Protein Features Documentation

Running findsitemetal

It's a bit more involved to run findsitemetal, since you need to:

  1. Generate a template list
  2. Set required variables
  3. Run findsite

Generating a template list

First, make sure that you've generated template lists for every UniProt ID of interest. We do this with LOMETS.

If you also used LOMETS, then you need to parse the init.dat files to produce a simple list of templates. You can do this by making a folder called LOMETSinitdatFiles were you put each init.dat file. Label each file with the UniProt id, for example:

(base) [js741@compute-e-16-233 LOMETSinitdatFiles]$ ll | head -3
total 587656
-rw-rw-r-- 1 js741 js741 1416739 May  4 22:11 O00444_init.dat
-rw-rw-r-- 1 js741 js741 1905099 May  4 22:11 O15020_init.dat

Then, make a folder called FINDSITE_input (at the same level as LOMETSinitdatFiles) and run LOMETS_to_FINDSITE_files_js.py (included in this repo and in /n/groups/drad/julian/findsite). That python script requires FINDSITE_PDB_mapping.csv, also included in the same directories, and will produce .txt files with template lists:

(base) [js741@compute-e-16-233 FINDSITE_input]$ ll | head -3
total 9864
-rw-rw-r-- 1 js741 js741  36 May  4 22:21 O00444.txt
-rw-rw-r-- 1 js741 js741  78 May  4 22:21 O15020.txt
(base) [js741@compute-e-16-233 FINDSITE_input]$ head -5 O00444.txt
1y8ga
2euea
1muoa
1ia8a
2ou7a

Those .txt files will be your template input when running findsite.

Setting required environmental variables

findsite requires a few environmental variables. As of May 5th, 2020, I export these three variables in my bashrc:

export FINDSITELIB=/n/groups/drad/findsitemetal-1.0/dat/pdb_library
export FINDSITEMAP=/n/groups/drad/findsitemetal-1.0/dat/my_mapping.cls
export FINDSITEDAT=/n/groups/drad/findsitemetal-1.0/dat

Running findsite

The basic format for running findsite is:

/n/groups/drad/findsitemetal-1.0/bin/findsitemetal\
    -s ${location of pdb file}\
    -t ${location of templates list file}\
    -o ${location of output files. Note that this isn't a directory: so, if this argument is '/home/findsite/my_findsite', then files like '/home/findsite/my_findsite.sites.dat' will be produced"

For example:

`/n/groups/drad/findsitemetal-1.0/bin/findsitemetal -s /n/groups/drad/julian/Carbonylation_May2020/all_relevant_pdbs/Q9Y6U3_5A1K_A.pdb -t FINDSITE_input/Q9Y6U3.txt -o /home/js741/CarbonylationSite_Prediction/Human_Proteins/findsite/LOMETS_to_findsitemetal/test_findsite/Q9Y6U3_5A1K_A`

Findsite does run relatively quickly (less than a minute per run, generally).

To help you run findsite in high-throughput, I provided a script here: /n/groups/drad/julian/findsite/run_findsite.sh as an example, but you may need to write your own script (since that one isn't super generalizable). (todo: make a more generalizable version of this script).

Profit

Now you should have findsite output files. Really, we just need the *.findsitemetal.sites.dat file.

For more information about parsing that file, see RLC's notes in Findsitemetal_notes_rlc.docx. I'll paste them here for quicker reference.

RLC Notes

Findsitemetal notes:

-First had to run LOMETS to get template lists for every protein

-Then ran findsitemetal to predict metal binding sites based on structures and template lists

-Results

- Don't need data from "\*.findsitemetal.alignments.dat" right now

-"\*.findsitemetal.templates.pdb" all-atom coordinates of
    template structures aligned and superimposed on the target
    protein structure (same coordinate system as target)

-"\*.findsitemetal.metals.pdb" has Cartesian coordinates of metal
    atoms from template structures aligned and superimposed on
    target protein structure

-   "\*.findsitemetal.sites.pdb" has Cartesian coordinates of
    predicted metal atoms in their binding sites, but since I'm
    taking my own approach to this based on centroid of metal
    binding residues, I don't need this file

-   "\*.findsitemetal.sites.dat" is probably the only file really
    needed. From here we can parse out:

    -   The number of the binding site: "SITE 1 7 1.0000 2.2058
        0.6435 0.2115 4" second column

    -   The number of metal binding residues in the site: "SITE 1 7
        1.0000 2.2058 0.6435 0.2115 4" last column

    -   Which metal(s) bind to each site: "METAL MG 1.000000
        0.764762" third column equal to 1 means the site does bind
        this metal, otherwise 0

    -   Which residues make up the binding site: "RESIDUE 25 T \* 7
        4.205 0.35060 0.74038 0.85714 0.63671 0.05634" if asterisk
        follows third column, it is a binding residue, else not

    -   Residue position(s) making up the binding site: "RESIDUE 25
        T \* 7 4.205 0.35060 0.74038 0.85714 0.63671 0.05634" second
        column

    -   Amino acid type making up the binding site: "RESIDUE 25 T \*
        7 4.205 0.35060 0.74038 0.85714 0.63671 0.05634" third
        column

    -   Parsed result columns look like: col1 = site\#, col2 =
        METAL1,METAL2..., col3 = AA\#,AA\#\...

