Difference between revisions of "Metachromatic leukodystrophy reference aminoacids"

From Bioinformatikpedia
(Muscle)
(ClustalW)
Line 212: Line 212:
 
=== ClustalW ===
 
=== ClustalW ===
   
ClustalW is a Multiple Alignment Tool that uses a guide tree to build a multiple sequence alignment.
+
ClustalW is a Multiple Alignment Tool that uses a guide tree to build a multiple sequence alignment. The guide tree is constructed using Neighbor-Joining.
 
<ref name="larkin2007">Larkin MA, Blackshields G, Brown NP, Chenna R, McGettigan PA, McWilliam H, Valentin F, Wallace IM, Wilm A, Lopez R, Thompson JD, Gibson TJ, Higgins DG.
 
<ref name="larkin2007">Larkin MA, Blackshields G, Brown NP, Chenna R, McGettigan PA, McWilliam H, Valentin F, Wallace IM, Wilm A, Lopez R, Thompson JD, Gibson TJ, Higgins DG.
 
(2007). Clustal W and Clustal X version 2.0. Bioinformatics, 23, 2947-2948. </ref>
 
(2007). Clustal W and Clustal X version 2.0. Bioinformatics, 23, 2947-2948. </ref>

Revision as of 17:54, 28 August 2011

Sequence

>sp|P15289|ARSA_HUMAN Arylsulfatase A OS=Homo sapiens GN=ARSA PE=1 SV=3
MGAPRSLLLALAAGLAVARPPNIVLIFADDLGYGDLGCYGHPSSTTPNLDQLAAGGLRFT
DFYVPVSLCTPSRAALLTGRLPVRMGMYPGVLVPSSRGGLPLEEVTVAEVLAARGYLTGM
AGKWHLGVGPEGAFLPPHQGFHRFLGIPYSHDQGPCQNLTCFPPATPCDGGCDQGLVPIP
LLANLSVEAQPPWLPGLEARYMAFAHDLMADAQRQDRPFFLYYASHHTHYPQFSGQSFAE
RSGRGPFGDSLMELDAAVGTLMTAIGDLGLLEETLVIFTADNGPETMRMSRGGCSGLLRC
GKGTTYEGGVREPALAFWPGHIAPGVTHELASSLDLLPTLAALAGAPLPNVTLDGFDLSP
LLLGTGKSPRQSLFFYPSYPDEVRGVFAVRTGKYKAHFFTQGSAHSDTTADPACHASSSL
TAHEPPLLYDLSKDPGENYNLLGGVAGATPEVLQALKQLQLLKAQLDAAVTFGPSQVARG
EDPALQICCHPGCTPRPACCHCPDPHA


Source


Database Searches

FASTA, BLAST and PSI-BLAST were run against the non-redundant database (NR). HHsearch was run through the web interface<ref>http://toolkit.lmb.uni-muenchen.de/hhpred</ref> against the PDB and Interpro database. The following parameter settings were used:

  • BLAST: blastall -p blastp -i refSeq.fasta -d /data/blast/nr/nr > blastp with refSeq.fasta being the file containing the reference sequence.
  • PSI-BLAST was run with the following parameter settings:
    • e-value cutoff 0.005, 3 iterations (Psi-blast1)
    • e-value cutoff 0.005, 5 iterations (Psi-blast2)
    • e-value cutoff 10E-6, 3 iterations (Psi-blast3)
    • e-value cutoff 10E-6, 5 iterations (Psi-blast4)
    • blastpgp -i refSeq.fasta -d /data/blast/nr/nr -e"e-value" -j "#iterations" > psiblast_"e-value"_"#iterations"
  • HHsearch: We used the online version of hhPred <ref>http://toolkit.lmb.uni-muenchen.de/hhpred</ref> with default parameters. One search was performed against PDB and one against Interpro.

Alignment results

We wrote a perl script to parse the output files of the individual programs and extracted identifier, alignment score and the percentage of identical residues within the alignment.

Mapping of identifier

The non-redundant database contains entries from various databases, including RefSeq, PDB, PIR, PRF, GenBank and Swiss-Prot. In order to compare results of NR database searches with the results of the HHpred searches, a mapping of the IDs is necessary. Furthermore, the entries in HSSP - which is used later to benchmark the alignment results - contains only references to the UniProtKB accession number (ACCNUM). To overcome this problem we downloaded a mapping table between the IDs from <ref>http://pir.georgetown.edu/pirwww/search/idmapping.shtml</ref>. This table was used - together with some short perl scripts - to map IDs between the databases and compare the results.

