The Toy Exon Finder

1
The Toy Exon Finder
2
The Toy genome

• The genome is very dense with genes, typically no
  more than 20 bp between successive genes on a
  chromosome
• The exons tend to be very small, on the order of
  20 bp
• The introns in multi-exon genes tend to be very
  small, about 20 bp
• The exons incorporate fairly strong codon biases
  and a significant GC bias
• The splice sites, start codons, and stop codons
  are flanked by positions with fairly strong base
  composition biases
• The codon usage and base composition statistics
  can be well characterized with some sample data
• Genes occur only on one strand, which we will
  call the Forward strand

3
Toyscan 1

• Use GC content to find exons
• Find all ORFs such that each ORF either
• Begins with a START and ends with a STOP
• Begins with a START and ends with a GT
• Begins with AG and ends with GT
• Begins with AG and ends with STOP
• Set threshold t such that if an exon has GC
  content below t, label it as noncoding
• For all remaining pairs of ORFs p1, p2, do
• If p1 and p2 overlap, then discard with ORF with
  lower GC content
• Output all ORFs that remain, calling them exons

4
Toyscan 2

• Use codon bias to find exons
• Codon frequencies for true exons are assumed to
  be known
• Stop codons not included so they have probability
  0
• Define codonBias function
• For an input ORF (given), score all 3 frames
• Ignore the fact that some frames have stop codons
  in them
• Score sum of log probabilities of all codons in
  that frame
• Probabilities are taken from the known
  probabilities
• Divide Score by number of codons n. This
  normalizes it.
• Output the highest score of the 3 frames

5
Toyscan 2 (cont.)

• Note the codonBias function achieves its maximum
  when the observed distribution within an ORF
  matches the correct distribution from real
  genes
• Define TOYSCAN_2 as
• A codon bias score threshold, t, is input
• For all ORFs, score them with the codonBias
  function
• If the score is lt t, delete the ORF
• For all remaining pairs of ORFs p1, p2, do
• If p1, p2 overlap then discard the ORF with the
  lower codonBias score
• Output all remaining ORFs as exons

6
Toyscan 3

• Use codon bias and weight matrix models (WMMs)
• Input includes WMMs for start, stop, donor, and
  acceptor sites
• Donor WMM includes 5 positions after GT
• Acceptor WMM includes 5 positions before AG
• Start codon WMM includes 5 positions before ATG
• Stop codon WMM includes 5 positions after
  TAA/TGA/TGA

7
Toyscan 3 (cont.)

• Score a weight matrix (scoreWMM)
• For each position i in the sequence S, sum the
  log probabilities of the bases in the interval
  (i,j) using the WMM, where j-i1 is the width of
  the WMM
• Score an ORF (scoreORF)
• choose the matrices to use on the left and right
  ends of the ORF
• E.g., internal exon has acceptor on left, donor
  on right
• Score WMM(left end) WMM(right end)
  codonBias
• return Score

8
Toyscan 3 (cont.)

• Now define Toyscan_3 as
• assume a scoring threshold, t, is provided
• You will have to experiment to find a good value
  for t
• Get all ORFs
• Score all ORFs using the scoreORF procedure
• If the score is lt t, delete the ORF
• For all remaining pairs of ORFs p1, p2, do
• If p1, p2 overlap then discard the ORF with the
  lower scoreORF score
• Output all remaining ORFs as exons

9
GFF format

• coding GC 49
• noncoding GC 50
• 1 toy-genome initial-exon 31 46 . . transgrp1
• 1 toy-genome final-exon 79 98 . . transgrp1
• 1 toy-genome single-exon 129 140 .
  . transgrp2
• 1 toy-genome single-exon 164 193 .
  . transgrp3
• 1 toy-genome single-exon 228 260 .
  . transgrp4
• 1 toy-genome single-exon 287 304 .
  . transgrp5
• 1 toy-genome single-exon 331 354 .
  . transgrp6
• 1 toy-genome single-exon 377 400 .
  . transgrp7
• 1 toy-genome single-exon 424 435 .
  . transgrp8
• 1 toy-genome initial-exon 475 488 .
  . transgrp9
• 1 toy-genome internal-exon 512 526 .
  . transgrp9
• 1 toy-genome final-exon 545 593 . . transgrp9