Your approach doesn't seem to be logical. You need to read a bit about processing your data from shotgun sequencing and make assemblies. And generally, people do not use protein sequences against genomes for similarity searching.

Well I need to search in several genomes of bacteria if there are some functional homologs to my set of proteins (proteins from specific PUL - Polysaccharide Utilization Loci) and then I need to look around these homologs to identify if I found PUL and what kind of PUL is it - if it's possible that it utilizes similar polysacharide as my reference. Due to search in different classes of bacteria I can't find homologs at DNA level but I can at protein level. That's the reason why I need to map protein sequences. At NCBI I found mainly shotgun sequences of bacteria genoms in which I have to search. These genoms are devided into dozens not connected scaffolds and that is problem for me now... how can I process these data from shotgun genomes? If they are not connected in NCBI database it means they can't be connected until there will be some reference full genome assembly?

You can simple do blastP (protein vs protein), and get the genome coordinates from your blast hit.

Dear martin

Just for the curiosuty try tblastn in ncbi. Using protein query and genome as reference. I hope that might help u