Summary of database searches

In this section, we give a short summary description of the search results of the individual programs and the compare them to each other.

Comparison of the methods

  • FASTA yielded with 4733 alignments the highest number of hits.
  • BLAST produced 252 alignments.
  • PSI-BLAST
    • Using an E-value cutoff of 0.005, PSI-BLAST produced 756 alignments for 3 iterations and 1257 for 5 iterations.
    • Using an E-value cutoff of 10E-6, PSI-BLAST produced 756 alignments for 3 iterations and 1257 for 5 iterations.
  • HHsearch produced 33 alignments for the search against PDB and 74 alignments for search against Interpro.

FASTA shows the highest number of alignments, probably due to the fact, that no e-value cutoff was chosen. Contrary, hhsearch has very few alignments. This could be ascribed to the fact that completely different databases were used for the alignments and Interpro and pdb just did not have as much homolguous sequences as the nr database. This is also supported by the benchmark with HSSP (see next section). Another interesting fact is that the results of PSI-BLAST depended for our parameter setting only on the number of iterations. Ragarding the results for the number of iterations, both e-value cutoffs yielded - except of some single exceptions the same aligned target sequences from the database - the same hits.

Overlap between methods
The number of shared target sequences between two methods is shown in the upper panel (self overlaps not shown). The lower panel depicts how many percent of the aligned target sequences of a given method (x-axis) are shared with the other methods.

The results of the four different PSI-BLAST runs show the highest overlap. The additional iterations find more related sequences and yield a higher number of alignments. Interestingly, almost all BLAST hits overlap with the PSI-BLAST and FASTA results. The overlaps of the searches show that the number of hits highly depends on the database. hhpred1 does not have a good overlap with any of the other methods, but hhpred2 shares a significant part of its results with PSI-BLAST.

Scores and identity of aligned residues

In general, the identity of aligned residues is very low, i.e. only very few highly similar hits are detected by the methods. These high scoring matches mainly represent homologs of Arylsulfatase A. This can be seen in the table of sequences which were chosen for the multiple sequence alignment. The majority of hits from an identity range of 99-90% and 89-60% are annotated as Arylsulfatases or Arylsulfatase A. Lowering the identity cutoff yields an increasing number of Sulfatases, which might be more distantly related to our query sequence.
The alignment scores show a very similar distribution. Lowest scores are produced by FASTA, which reflects the low sensitivity of the method for the detection of true homologs. A lot of these matches might be classified as false positive, i.e. they are not evolutionary related to the query sequence. The BLAST scores are a bit elevated compared to FASTA. The highest scores are derived from the PSI-BLAST searches and an increase can be seen when the number of iterations is raised.

The density of alignment scores and percentage of identical residues within the alignments are plotted
HSSP

HSSP is a database which contains information about homologuous protein sequences for each protein in the PDB database, which are very likely to have a similar structure as the query. HSSP uses a position-weighted dynamic programming method for sequence profile alignment of PDB entries against the Swissprot database. Therefore the database can also be used to infer homology based secondary structure predictions.
Now we want to use the homology information in HSSP to benchmark our alignment results. The table below depicts the recall and precision of homologs in HSSP of our query by our alignments. FASTA shows the highest recall (92 %), which could lead to the misleading interpretation, that this method performs best. The high value can be ascribed to the fact, that FASTA reports a very high number of alignments - with a lot of false positive - and a selection of true homologs from this results without prior knowledge is quite challenging. This is reflected by the precision of the method, which is only 23 %, i.e. only 23 % of the hits in fasta are true homologs regarding the HSSP annotation.
BLAST perfoms a little bit better than FASTA and PSI-BLAST perfoms best. The latter shows a recall of around 20 % - which is still quite low - but a precision of 65 %, i.e. 64 % of the hits using the PSI-BLAST method are true homologs.
As hhpred was searched against other databases, which contained much less entries than the NR-database used for the other methods, it is not directly comparable to the other results. It performs much worse when hits are compared to all HSSP homologs. But if we only take HSSP homologs, that are contained in the PDB hhpred perfoms best. It even recalls all 12 HSSP "pdb-homologs".

