Hi,

The Ensembl pipeline starts with protein sequences (the longest protein for every gene in Ensembl), and calculates BLAST reciprocal hits. After that, M-coffee and TreeBest are used to make the tree, and to determine homology relationships. The full pipeline is here:

http://www.ensembl.org/info/docs/compara/homology_method.html

Without knowing what group of proteins you are starting with, or your PSI BLAST parameters, I'm guessing that initial step of BLAST+Smith Waterman is picking up more relationships. You're welcome to try our pipeline.

I would go with more ortholog control database like: orthoMCL or inparanoid, to blast against their databases and look for orthologs. This will give you a better over-all look for orthologs of your gene. Ensembl is nice, but I prefer relying on either ENTREZ refseq proteins or uniprotKB, to avoid duplication, and more controled blast and orthologs groups. plus, they have better interface.

Most probably the differences between your gene tree and the Ensembl genetree are due to differences in the methods used to analyse such data.

The parameters used to include or exclude sequences into a gene family can make it small but tightly aligned or large but sparsely aligned.

Ensembl uses a combination of multiple aligners and TreeBest to generate the tree from the alignments:

http://www.ensembl.org/info/docs/compara/homology_method.html