and why not to use standard blastall?

to create db just type:

formatdb -i your_sequences.fasta

to make search ( -a no of cores to use; -m tabular output):

blastall -pblastp -d your_sequences.fasta -i your_sequences.fasta -a no_of_threads -m8 -o blast_output

Of course, you can use on-line bl2seq and download the results from there. But command line would be much more convenient I suppose.

If you want to get pairwise similarity between all 1000 protein sequence in your dataset, try this:

1. Align your fasta sequence using a standard sequence alignment program
2. Use alistat (Written by Sean Eddy as part of SQUID - a C function library for sequence analysis) to get pairwise information.

Read Alistat man pages here. You can get various alignment statistics.

A percent pairwise alignment identity is defined as (idents / MIN(len1, len2)) where idents is the number of exact identities and len1, len2 are the unaligned lengths of the two sequences. The "average percent identity", "most related pair", and "most unrelated pair" of the alignment are the average, maximum, and minimum of all (N)(N-1)/2 pairs, respectively.

If you want to query a set of sequence against a custom-database of 1000 sequences: try A: How To Get Pair-Wise Sequence Similarity Score For > 1000 Protein. .

Probably the best choice is swps3. It performs vectorized Smith-Waterman alignment. 1000x1000 should not take long. Also, you may consider a multiple alignment tool (e.g. muscle) if these proteins are known to come from a gene family.