Method Recall (GI) Recall (pdb) Precision (GI)
FASTA 0.92 0.67 0.23
BLAST 0.11 0.42 0.54
Psi-blast1 0.21 0.42 0.65
Psi-blast2 0.23 0.5 0.62
Psi-blast3 0.21 0.42 0.65
Psi-blast4 0.23 0.5 0.62
hhpred (pdb) 0.01 1 0.11
hhpred (interpro) 0.01 0.92 0.12

Multiple Alignments

For building the multiple Alignments the following sequences were chosen:

SeqIdentifier Seq Identity source Protein function
99-90% Sequence Identity
gi109094666 96.6% Macaca mulatta Arylsulfatase A isoform 2
gi281339526 90.8% Ailuropoda melanoleuca unknown
gi47522624 91.5% Sus scrofa Arylsulfatase A
gi149759319 89.7% Equus caballus Arylsulfatase A
gi301763795 90.8% Ailuropoda melanoleuca Arylsulfatase A
89-60% Sequence Identity
115497982 87.3% Bos taurus Arylsulfatase A precursor
gi118081865 63.4% Gallus gallus Arylsulfatase A
gi126339031 74.3% Monodelphis domestica Arylsulfatase A
gi114326188 88.4% Canis lupus familiaris Arylsulfatase A
gi164519052 85.6% Rattus norvegicus Arylsulfatase A
59-40% Sequence Identity
gi223936859 43.9% Bacterium Ellin514 Sulfatase
1P49 39.0% Homo Sapiens Steryl-Sulfatase
gi120537984 56.0% Xenopus laevis unknown
gi301625378 55.5% Xenopus (Silurana) tropicalis Arylsulfatase A
gi86142609 40.0% Leeuwenhoekiella blandensis MED217 Arylsulfatase A
39-20% Sequence Identity
1FSU 28.0% Homo Sapiens N-Acetylgalactosamine-4-Sulfatase
2VQR 20.0% Rhizobium leguminosarum Sulfatase
3ED4 32.0% Escherichia coli Arylsulfatase
gi113971721 29.0% Shewanella sp. MR-4 Sulfatase
gi310635680 36.0% Planctomyces brasiliensis DSM 5305 Sulfatase


The sequences with <20% and >99% sequence identitiy were ignored and 5 samples were randomly picked from the other ranges. So 20 sequences were available for the multiple alignments. Unfortunately no sequences in the range between 99-90% with known 3D-structure were found, so only sequences without known structure were used here. For the range between 59-40% also no pdb-structure was found, so we used a sequence with 39% sequence identity to have at least one pdb-structure.

Cobalt

Cobalt is a progressive Multiple Alignment Tool which uses a collection of pairwise constraints which are derived from different databases. To reduce computation time, sequence clusters are formed and then only one sequence out of the cluster is used for finding conserverd domains and motif matches. <ref name="papadopoulos2007">Papadopoulos JS and Agarwala R (2007) COBALT: constraint-based alignment tool for multiple protein sequences, Bioinformatics 23:1073-79</ref>

Command

time /home/student/Downloads/ncbi-cobalt-2.0.1/cobalt -i MSA_seqs.fasta -norps T > alignments/MSA_cobalt.aln

time
real 0m6.691s
user 0m3.550s
sys 0m0.240s

Gaps in secondary structure & JalView

View of Alignment built by Cobalt, shown in Jalview
Position in reference
HELIX 39, 70, 210, 290, 291, 451
STRAND 29, 154, 190, 242, 244, 327, 387, 427
TURN 163, 320

ClustalW

ClustalW is a Multiple Alignment Tool that uses a guide tree to build a multiple sequence alignment. The guide tree is constructed using Neighbor-Joining. <ref name="larkin2007">Larkin MA, Blackshields G, Brown NP, Chenna R, McGettigan PA, McWilliam H, Valentin F, Wallace IM, Wilm A, Lopez R, Thompson JD, Gibson TJ, Higgins DG. (2007). Clustal W and Clustal X version 2.0. Bioinformatics, 23, 2947-2948. </ref>

Command

time clustalw

time
real 3m28.533s
user 0m9.500s
sys 0m0.110s


Gaps in secondary structure & JalView

View of Alignment built by ClustalW, shown in Jalview
Position in reference
HELIX 39, 70, 290
STRAND 190, 325
TURN 240, 321

