Why do other chicken EST sets appear to have more known genes in them?

It has been brought to our attention that our collection of chicken ESTs have a much lower fraction of hits with the known databases. This certainly seems to be the case. We have analysed a few subsets of chicken ESTs in the public domain.

EST or Contig set	Number of sequences	BLAST hits vs Swiss-Trembl
EST or Contig set	Number of sequences	1E-03	1E-06	70 bits
Riken set 1	7,410	57%	54%	49%
EMBL EST set 2	23,026	74%	71%	67%
Assembled set 2	10,068	67%	63%	60%
BBSRC ESTs	330,388	50%	48%	45%
BBSRC assembly	85,486	39%	37%	35%
BBSRC+Genbank contigs (All)	97,221	42%	39%	35%
BBSRC+Genbank contigs (BBSRC only)	73,023	35%	32%	28%
BBSRC+Genbank contigs (Genbank only)	8,637	43%	39%	34%
BBSRC+Genbank mixed contigs	15,561	75%	72%	70%

Notes: The Riken set 1 corresponds to those ESTs published from the Buerstedde group, whilst the EMBL EST set 2 include the Riken set plus further ESTs deposited from the Buerstedde group not originally listed in the paper. This set were assembled using PHRAP to produce 10,068 contings (Assembled set 2). The BBSRC project ESTs and contigs are considered next. Finally, the last 5 rows refer to contigs produced by assembling the BBSRC ESTs and 60,000 chicken ESTs in Genbank to give 97,221 contigs. Only 15,561 of these are shared by both the BBSRC and Genbank ESTs.

As can be seen, the BBSRC set has a relatively low percentage of "known" genes in it compared to other sets. We believe this is for a number of reasons:

Our collection of ESTs is much larger than previous sets and the shear size of the collection suggests we are less likely to find known genes. If our contig set contained the same fraction of "known" genes as the "Assembled set 2" then we would have over 55,000 different genes - a number that is likely to easily exceed the total number in the chicken genome !
Our libraries went through extensive, stringent normalisation procedures which will have removed many of the abundant (and therefore common) genes. Hence, we discover more unusual genes (which are therefore less likely to have matches in the database).
Our strategy focussed on minimising the number of known genes. We had several stages during our sequencing cycle where we assessed the "uniqueness" of a library in comparison with all others, which led us to pick clones from the most informative ones in our opinion. Thus we focussed on tissues and libraries in particular which were not yielding lots of well-known genes with characterised homologues.
Our ESTs do appear to be unusual/rare clones. Considering the data above, contigs generated in the BBSRC+GENBANK assembly which are shared by both projects have a very high "hit" rate to the databases. The contigs which are less common (ie. Unique to either BBSRC or GENBANK) have much lower hit rates.
Our assembly protocol differs slightly from others used. Although we have used PHRAP which is believed to over-assemble contigs, we use a BLAST-based pre-filter which attempts to keep genuinely dissimilar clones apart. Nevertheless, this may lead to an effect on the number of contigs which affects the number of contigs with BLAST hits. We note that in general, the overall percentage with BLAST hits falls off when a set of ESTs is assembled as all the redundant clones with the same BLAST hit are all brought together in one contig.