotation Seven predicted Genscan genes
Each one would be investigated

33. Investigate 10.4
All putative genes will need to be analyzed; we will focus on 10.4 in this example
To zoom in on this gene enter:
chr10: in position box
Then click jump button

34. Step 1: Find Ortholog
If this is a real gene it will probably have at least some homology to a D. melanogaster protein
Step one: do a BLAST search with the predicted protein sequence of 10.4 to all proteins in D. melanogaster
In Genome Browser (
Click on one of the exons in gene 10.4
On the Genscan report page click on Predicted Protein
Select and copy the sequence
On the flybase blast page (
Do a blastp search of the predicted sequence to the D. melanogaster “Annotated Proteins” (Database)

35. Step 1: Find Ortholog
The flybase blastp results show a significant hit to the “A”, “B” and “C” isoforms of the gene “mav”
Note the large step in E value from Mav isoforms to next best hit gbb; good evidence for orthology

36. Step 1: Results of Ortholog Search
The alignment looks right for virilis vs. melanoaster- regions of high similarity interspersed with regions of little or no similarity
Same chromosome supports orthology
We have a probable ortholog: “mav”

37. Step 2: Gene Structure Model
We need information on “mav”; What is mav? What does its gene look like?
If this is the ortholog we will also need the protein sequence of each exon
Use the Gene Record Finder
Available off the “Projects” menu at
Projects Annotation Resources  Gene Record Finder

38. Step 2: Gene Structure Model
Search for “mav” (search box)
Only one mav so select and hit the button:

39. Step 2: Gene Structure Model
Sequence of exons in the mav gene in D. mel

40. Step 2: Gene Structure Model
We now have a gene model (three isoforms each with 2 CDS).
In this case, we will annotate isoform A only.
For your project report you would annotate ALL isoforms for all genes you identify in your fosmid/contig.
Isoforms B and C only differ in non-coding regions. Simply make B model, duplicate and rename for submission

41. Step 3: Investigate Exons
The last section of the Gene record includes the exons:

42. Step 3: Investigate Exons

43. Step 3: Investigate Exons
Use exon to search fosmid with exon
Where in this fosmid are sequences which code for amino acids that are similar
Best to search entire fosmid DNA sequence (easier to keep track of positions) with the amino acid sequences of exons
In the genome browser tab, go to the browser “chr10” of D. virilis; click the DNA button, then click “get DNA”
In the Gene Record Finder tab, make sure you have the peptide sequence for the melanogaster mav gene exons
These two tabs now have the two sequences you are going to compare

44. Step 3: Investigate Exons
Copy and paste the fosmid genomic sequence obtained from genome browser tab into the top box 1 of blastx
Copy and paste the peptide sequence of exon 1 from “gene record finder” tab into bottom box 2 of blastx
Open the “algorithm parameters” section: turn off the filter; leave other values at default
Click “BLAST” button to run the comparison

45. BLASTX Comparison

46. Step 3: Investigate Exons
Sometimes you have not find any similarity
If so: change the expect value to 1000 or even larger and click “BLAST”, keep raising the expect value until you get hits, then evaluate hits by position
This will show very weak similarity which can be better than nothing

47. Step 3: Investigate Exons
We have a weak alignment (50 identities and 94 similarities), but we have seen worse when comparing single exons from these two species
Notice the location of the hit (bases to 17504) and frame +3 and missing 92 aa

48. Step 3: Investigate Exons
A similar search with exon 2 sequences gives a location of chr10: and frame +2 with one amino acids missing

49. Step 3: Investigate Exons
For larger genes continue with each exon, searching with BLASTX for similarity (adjusting e cutoff if needed) noting location, frame and any missing amino acids
Very small exons may be undetectable, move on and come back later
Draw these out if this will help
Note unaligned amino acid as well
16,866
17,504
18,476
19,744 93
271 2
430
Frame +3
Frame +2

50. Step 4: Create a Gene Model
Pick ATG (met) at start of gene, first met in frame with coding region of similarity (+3)
For each putative intron/exon boundary compare location of BLASTX result to locate exact first and last base of the exon such that the conserved amino acids are linked together in a single long open reading frame
Exons: ;
Intron GT and AG present

51. Step 4: Create a Gene Model
For many genes the locations of donor and acceptor sites will be easily identified based on the locations of the alignments of the individual exons.
However when amino acid conservation is not sufficient to allow for the identification of intron/exon boundaries then other evidence must be considered. See the handout “Annotation Instruction sheet” for more help.