Muscle

Muscle is a Multiple Alignment Tool that builds a guide tree by an approximation algorithm. Then the sequences are progressively aligned. The alignment is then refined using the calculated pairwise distances. As final step a tree-dependent refinement takes place where the tree is partitioned in two partitions and the profiles for the two subsets are again aligned. <ref name="edgar2004">Edgar, R.C. (2004) MUSCLE: multiple sequence alignment with high accuracy and high throughput.Nucleic Acids Res. 32(5):1792-1797</ref>

Command

time muscle -in MSA_seqs.fasta -out MSA_muscle.aln

time
real 0m4.236s
user 0m2.550s
sys 0m0.090s

Gaps in secondary structure & JalView

View of Alignment built by muscle, shown in Jalview
Position in reference
HELIX 39, 70, 132, 139, 197, 246, 291
STRAND 122, 154, 181, 189, 243, 244, 327, 375, 387, 426
TURN 164, 320

T-Coffee

standard parameters

Command

time t_coffee MSA_seqs.fasta

time
real 1m20.451s
user 0m57.410s
sys 0m1.580s
Gaps in secondary structure & JalView
View of Alignment built by T-Coffee with standard parameters, shown in Jalview
Position in reference
HELIX 69, 131, 139, 197, 288, 289, 414, 453
STRAND 153, 160, 188, 189, 191, 327, 374, 426
TURN 236, 239, 241, 384

3d-coffee

Command

time t_coffee MSA_seqs.fasta -mode expresso -pdb_type dn

time
real 21m38.094s
user 11m27.690s
sys 1m32.880s
Gaps in secondary structure & JalView
View of Alignment built by T-Coffee with 3D-coffee, shown in Jalview
Position in reference
HELIX 69, 131, 137, 194, 205, 211, 246, 290, 450, 452, 453, 459, 466
STRAND 66, 183, 187, 189, 190, 231, 242, 317, 324, 325, 373, 387
TURN 241, 321, 496, 499


Summary

Gaps

For each method used in this section, the number of gapped columns (a column is defined as gapped, if it contains at least one gap), the number of conserved columns (all entries in the row have the same amino acid) and the alignment length were calculated.

Cobalt Muscle ClustalW T-Coffee T-Coffee 3D
#gapped columns 415 411 286 523 600
#conserved columns 24 26 22 27 25
length(alignment) 715 753 668 862 896

ClustalW yields the most compact, i.e. shortest multiple sequence alignment (MSA). It exhibits the lowermost overal number of gaps and number of gaps within the secondary structure of the human Arylsulfatase A. The number of 100% conserved columns consequently decreases a bit in comparison to the other methods. Further on the JalView representation shows the most homogenuous alignment of conserved blocks.

Conservation of active sites

As defined before, we only call a Position conserved if has the same amino acid in all sequences.

Cobalt Muscle ClustalW T-Coffee T-Coffee 3D
active site (Pos. 125) 20 / 21 1 / 21 20 / 21 20 / 21 20 / 21
Metal binding site (Pos. 29) 21 / 21 3 / 21 21 / 21 21 / 21 21 / 21
Metal binding site (Pos. 30) 21 / 21 1 / 21 20 / 21 20 / 21 20 / 21
Metal binding site (Pos. 69) 16 / 21 13 / 21 16 / 21 16 / 21 16 / 21
Metal binding site (Pos. 281) 21 / 21 1 / 21 21 / 21 21 / 21 21 / 21
Metal binding site (Pos. 282) 19 / 21 1 / 21 19 / 21 19 / 21 19 / 21
Binding site (Pos. 123) 20 / 21 1 / 21 20 / 21 20 / 21 20 / 21
Binding site (Pos. 150) 17 / 21 1 / 21 16 / 21 17 / 21 16 / 21
Binding site (Pos. 229) 20 / 21 1 / 21 21 / 21 21 / 21 21 / 21
Binding site (Pos. 302) 21 / 21 2 ( 21 21 / 21 21 / 21 20 / 21

As shown in the table above, by only looking at important sites, the alignments are very similar, except the muscle-alignment. So the important sites are really well-conserved also in more distant related proteins.

Computation Time

The computation time differed significantelly between the different programs. 3d-coffee took the most time (as expected) while Cobalt and Muscle where the fastest programs on our testset.

Plot of computation times by differenet MSA-programs on ARSA

References

<references />