Lab Journal of Task 3 (MSUD)

For task 3 we have used the reference sequence of BCKDHA and other given example proteins.

Secondary structure

Prediction and assignment

• PSSMs were created with Psi-Blast:

blastpgp -d /mnt/project/pracstrucfunc13/data/big/big_80 -i P10775.fasta -j 2 -h 10e-10 -Q P10775_big80.blastPsiMat

blastpgp -d /mnt/project/pracstrucfunc13/data/swissprot/uniprot_sprot -i P10775.fasta -j 2 -h 10e-10 -Q P10775_SwissProt.blastPsiMat

• ReProf was run for P10775 with a simple fasta file and with a PSSM (generated with big_80 and SwissProt, respectively) as input:

reprof -i P10775.fasta

reprof -i P10775_big80.blastPsiMat

reprof -i P10775_SwissProt.blastPsiMat

• PsiPred and DSSP were run on the following servers: PsiPred, DSSP_server.

• The PDB files used as input for DSSP are located at /mnt/home/student/schillerl/MasterPractical/task3/pdb_structures/. If there were more than one PDB structures available those which the highest coverage over the sequence and with the highest resolution were taken preferentially. These structures were used: 2BFD (P12694), 2BNH (P10775), 1AUI (Q08209) and 1KR4 (Q9X0E6).

• To parse the output of ReProf and PsiPred, we used Sonja's script. For parsing the DSSP output, we used the following Python script (which is located at /mnt/home/student/schillerl/MasterPractical/task3/dssp/parse_dssp_output.py):

<source lang=python> Parse DSSP output to a sequence of H, E and L (secondary structure). If there is more than one sequence in the DSSP output, only the first sequence is parsed.

Call with python parse_dssp_putput.py <dssp output file> <sequence length> <output file>

@author: Laura Schiller

import sys

seq_len = int(sys.argv[2]) dssp_file = open(sys.argv[1]) out_file = open(sys.argv[3], "w")

1. translation of DSSP secondary structure to H, E and L

secstr = { 'H': "H",

           'G': "H", 
           'I': "H", 
           'E': "E", 
           'B': "E", 
           'S': "L", 
           'T': "L", 
           ' ': "L" }

line = dssp_file.readline() while not line.startswith(" # RESIDUE"):

   line = dssp_file.readline()

line = dssp_file.readline()

current_position = 1 while (line):

   try:
       position = int(line[7:10])
   except ValueError:
       if(line[13:15] == "!*"): # new amino acid chain
           break
       else:
           line = dssp_file.readline()
           continue
   structure = secstr[line[16:17]]
   if(current_position != position):
       for i in range(position - current_position):
           out_file.write("-") # no DSSP output for that residue
           current_position += 1
   out_file.write(structure)
   current_position += 1    
   line = dssp_file.readline()

for i in range(seq_len - current_position + 1):

   out_file.write("-")

dssp_file.close() out_file.close() </source>

Note: for the proteins P12694 and Q9X0E6 the residue numbering in the DSSP output did not agree with the fasta sequences, so the output of the above script for these proteins was manually modified to fit the fasta sequences (for P12694 the beginning of the sequence was missing in the DSSP output, so there was added sufficiently many '-' for this part; for Q9X0E6 there were some additional residues in the beginning and end in the DSSP output, which are missing in the fasta sequence, so these parts were deleted). The reason for these discrepancies may be, that the PDB files that were used as input for DSSP don't exactly match the fasta sequences from UniProt.

Evaluation of prediction approaches

• The ReProf predictions were compared with the DSSP assignment with the following Python script (located at /mnt/home/student/schillerl/MasterPractical/task3/evaluate_secstr_reprof.py), which calculates the recall, precision and f-measure of the predictions. Positions that that lack a DSSP assignment (parsed as '-' by the above script) were ignored for the calculation.

Recall and Precision are defined as follows:

• recall = TP / (TP + FN)

• precision = TP / (TP + FP)

• f-measure = 2 * recall * precision / (recall + precision)

where TP means true positive, FP false positive and FN false negative.

<source lang="python"> dssp_file = open("./dssp/P10775_secstr.txt") dssp = dssp_file.readline() dssp_file.close()

for reprof_run in ["./reprof/P10775_secstr.txt", "./reprof/P10775_big80_secstr.txt", "./reprof/P10775_SwissProt_secstr.txt"]: reprof_file = open(reprof_run) reprof = reprof_file.readline() reprof_file.close()

assert len(dssp) == len(reprof)

sum1 = {'E': 0, 'H': 0, 'L': 0, 'all': 0} sum2 = {'E': 0, 'H': 0, 'L': 0, 'all': 0} found = {'E': 0, 'H': 0, 'L': 0, 'all': 0} right = {'E': 0, 'H': 0, 'L': 0, 'all': 0}

for i in range(len(dssp)): for secstr in ['E', 'H', 'L']: if dssp[i] == secstr: sum1[secstr] += 1 if reprof[i] == secstr: found[secstr] += 1 if reprof[i] == secstr: sum2[secstr] += 1 if dssp[i] == secstr: right[secstr] += 1

for sum in [sum1, sum2, found, right]: sum['all'] = sum['E'] + sum['H'] + sum['L']

recall = {'E': 0.0, 'H': 0.0, 'L': 0.0, 'all': 0} precision = {'E': 0.0, 'H': 0.0, 'L': 0.0, 'all': 0}

print "-----------" print "%s:" % reprof_run print "-----------"

for secstr in ['H', 'E', 'L', 'all']: recall[secstr] = (float(found[secstr]) / sum1[secstr]) print "Recall for %s: %f" % (secstr, recall[secstr]) precision[secstr] = (float(right[secstr]) / sum2[secstr]) print "Precision for %s: %f" % (secstr, precision[secstr]) print "F-measure for %s: %f" % (secstr, (2 * precision[secstr] * recall[secstr] / (precision[secstr] + recall[secstr]))) </source>

Disordered protein

IUPred

• Predictions were performed through the web server of IUPred. Graphical profiles of the results were downloaded.
• Output of IUPred are stored in the directory /mnt/home/student/weish/master-practical-2013/task03/02-disordered-protein/iupred
• We have also performed the prediction from command-line, following is the bash script:

<source lang="bash">

1. !/bin/sh -e

INPUT=$HOME/master-practical-2013/task03 OUTPUT=$HOME/master-practical-2013/task03/02-disordered-protein/iupred PARAMS="long short glob"

if [ ! -d $OUTPUT ]; then

       mkdir $OUTPUT

fi

for seq in $INPUT/*.fasta do

       filename=`basename $seq`
       for param in $PARAMS
       do
               iupred $seq $param > $OUTPUT/iupred_${filename}_$param.tsv
       done

done </source>

MetaDisorder(MD)

• As the man page of metadisorder describes, the prediction of disordered region is based on the results of other programs such as NORSnet, PROFbval etc. Rather than directly call metadisorder we have used the wrapper program predictprotein as is described on the exercise page.
• Comparison to DisProt database: TODO

Following script was called for the task: <source lang="bash">

1. !/bin/sh -e

INPUT=$HOME/master-practical-2013/task03 OUTPUT=$HOME/master-practical-2013/task03/metadisorder EXE=predictprotein

1. make output directory

if [ ! -d $OUTPUT ]; then

       mkdir $OUTPUT

fi

1. call metadisorder for all query sequences

for seq in $INPUT/*.fasta do

       filename=`basename $seq`
       $EXE --seqfile $seq --target metadisorder -p metadisorder_$filename \
       -o $OUTPUT

done echo Done! </source>