52. Splicing
The splice of life: Alternative splicing and neurological disease, B. Kate Dredge, Alexandros D. Polydorides & Robert B. Darnell, Nature Reviews Neuroscience 2, (January 2001)

53. Alternative Splicing
Splicing mutations can arise by alteration of conserved splice donor and splice acceptor sequences or by activation of cryptic splice sites
(A) Mutations at conserved splice donor (SD) or splice acceptor (SA) sequences result in intron retention where there is failure of splicing and an intron sequence is not excised; or in exon skipping where the spliceosome brings together the splice donor and splice acceptor sites of non-neighboring exons.
(B) Sequences that are very similar to the splice donor or splice acceptor sequences may coincidentally exist in introns and exons (sd and sa). These sequences are not normally used in splicing and so are known as cryptic splice sites. A mutation can activate a cryptic splice site by making the sequence more like the consensus splice donor or acceptor sequence and the cryptic splice site can now be recognized and used by the spliceosome (activation of the cryptic splice site).
From “Human molecular genetics”, Tom Strachan and Andrew Read

54. Alternative Splicing

55. Revisit “Phase”
Introns and internal exons are divided according to “phase”, which is closely related to the reading frame.
An intron which falls between codons is considered phase 0
An intron after the first base of a codon, phase 1
An intron after the second base of a codon, phase 2
Internal exons are similarly divided according to the phase of the previous intron (which determines the codon position of the first base-pair of the exon, hence the reading frame).

56. Step 5: Confirm Gene Model
We will do this later once you have a real model of your own to check
As a final check:
enter coordinates into “gene model checker” (gep.wustl.edu, projects  annotation resource  gene model checker) to confirm it is a valid model
Use custom tracks to view model and double check that the final model agrees with all your evidence
Examine dot plot to discover possible errors

57. Considerations
Some exons are very hard to find (small or non-conserved) keep increasing E value to find any hits; restrict location based on flanking exons
Donor “GC” seen on rare occasions
We have seen a couple of examples where the only reasonable interpretation was that the basic gene structure had changed (i.e. intron was gained or lost)
Use the techniques described in the Handout as well as discussions with your peers when things get difficult. Research is almost always a collaborative effort.

58. Complex Genes with Many Isoforms
Some genes will have many isoforms and many exons. The technique described for mav does not scale well to these large complex genes.
The recommended protocol in these cases:
Identify all unique exons and which isoforms have which exons (Gene record finder)
Map each unique exon
Build gene models of each isoform based on which exons it contains; re-use previously found exons

59. Homework 6 Read the “Annotation Instruction Sheet”
Go through “A Simple Annotation Problem” and do the exercises
Get familiar with “Annotation Report”
Read requirements for your annotation project report
Work on an annotation project (details are given in the next slide)
Read “Gene and Disease” (ebook) on NCBI website, select one genetic disease and prepare roughly 10 slides for your presentation given in next lecture (3/12/2012, provide the slides to me before the class)

60. Annotation Procedure
You are provided a zip file named “derecta_2nd3Lcontrol_Nov2011_fosmid22.zip”, unzip it you will get a folder named “derecta_2nd3Lcontrol_Nov2011_fosmid22”
Go into the provided folder, go to subfolder “src”, get sequence named “fosmid22.fasta.masked”. Use only this masked genomic sequence when the genomic sequence is needed. This sequence is produced by repeatmasker and the corresponding report can be found in subfolder “analysis”  “Repeats”
Go to select “D. erecta” for genome, “Nov.2011” for assembly and type “fosmid22” in the “position or search term” box, then click submit
In the genome browser, click one predicted gene from GenScan, obtain the predicted protein sequence in the next page
Use blastp on the flybase website (use “Annotated Protein” database), search this predicted protein against annotated D. mel proteins. Find the best match and determine the gene name in D. mel. Adjust blast parameters when needed
Search this gene in the GEP gene record finder, find D. mel gene details (exon amino acids sequences, gene structure etc.).
Search the masked genomic sequence against each amino acid sequence of exon in each isoform of D. mel using blastx on the NCBI website (The genomic sequence as query and each exon as subject). Record the matched regions (in the exon and in the genomic sequence), frame
Determine precise exon boundary by using signals (ATG, GT, AG, TAA,…), phase, conservation and frame information in the genome browser
Fill out the project report form