Gene annotation of genomes

ABSTRACT

We analyze an anonymous human genome contig using a variety of computational tools. We first download the sequence. Then, first locate repetitive elements along the genomic sequence. After masking the sequence accordingly, we use a number of gene prediction programs to obtain an initial delimitation of putative genes encoded along the sequence. We then validate predictions comparing them against databases of known coding sequences (ESTs), which also help to infer gene variability by means of EST spliced alignments.

Programs and languages

In this practical we use several well known bioinformatics programs, as well as gawk, perl and UNIX commands and scripts. Being fluent on the command-line is important.

Downloading the sequence

We are going to work with the contig AC091491.embl

Obtaining the sequence in fasta format

the following command does the job

grep '^ ' AC091491.embl | sed 's/[ 0-9]//g' > AC091491.fa

however, the fasta format requires each line to be preceeded by the '>seqname' line. We use emacs to introduce this line, and make the file AC091491.fa conform the fasta format.

Of course, there is a number of ways in which we can convert the original file AC091491.embl to the fasta formated one AC091491.fa, without the need of editing the file. Can you think of one such way?

It may be useful to have also the sequence in tabular format. The following script will convert the fasta file into a tabular file.


awk '{printf "%s", $0}'  AC091491.fa > AC091491.tbl

Again, we need to use emacs to introduce a \tab between the sequence name and the sequence itself, to get AC091491.tbl in the prober tabular format. And, again a number of simple unix commands will do the whole conversion automatically. Can you think on such command? (Check here).

The tabular format allows to perform some preliminary analysis on the problem sequence

length of the sequence


awk '{print length($2)}' AC091491.tbl 
152882 

g+c content

awk '{print $2}' AC091491.tbl | fold -1 | sort | uniq -c | gawk '{print $2, $1/152882}' 
a 0.323073
c 0.200802
g 0.189257
t 0.286868

identify repeatitive elements in the sequence

A large fraction of genomic DNA belongs to relative small number of repeat families. A popular program to identify repeats in DNA sequences is RepeatMasker. There are a number of web servers which offer RepeatMasker analysis. We will use the EMBL RepeatMasker server.

RepeatMasker analysis produces a number of files. Among them:

By taking a look at this last file, we see that close to 40% of our sequence belong to repetitive DNA.

It may be interesting to visualize the distribution of repeats along the sequence. There are a number of available tools to visualize genome annotations. We will use here gff2ps. There is a gff2ps web server at the pasteur institute, but the software can also be installed locally.

gff2ps requires the input file in gff format. The following awk script will do the conversion:


grep AC091491 AC091491.seq.out | awk 'BEGIN{OFS="\t"}{print $5, $11, "repeat", $6,$7,".", ".", "."}' > AC091491.seq.out.gff

Now we can run gff2ps: be careful about the path of the perl interpreter (try which perl);

gff2ps AC091491.seq.out.gff > AC091491.seq.out.ps

which we can visualize using ghostview:

gv  AC091491.seq.out.ps

"ab initio" gene prediction

We use a number of gene predictions programs to predict the genes along the sequence. For instance, we will access the imim geneid server and the mit genscan server. We save the predictions in the files: Both predictions share a number of exons, but they are not equivalent. It will be useful to visualize both of them simultaniously. If we use gff2ps, we need first to convert them to gff. The geneid server already produces gff output. Thus we can run again the sequence on the server and obtain the gff formatted predictions, for instance in the file AC091491.geneid.gff. For genscan, we can type, for example:

gawk 'BEGIN{OFS="\t"}$2 ~ /Term|Intr|Init/ {print "AC091491", "genscan", $2, start=($4<$5 ? $4 : $5), end=($5<$4 ? $4 : $5),$13,$3,$7, $1}'  AC091491.genscan  | sed 's/\.[0-9][0-9]$//' > AC091491.genscan.gff

Now we can run gff2ps and visualize the two predictions simultaniously:

gff2ps AC091491.geneid.gff AC091491.genscan.gff > AC091491.genepredictions.ps

Gene prediction validation

We will try now to validate the gene predictions, and investigate if there exist additional information which can help us to infer the most likely exonic structure for this gene.

Est database searches

We compare the query sequence agains the database of human ESTs using blast. We can use the ncbi blast server. Because we are using a large sequence, and we are only interested in nearly identical matches, we use megablast. Since we will post-processing the blast output, we will initially ask for the output to be produced in plain text.

