Harvard:Biophysics 101/2007/Notebook:Xiaodi Wu/2007-3-20

• Okay... looking at all the low-level sequence, blast, chromosome, and SNP parsing you guys have planned, I'm a little worried that you guys are making this too difficult. You can certainly stick with your original plans, but here is what I would do. First, recall that NCBI already hooked up blast with dbSNP at SNP BLAST. This allows you to perform blastn or tblastn and get SNP ids directly out of your query. I ran the March 13 example with all the default settings, which you can retrieve by inputing this request ID into the Retrieve results page.

1174240596-3341-64945077497.BLASTQ2

This gets you directly to the SNP id rs11200638 that you found in your previous homework with a single step. The tricky part may be figuring out how to interact with SNP BLAST from your scripts. I would recommend looking at the source of that page, specifically the hidden input tags, e.g. <INPUT TYPE="hidden" NAME="WWW_BLAST_TYPE" VALUE="SNP">. Those of you with cgi skills should sign up for this part and do it as early as possible. Once you can retrieve SNP ids, they can be passed to an OMIM module. Make sure everyone sychronizes on the input and output formats at each step so you can modularize everything and split up the tasks efficiently. —smd 14:39, 18 March 2007 (EDT)

• I'm going to start pseudocoding a bit with links to individual modules so people can start working on it. Primarily, I would like to start hammering down the specs so that when we write these, we can interface. --Mike 12:41, 18 March 2007 (EDT)

• I cleaned up this page a little. Please sign your comments with --~~~~. --smd 11:47, 18 March 2007 (EDT)

• Hey all, here's a sketch of the algo/pseudocode I propose for this...feel free to flesh it out and modify/edit/discuss

Blast original query

• Take sequence: look up on Blast (see http://www.dalkescientific.com/writings/NBN/blast_searching.html);
• Perhaps call this function sequence_lookup(str), returning some sort of object.
• For now, let's say the object includes best_gene_hit and best_genomic_position_hit, which includes chr (chromosome) and chrpos (position on chromosome), and the p-values for each match (or whatever they call them -- I think it might be called 'expected value')

Maybe we should call the object a blastcomparison. It would consist of a list of Match objects which each in turn contain the chromosome, chrpos, adjacent_genes (maybe itself a list) and p-values. Calling sequence_lookup(str) will then return a blastcomparison. (Yeah...I'm a C/java guy too...I want to call everything a vector...) --Mike 00:27, 18 March 2007 (EDT)

class BlastMatch
    chromosome (an integer)
    chrpos (a list where chrpos[0] is the start and chrpos[1] is the end)
    genes (a list of genes which span the sequence, if any)
    adjacentfeatures (a list of nearby features, if any)
    p (the p-value)
    def gene_match
       if(length(gene) > 0)
           return true (or alternately return the first match with highest p)
       else
           return false (or alternately -1 or something.  I think bool is easier)
    def feature_match
       same idea as the gene_match

class BlastComparison 
    all_comparisons (a list of BlastMatch types)
    def __init__()
          self.all_comparisons = []
    def populate (string):
       Query blast
       Parse results
       list_position = 0
       for i in all_blast_matches:
           self.all_comparisons[list_position].chromosome = i.chromosome
           Repeat to populate all features of BlastMatch

Notes:

• The output of initializing and then populating a blast comparison with a string will return a list of matches. It probably makes most sense to just go with the match with the highest p-value (the first one), but maybe we want to think of some other criterion to figure out which one to pursue in dbSNP/OMIM --Mike 12:57, 18 March 2007

Parse blast results

Has anyone gotten the snp search code zach has the link to in case 2 working? I'm assuming we can plug anything into the snps = GenBank.search_for('rs8192602', 'snp') segment to search by CHRPOS, gene name, etc. --Mike 13:22, 18 March 2007
Case 1: Direct gene hits

• If there is a gene hit (i.e. if sequence_lookup(str).best_gene_hit exists and has a p-value (expected value) above some threshold, then consider the given sequence to be part of a gene. Then...

