PSSM (Position Specific Scoring Matrix) from NCBI convert to matrix format

Question

[gaiboyan23]

When I run psi-blast (3 iterations) to download the PSSM (Position Specific Scoring Matrix), the files come up as ASN.1 files. How can i convert these ASN.1 files to matrices. The PSSMs should look like matrices with 20 columns (for 20 amino acids) and L rows (L=length of protein)

However, the files I download are ASN.1 files, which look like this:

numRows 28,         
numColumns 555,         
swissprot {accession "q2m32 9"}             
inst {          
repr raw,           
mol aa,         
length 555                  
{ 0, 10, 0 },           
{ 634267618256118,  10, -16 },
{ 0, 10, 0 },           
{ 129176519709001,  10, -16 },
{ 182041473572035,  10, -16 },
{ 292541684289281,  10, -16 }
.....

How can I download/convert this ASN.1 format to what a PSSM matrix is supposed to look like?

Answer 1

[bjwiley23]

I have been looking for a resource to indicate what the 28 rows align with so I can programattically do this but have not found a resource yet. In the mean time you can upload your Scoremat.asn file to the pssm_viewer, choose the file, then click 'matrix view', and 'download matrix to file'.

EDIT: I got a response from Christiam C. at NCBI. The 28 rows are based off the order of the array NCBISTDAA_TO_AMINOACID (see definition below): https://www.ncbi.nlm.nih.gov/IEB/ToolBox/CPP_DOC/lxr/source/src/algo/blast/core/blast_encoding.c#L115

I confirmed the order with a PSI-blast.

[leeeiqi]

Hello！Do you have a script for batch conversion of Scoremat.asn to PSSM matrix format? Thank you for taking the time to answer me.

[leeeiqi]

Hi, can you help me? I'm at my wit's end. Could you show me an example of Python parsing? Thank you very much !!!!!!

[bjwiley23]

Hi @leeiqi,

I don't think I have the python script anymore. To do this you really just need to see the order of the amino acids in https://www.ncbi.nlm.nih.gov/IEB/ToolBox/CPP_DOC/lxr/source/src/algo/blast/core/blast_encoding.c#0115 (see line 115) in which there are 28 I believe and have that amino acid order. So you need to parse out all the numbers using a regex matcher between

finalData {
      scores {

and

},
      lambda { 0, 10, 0 },

It should have a length that is a multiple of 28 and your amino acid length. So its a 1-d vector of (28 x n) and then you need to reshape it with python numpy so that it is 2-d as in (n x 28) where you have n rows and 28 columns. Then only keep the amino acids you need using list comprehension. Like you don't need '-','Z','U','*','O', 'J'... So those columns you would remove by their number and you have the PSSM!