est2genome aids the prediction of genes by sequence homology. It aligns a set of spliced nucleotide sequences (ESTs cDNAs or mRNAs) to an unspliced genomic DNA sequence, inserting introns of arbitrary length when needed. Where feasible introns start and stop at the splice consensus dinucleotides GT and AG.
By default, est2genome makes three alignments: First it compares both strands of the spliced sequence against the forward strand of the genomic sequence, assuming the splice consensus GT/AG (ie in the forward gene direction). The maximum-scoring orientation is then realigned assuming the splice consensus CT/AC (ie in the reversed gene direction). By default, only the overall maximum-scoring alignment is reported, and then if it scores higher than a specific minimum threshold score. Optionally, all comparisons may be reported.
The program outputs a list of the exons and introns it has found. The format is like that of MSPcrunch, ie a list of matching segments. This format is easy to parse into other software. The program also indicates, based on the splice site information, the gene's predicted direction of transcription. Optionally the full sequence alignment is printed as well.
The algorithm has the following steps:
|
|
The score for Exon segments is the alignment score excluding flanking intron penalties. The Span score is the total including the intron costs.
The coordinates of the genomic sequence always refer to the positive strand, but are swapped if the est has been reversed. The splice direction of Introns are indicated as +Intron (forward, splice sites GT/AG) or -Intron (reverse, splice sites CT/AC), or ?Intron (unknown direction). Segment entries give the alignment as a series of ungapped matching segments.
parameter default description
match 1 score for matching two bases
mismatch 1 cost for mismatching two bases
gap_penalty 2 cost for deleting a single base in
either sequence,
excluding introns
intron_penalty 40 cost for an intron, independent of
length.
splice_penalty 20 cost for an intron, independent of
length and starting/ending on
donor-acceptor sites.
space 10 Space threshold (in megabytes)
for linear-space recursion. If the
product of the two sequence
lengths divided by 4 exceeds this then
a divide-and-conquer strategy is used
to control the memory requirements.
In this way very long sequences can
be aligned.
If you have a machine with plenty of
memory you can raise this parameter
(but do not exceed the machine's
physical RAM)
However, normally you should not need
to change this parameter.
There is no gap initiation cost for short gaps, just a penalty
proportional to the length of the gap. Thus the cost of inserting a
gap of length L in the EST is L*gap_penaltyand the cost in the genome is
min { L*gap_penalty, intron_penalty } or
min { L*gap_penalty, splice_penalty } if the gap starts with GT and ends with AG
(or CT/AC if splice direction reversed)
Introns are not allowed in the EST. The difference between the
intron_penalty and splice_penalty allows for some slack in marking the
intron end-points. It is often the case that the best intron
boundaries, from the point of view of minimising mismatches, will not
coincide exactly with the splice consensus, so provided the difference
between the intron/splice penalties outweighs the extra mismatch/indel
costs the alignment will respect the proper boundaries. If the
alignment still prefers boundaries which don't start and end with the
splice consensus then this may indicate errors in the sequences.
The default parameters work well, except for very short exons (length less than the splice_penalty, approx) which may be skipped. The intron penalties should not be set to less that the maximum expected random match between the sequences (typically 10-15 bp) in order to avoid spurious matches.
The original program was est_genome, written by Richard Mott at the Sanger Centre. The original version is available from ftp://ftp.sanger.ac.uk/pub/pmr/est_genome.4.tar.Z