Gene Finding

Question 1

Describe the eukaryotic genome structure.

Answer 1

• Exons and introns that make up the ORF
• The promoter is the nucleotide sequence to which RNA polymerase binds to initiate transcription
• Initiation codon (usually ATG)
• Termination codons (TAA, TAG, TGA)
• LINEs, SINEs, LTR elements

Question 2

What is TRY4?

Answer 2

• A gene that is responsible for the synthesis of trypsinogen (trypsin precursor)
• Discontinuous gene (split into 5 exons and 4 introns)

Question 3

What are V28 and V29-1?

Answer 3

• Discontinuous gene segments that specify part of the beta T-cell receptor protein
• Not complete genes and must be linked to each other by gene splicing
• Results in a permanent genome change during cell differentiation

Question 4

What does gene finding aim to find?

Answer 4

• Which regions code for proteins
• Gene start and end regions
• Exon/intron boundaries
• Regulatory sequences

Question 5

Describe prokaryotic genome finding.

Answer 5

Prokaryotes have small genomes (0.5-5 Mbp) with a high coding density (>90%) and no introns. This makes gene identification relatively easy (~99% success rate). Problems include overlapping ORFs (due to the prokaryotic genome being so small), short genes missing the cutoff point (roughly 50 aa), and finding promoters.

Question 6

Describe eukaryotic genome finding.

Answer 6

Eukaryotes have large genomes (10-120,000 Mbp) with a low coding density (<50%, 2-3% in humans), and they contain introns. This makes gene identification relatively difficult (~50% success rate). There are many problems in eukaryotic genome finding.

Question 7

Methods of gene finding

Answer 7

• Ab initio methods
• Similarity-based methods
• Integrated approaches

Question 8

What are Ab initio methods?

Answer 8

Making predictions based on typical gene features such as splice signals and sequence composition. Regions to look for include initial 5’ exons, internal exons, and final 3’ exons.

Question 9

What are similarity-based methods?

Answer 9

• Predict genes by adding information from the whole genome sequence from related species
• The similarity between the query sequence and the known coding sequence are used to infer gene structures
• Allows predicted exon sequences located by ORF scanning to be tested for functionality
• Results are influenced by the availability of close homologues

Question 10

Problems with similarity-based methods

Answer 10

• Different genes are expressed differently in different tissues so you can’t just sequence mRNA from a random tissue, you need to sequence everything under a lot of different conditions
• Might be different due to alternative splicing

Question 11

ORF scanning in prokaryotes

Answer 11

• Search DNA for start and stop codons (may end up with false positives)
• Search for promoters
• Since most genes are >50 codons you apply a cutoff of ~100, but you also lose the short genes

Question 12

ORF scanning in eukaryotes

Answer 12

• Search DNA sequence for start and stop codons
• Search for promoters
• Search for intron/exon boundaries (spliced mRNA does not contain introns)
• Since most genes are >50 codons you apply a cutoff of ~100

Question 13

Codon usage in genomes

Answer 13

• Codons are used unequally, this is a universal feature of genomes
• You can use this to differentiate coding and non-coding regions (e.g. humans use GTG 4x more than GTA for valine)
• Real exons are expected to show codon bias but introns should not

Question 14

ORF scanning and moving windows

Answer 14

Sequence information only is used to identify coding exons through integrating coding statistics. We want to calculate the likelihood that a triplet is in a coding region and plot a graph of it (above zero is likely below zero is unlikely)

Question 15

ORF scanning: exon/intron boundaries

Answer 15

Exon-intron boundaries have distinctive sequence features.
- Upstream boundary: invariant GT and consensus sequence
- Downstream boundary: T or C, any amino acid, then CAG

Question 16

ORF scanning: upstream regulatory sequences

Answer 16

Locate where genes begin using distinct sequence features (e.g. recognition signals for DNA-binding proteins). Regulatory sequences are variable and difficult to incorporate into gene prediction algorithms)

Question 17

Best Ab initio methods

Answer 17

Based on HMMs, a machine learning approach that takes sequences and encodes them in a statistical framework

Question 18

Examples of Ab initio methods

Answer 18

GENSCAN, HMMgene, GeneMark

Question 19

What is GenScan?

Answer 19

GenScan identifies complete intron/exon structures of genes in genomic DNA and predicts multiple genes, partial and complete genes. It uses HMM to model gene structure and has separate HMMs for exons, introns, and intergenic regions. There are different parameters for regions with different GC content.

Question 20

P values in GenScan

Answer 20

P is the probability that the exon is correct.
When P>0.99, the exon is almost exactly correct.
0.50<=P<=0.99, the exon is correct most of the time.
P<0.50, not reliable

Question 21

Sensitivity (nucleotide level accuracy)

Answer 21

no. of correct exons/no. of actual exons

Question 22

Specificity (nucleotide level accuracy)

Answer 22

no. of correct exons/no. of predicted exons

Question 23

Sensitivity (exon level accuracy)

Answer 23

true prediction/(actual exons + missed exons)

Question 24

Specificity (exon level accuracy)

Answer 24

true prediction/(true prediction + false prediction)

Question 25

Advantages of Ab initio gene prediction

Answer 25

• Good at predicting nucleotides (>90%)
• Improve accuracy by combining methods
• Easier for prokaryotes because there are no introns

Question 26

Disadvantages of Ab initio gene prediction

Answer 26

• Moderate at finding exon boundaries (70-75% correct per exon)
• Poor at predicting complete gene structure (<50%)
• Need to identify upstream regulatory sequences

Question 27

What did Rogic et al. (2002) publish

Answer 27

A way to improve gene prediction accuracy y combining GenScan and HMMgene.
- OR returns regions predicted by either program (more sensitive, less specific), leading to overprediction
- AND returns regions predicted by both programs (less sensitive, more specific), leading to underprediction
- EUI (exon union-intersection), OR if above a given significance threshold, AND otherwise

Question 28

Examples of similarity-based gene prediction methods

Answer 28

AAT, GeneWise, SGP2, Rosetta

Question 29

What is SGP2

Answer 29

Synthetic Gene Prediction, the query sequence is compared against sequences from the informant genome. The results of the comparison are used to modify exons predicted by ab initio prediction. It uses an integrated approach.

Question 30

How do we confirm putative genes?

Answer 30

1. If the gene is matched to one or more ESTs from the same organism.
2. There is a similarity of the nucleotide/translated protein sequence to sequences in databases.
3. There is a match for the protein sequence in a secondary databank.
4. There is an association with predicted promoter sequences.

Question 31

Problems with gene finding in general

Answer 31

• Imprecise or incomplete
• Splicing incorrect
• False positives
• Failure to identify true genes
• Doesn’t account for PTHMs (ligands, glycosylation, methylation, peptide excision)