• Translate gene (call this function translate_in_frame(str); I have an algo that goes through all the frames and finds the most likely ORF; works beautifully but a little slowly, but it will do -- we don't have to write this part of the algo), and locate mutations (locate_mutations(str, ref_str), returning a list (what in C would be an array -- I might slip into C lingo every so often so this is what I mean) containing the type of mutation (point mutation, insertion, deletion) in both a.a. and DNA sequence; again, I think we all have an algo for this)

• I have something that searches for orfs and then returns an object orf_list which specifies their positions in the original search string and parses into codons. In other words, each orf is a list of codon objects which in turn contain the sequence and Aa translation of that codon. My mutation detection function does pairwise comparisons of strings to identify types of mutations when one is considered a refseq. It then returns a list of Mutation objects (...I went object crazy...) --Mike00:27, 18 March 2007 (EDT)

• I also have code that can take a sequence compare it against a reference and then return the found mutations (of the type of interest for the March 13th assignment) within the nucleic acid seuence as well as the protein sequence. The code could also be easily modified to determine whether a given amino acid change is 'significant' (hydrophilic->hydrophobic, etc.) --Resmi 08:54, 18 March 2007 (EDT)

• As GMC mentioned in class, PolyPhen already scores the significance of amino acid changes (see examples). Consider figuring out ways to use existing resources, rather than re-implementing those tools. —smd 15:59, 18 March 2007 (EDT)

Case 2: non-gene Blast hits

• If there is no gene hit (like the example of 13 March, which was non-coding, supposedly), take the best_genomic_position

• Again, locate mutations, but only in the nucleotide sequence (locate_noncoding_mutations(str, ref_str)) and also maybe do a tblastx (or just blastx) [hrm...is this too much?]

• Look up the chromosome position in dbSNP and also this database, potentially: http://projects.tcag.ca/variation/

• This is looking better: At http://lists.open-bio.org/pipermail/biopython/2006-March/002971.html there is a modified biopython script to deal with SNP lookup... directly gives us the code, which is good.--Zsun 00:06, 18 March 2007 (EDT)

• Fnd the IDs of known SNPs and CNVs, compare to what we have about our own sequence, and then search OMIM with this info (call the function omim_noncoding_search, with parameters TBD)

• I think this would be an easier way to deal with OMIM since presumably if we parse the OMIM database into entries, they will be organized in this fashion → so less searching through entries until we find the specific one we are interested in. --Resmi 08:54, 18 March 2007 (EDT)

Perform OMIM search

• Look up these mutations for the gene on OMIM (call this function omim_gene_search(genbank_id, muts), where muts is a list of mutations from 2a to look for; for genes they are listed in OMIM in the format {amino acid}{position}{amino acid} instead of {nucleotide}{position}{nucleotide}; see http://eutils.ncbi.nlm.nih.gov/entrez/query/static/eutils_help.html for info on how to search all NCBI db's)

• This is not trivial; I could not find any way to program a link to OMIM, and there are no references online. The best I could find was the ability to download the OMIM database at: ftp://ftp.ncbi.nih.gov/repository/OMIM/ , thereby creating a 50MB text file which can then be parsed (that would be painful). Otherwise, there is a MySQL script which does the same, but that requires we know MySQL... http://search.cpan.org/~BIRNEY/bioperl-1.2.3/Bio/Phenotype/OMIM/OMIMparser.pm . --Zsun 00:00, 18 March 2007 (EDT)

• Yeah....I'm pretty against working with OMIM unless we can get a fast server to run this on and just let it sit... I still have to look at the eutils but that looks promising (Zach, you said it was already implemented in Biopython?) --Mike 00:27, 18 March 2007 (EDT)

• Well, EUtils does look like the most promising path. I'm sorry for not highlighting that earlier, but the link that I pointed to above (http://eutils.ncbi.nlm.nih.gov/entrez/query/static/eutils_help.html) has a good amount of info on how this works, though I can't say that I've looked at it closely yet --wuxiaodi 03:12, 18 March 2007 (EDT)
• EUtils is indeed implemented in Biopython. I got their example case working on my computer; unfortunately, that's all the documentation they have! Interestingly, EUtils seems to have a class equivalent to the "Links" button. This could be a nicer way to move from the geneID to all the other stuff we want. Could be. I'm going to have to go source-code diving to figure out the syntax/what it actually does...will report back. --Kay 12:22, 18 March 2007 (EDT)