After running the blast search, we store the results in, for instance, the file AC091491.blast.est

These are difficult to analyze. We will first visualize them using gff2ps. This requires converting the blast result to gff. It is possible to write a relatively easy perl script to do the conversion. Here, we will borrow one such script written by Josep F. Abril. The name of the script is parseblast. Then, we just type


parseblast -Gi AC091491.blast.est | gawk 'BEGIN{OFS="\t"}{$2=$9;$6=$7=".";print}' > AC091491.blast.est.gff

And now we are ready to plot the est matches along the query sequence:


gff2ps AC091491.blast.est.gff > AC091491.blast.est.ps

As you can see, there is a large number of singleton EST matches. Because artefacts during EST library construction are not uncommon, usually only spliced ESTs are considered good evidence for the existence of a gene, when identified along a genomic sequence.

Let's consider, thus, only the spliced EST matches. To obtain them, we just need to identify those hps's appearing more than once in the gff formatted est search file. You can write a simple perl script for this. I wrote an awk script (save it in a file: getsplicedhsp.awk), instead, and named the resulting gff file of splices est matches, as AC091491.blast.est.spliced.gff


gawk -f getsplicedhsp.awk AC091491.blast.est.gff > AC091491.blast.est.spliced.gff

Now we can visualize est matches altogether with gene predictions


gff2ps AC091491.blast.est.spliced.gff AC091491.geneid.gff AC091491.genscan.gff > AC091491.est+genepredictions.ps

We can convert the postscript output to gif (or jpeg) format, and include it within the html document.

convert -antialias -rotate 90 AC091491.est+genepredictions.ps AC091491.est+genepredictions.gif 

After looking at the plot, it appears, in this case, that the geneid prediction is quite consistent with the EST matches, with the exception of exon 68656-68712. genscan appears to predict a number of additional exons. Note, however, that the large exon predicted both by genscan and geneid is only partially supported by EST matches. In general, this does not necessarily means that the exon predicted by the "ab initio" gene prediction is incorrect. Because of the mechanism by means of which ESTs are obtained, 3' ESTs do generally support the 3' end of a gene, while 5'ESTs are often internal to the gene, and do not cover its 5' end. The ESTs matching our query sequence are


>gb|AA772973.1|AA772973 af33h09.s1 Soares_total_fetus_Nb2HF8_9w Homo sapiens cDNA clone
             IMAGE:1033505 3' similar to TR:Q15111 Q15111 PHOSPHOLIPASE C. ;.
          Length = 471

>emb|AL526158.1|AL526158 AL526158 LTI_NFL003_NBC3 Homo sapiens cDNA clone CS0DC015YO17 5 prime.
          Length = 926

>gb|BG505795.1|BG505795 601860281F2 NIH_MGC_61 Homo sapiens cDNA clone IMAGE:4072604 5'.
          Length = 630

>gb|R41142.1|R41142 Hk907-f Adult heart, Clontech Homo sapiens cDNA clone k907-f.
          Length = 355

>gb|W21387.1|W21387 zb59h03.r1 Soares_fetal_lung_NbHL19W Homo sapiens cDNA clone
            IMAGE:307925 5'.
          Length = 254

Because supported by a 3'EST (AA772973), we can we quite confident of the 3' end of the gene. However, despite a number of 5'ESTs, we can not guarantee the 5'end of the gene, which could extend quite beyond this very large exon (as, both, the genscan and the geneid prediction suggest) Thus, we conclude that a likely exonic structure for the (apparently) only gene in contig AC091491 is

AC091491	prediction	Terminal	9279	9458	 .	-	0
AC091491	prediction	Internal	31099	31235	 .	-	2
AC091491	prediction	Internal	56242	56317	 .	-	0
AC091491	prediction	Internal	84280	84483	 .	-	0
AC091491	prediction	Internal	87031	89517	 .	-	0

There is one EST (BG505795) suggesting an alternative form for this gene, in which the short exon 56242-56317 is missing. There are two spliced ESTs (BI021866 and W21387) wich support exonic structure incompatible with the exonic structure supported by most ESTs and predicted by the "ab initio" gene prediction programs, and are deemed as artefactual.

Analysis of the predicted proteins

Now is time to try to caracterize the predicted protein product. We should start by comparing the deduced amino acid sequence against the set of all known proteins, using blastp. We will discover that the amino acid sequence predicted here corresponds to a well known protein.
Roderic Guigó, rguigo@imim.es
2002-02-24