04 - Bacterial Gene Prediction

1. Obtain new genomic DNA sequence

2. Translate all 6 ORFs and compare them to the protein database

3. Similarity search of expressed sequence tag (EST) DB of the same organism, or cDNA sequences if available

4. Gene prediction tool to locate genes

5. Analyse the regulatory sequence in the gene

Identifying the regions of genomic DNA that encode genes

start: ATG

stop: TAA, TAG, TGA

(A can also be U)

• Intrinsic:

  • sufficient length of ORFs, long ORFs rarely occur by chance

  • special codon patterns in comparison to non-coding regions

  • ribosomal binding sites upstream of start codon

• extrinsic:

  • Similarity to known gene products

sequence of DNA -> can be translated into protein

• starting with start codon

• ending with stop codon

Purine-rich sequence in bacterial mRNA —> recognized by the ribosome for translation initiation

5-10 basepairs upstream from the start codon

conserved sequence found in bacterial promoters

typically located at -10 relative to the transcription start site (upstream)

= sequence upstream of start codon

—> helps direct the ribosome to correct translation start position

Codon usage bias refers to preference for certain codons over others in coding sequences

—> can help distinguish coding regions from non-coding regions

= statistical model —> represent probability of sequences

particularly useful for identifying gene structures in genomic data

• GeneMark

  • non-homogenous HMMs for coding regions

  • homogenous HMMs for non-coding regions

  • coding capacity of sliding windows is deduced through a Bayesian decision rule

• GLIMMER

  • interpolated HMM

  • takes into account DNA oligomers of varying lengths dependent on the local composition of the sequence

• EcoParse

  • HMM

  • maximum likelyhood of a sequence in coding and non-coding

  • no sliding window

—> 3 ORFs per strand

• Factors for unequal codon usage in coding regions:

  • Unequal AA usage

  • Unequal number of codons for different AA

  • codon preference

• Factors for a coding RF1:

  • AA composition in RF1-3

  • Codon composition of RF1-3

  • Positional base frequency: freq. with which each of the four bases occupies each of the three positions within codons

• able to model “grammar”

• words are codons

• random walk starting in the middle of any of the HMMs

• Choosing one of the 61 codon models repeatedly results in a 'random gene‘

• Gene termination

• Intergenic region

• Start codon HMM

• Transition is made back to the central state and the whole process repeated

—> Result: a sequence of nucleotides that is statistically similar to a contig of E. coli DNA (or any other training set) consisting of a collection of genes interspersed with intergenic regions

Most likely random walk:

• Viterbi algorithm

• find the most probable sequence of gene structures (hidden states) that explains the observed nucleotide sequence

• identification of coding regions, introns, exons, and other genomic features

• Class I genes

  • intermediate codon usage bias

  • low/ intermediate level of expression

  • environmental triggers —> high expression of some genes (rare)

• Class II genes

  • high codon usage bias

  • highly expressed under exponential growth conditions

• Class III genes

  • low codon usage bias

  • plasmids and insertion sequences

  • can be expressed at a fairly high level

  • includes genes coding for fimbriae, major pili, membrane proteins, …

• used to verify intrinsic methods

• e.g. ORPHEUS