-What features to extract

-   Total number of metal-specific binding sites on the protein
    (this should have a column for each type of metal, and their sum
    as well, and will consist of integers)

-   Local metal ion contacts (should have a column for each type of
    metal, their sum, and will be integers)

    -   Approximate the locations of bound metal atoms as the
        centroid of all atoms (or if possible, just sidechain atoms)
        of all metal binding residues for each site; it is not known
        which specific amino acid atoms metals may form bonds with
        because it is difficult to know the partial charges on the
        protein, protonation states, etc., so an approximation is
        probably appropriate; this method leads to very slightly
        different centroids than those output by findsitemetal
        itself but is also extensible to metal binding sites from
        UniProt

    -   Use Bubble method (5-angstrom radius) and count the number
        of metal atoms for each type of metal within this radius
        (distance from centroid to potential CS atom \< \[5
        angstroms + atomic radius of metal\]) for a given RKPT
        residue

    -   Metal atomic radii (Chimera default ionic radii,
        coordination number = 6; from R. D. Shannon (1976).
        \"Revised effective ionic radii and systematic studies of
        interatomic distances in halides and chalcogenides\". Acta
        Crystallogr A. 32: 751--767.):

        -   CA2+ = 1.00 angstroms

        -   CO2+ = 0.65 angstroms

        -   CU2+ = 0.73 angstroms

        -   CU1+ = 0.77 angstroms

        -   FE2+ = 0.61 angstroms

        -   K1+ = 1.38 angstroms

        -   MG2+ = 0.72 angstroms

        -   MN2+ = 0.83 angstroms

        -   MO4+ = 0.65 angstroms

        -   NA1+ = 1.02 angstroms

        -   NI2+ = 0.69 angstroms

        -   ZN2+ = 0.74 angstroms

Decide whether to say findsitemetal "FE" can stand-in for UniProt "Iron sulfur" clusters

  • "TP" = 468 = Count number of overlapping residues that findsitemetal predicts bind FE and UniProt says bind an iron-sulfur cluster

  • "FN" = 249 = Count number of residues that UniProt says bind an iron-sulfur cluster but findsite metal does not predict bind FE

  • If TP >> FN, then when merging UniProt and findsitemetal sites, treat iron-sulfur clusters like they are just free FE, otherwise, these can't be merged and probably should toss out iron-sulfur cluster data from UniProt; TP >> FN can be evaluated by seeing if FE-ironSulfur(TP/FN) >= FE-FE(TP/FN)

    • FE-FE(TP) = 76

    • FE-FE(FN) = 32+41 = 73

  • FE-ironSulfur(TP/FN) = 468/249 = 1.88

  • FE-FE(TP/FN) = 76/73 = 1.04

  • This result suggests that if FE binding prediction by findsitemetal is considered to be predictive of iron-sulfur cluster binding sites, it performs substantially better at predicting FE binding overall, but since findsitemetal does not provide for the orientation of iron-sulfur clusters, to be conservative, I'll treat those binding sites as binding free FE. This means that I'll use the ionic radius of free FE2+ instead of the substantially larger size of the iron-sulfur cluster, and the position of the FE should be approximately the centroid of the position of the iron-sulfur cluster. RKPT will have to be closer to these FEs in order to count towards their Bubble feature for proximal metal binding than they would if the binding position of the iron-sulfur cluster was known.

Merging UniProt metal binding site data with findsitemetal predictions

  • If sites between UniProt and findsitemetal overlap at least partially in both residues and type of metal, merge them into one site by taking the union of residues and metals for both sites

  • If sites between UniProt and findsitemetal overlap only in partially in residues and do not overlap at all in metals, consider the two as distinct sites that bind different metals; very few (76) such cases in my protein structures

  • If sites between UniProt and findsitemetal overlap only in metals and not at all in residues, consider the two as distinct sites that bind the same type of metal but not the same actual ion simultaneously

  • After merging in this fashion, make sure to include all findsitemetal results for proteins not annotated with metal binding sites in UniProt and all UniProt binding sites that either don't overlap with any findsitemetal results or for proteins with no predicted findsitemetal results

  • De-duplicating

    • If there are entirely identical final binding sites because of multiple findsitemetal sites that use the same binding residues, remove the higher-numbered site (or one with more metals associated; or combine metals for otherwise identical sites)

    • If there are entirely identical final binding sites because of merging 2 or more uniprot sites with findsitemetal sites, de-merge them and keep only the uniprot sites as they were in uniprot; this is especially important because not doing so will reduce the count of real metal binding sites to be used in the Bubble method

← Running uniprotmetalSPPIDER Overview →