Sequence Alignments HEXA
Use of database searching tools
/bin/fasta36 seq.fasta /data/blast/nr/nr > fasta_out.txt
blastall -p blastp -d /data/blast/nr/nr -i mult_seq.fasta > blast_out.txt
blastpgn -i seq.fasta -j <#iterations> -h <e-value threshold> -d /data/blast/nr/nr > psiblast_out.txt
For the statistical analysis we wrote a script which shows the distribution of the E-Value and the identity as well as the different aligned sequences. Furthermore, we created a Venn diagram to present the overlap between the results of the different searching methods (with http://bioinfogp.cnb.csic.es/tools/venny/index.html). First, we compared the methods BLAST, FASTA and PsiBlast (PsiBlast with 3 and 5 runs and E-Value cutoff of 10E-6). Then we looked for the overlap of all done PsiBlast runs.
Overlap of the aligned sequences
As it can be seen on Figure 1, FASTA found a large number of matches which are not found by the other methods. By comparison, the number of hits which were not found by BLAST or PsiBlast as well, is about 1400. This is much higher than the number of sequences which is found by FASTA and BLAST together. This leads to the conclusion that FASTA aligns many more sequences which are probably worse or even wrong.
The both different PsiBlast variants deliver the same hits which are all found by FASTA, as well. Furthermore, all resulting BLAST-sequences were aligned by FASTA and the most of them also by PsiBlast.
Besides, we decided to compare different runs of PsiBlast. We compared PsiBlast with 3 iterations and two different E-Value Cutoffs (0.005 and 10E-6) and also two PsiBlast runs with 5 iterations and the same two E-Value cutoffs as before, which can be seen on Figure 2. The Venn diagram shows that the different results overlap mostly. Only a few ones differ from the other. This leads to the conclusion, that PsiBlast with different iteration numbers and E-value receives usually a similar result. In summary the BLAST-methods agree with each other. In contrast the FASTA-method delivers much more sequences which do not correspond to the other methods.
Distribution of the sequence identity and the e-value
The following plots show the distribution of the sequence identities and the E-values of all used methods. Both values (x-axis) and their frequencies (y-axis) were extracted from the corresponding output-files.
The first image (Figure 3) shows the distribution of the sequence identities. The first plot visualize the distribution for BLAST. Here could be seen that these identity-distribution is very balanced which means the low identities are approximate same common as the very high ones. It is also the same for the HHSearch-plot. Contrary, the FASTA-distribution has often a very high frequency for very small identities. This means that FASTA aligns many sequences although they have only a small sequence identity. This could explain why FASTA receives so many hits which do not agree with the other sequence searching tools (see Venn diagram).
The last four plots represent the corresponding distribution for the different PsiBlast runs which are very similar. This is another indication that PsiBlast received very similar results for different parameters. Their distribution is also very balanced. There are high frequencies for small, middle and high sequence identity values.
The second image (Figure 4) shows the distribution of the E-values. The E-value is a measurement for the probability that a hit is found by chance. Therefore, the smaller the E-value the better the alignment. All plots, except the one for FASTA, have high frequencies for small E-values whereby BLAST receives the smallest E-values. The E-values of FASTA have range from 0 to 8 where in contrast the other methods have no E-value higher than 1. Furthermore, the highest BLAST E-value is about 1e-29 which is still very low. In summary this shows again that BLAST delivers the best results and FASTA the worst ones.
True positive hits
HSSP (Homology-derived Secondary Structure of Proteins) lists proteins which are homologue and have a similar secondary structure. Therefore, we used the HSSP alignment to check our results. The overlapping sequences are the true positives. FASTA (Figure 5) has a greater overlap than the other methods (about 10 sequences more). The BLAST result (Figure 6) and the results of the Blast variants (PsiBlast runs Figure 7 - 10) show very similar overlap. Therefore, in this case FASTA gave the most true positive hits (although FASTA has also a hugh number of false positive predictions).
With the results of these analyses, we created our file for the multiple alignments.
|99%-90% Sequence Identity|
|89%-60% Sequence Identity|
|59%-40% Sequence Identity|
|867691|gb|AAA68620.1||55%||PsiBlast, 3 Iterations, E-Value Cutoff = 0.005|
|39%-20% Sequence Identity|
|299139410|ref|ZP_07032585.1||36%||PsiBlast, 3 Iterations, E-Value Cutoff = 0.005|
|166159759|gb|ABY83272.1||32%||PsiBlast, 5 Iterations, E-Value Cutoff = 0.005|
|212691177|ref|ZP_03299305.1||22%||PsiBlast, 3 Iterations, E-Value Cutoff = 10E-6|
Back to [Tay-Sachs Disease]
To use cobalt, we had to install the program on our virtual box first.
- Download Cobalt from [here] (ncbi-cobalt-2.0.1-x64-linux.tar).
- Uncompress the archive file with tar xfz ncbi-cobalt-2.0.1-x64-linux.tar and change directory to the uncompressed cobalt directory.
- Now you find an executable file called cobalt
Call: ./cobalt -i mult_seq.fasta -norps T > cobalt_out.aln
clustalw -infile=mult_seq.fasta > clustalW_out.aln
muscle -in mult_seq.fasta -out muscle_out.aln -clw
t_coffee -seq mult_seq.fasta
- T-Coffee (3D)
t_coffee -seq mult_seq.fasta -mode expresso
Next, we wanted to know if there is a strong conservation between the sequences in our multiple sequence alignment. We compared each position of the alignment to find and count the conserved columns. Furthermore, we counted also the number of gaps in our alignment.
|Alignment methods||Conserved Columns|
|Gaps||100% cons||>90% cons||>80% cons||>70% cons||>60% cons||>50% cons||>40% cons|
The different methods have a different amount of gaps in their alignments. They number of gaps differ between 346 (ClustalW) and 609 (T-Coffee). There is a clear trend by looking at the different conservation levels (100% conservation is not that frequent as lower conservation). Interestingly, there do not exist columns which have a conservation between 40% and 50%.
For the identification of gaps in secondary structure elements we wrote a script which compares the alignment sequence of hexasaminidase with the secondary structure sequence from the [PDB] visualized in Figure 11. The results for all multiple alignment tools are listed in the following table.
|Alignment methods||Gaps in Secondary Structure Elements|
|Sum of Gaps||Helix||Extended||Coil|
We found several functional residues in the [Uniprot] database:
|Disulfide bond||58 <-> 104|
|Disulfide bond||277 <-> 328|
|Disulfide bond||505 <-> 522|
Because these residues are functionally important, they should be conserved. We compared the different alignments and looked if these residues are conserved.
|residue position||Cobalt||ClustalW||Muscle||T-Coffee||3D T-Coffee||dominated substitution|
|E (active site)||323||conserved (21/21)||conserved (21/21)||conserved (21/21)||conserved (21/21)||conserved (21/21)||none|
|N (Glycolysation)||115||non-conserved (12/21)||non-conserved (13/21)||non-conserved (13/21)||non-conserved (13/21)||non-conserved (13/21)||Serine|
|N (Glycolysation)||157||non-conserved (16/21)||non-conserved (16/21)||non-conserved (16/21)||non-conserved (16/21)||non-conserved (16/21)||Proline and Serine|
|N (Glycolysation)||295||non-conserved (14/21)||non-conserved (14/21)||non-conserved (14/21)||non-conserved (14/21)||non-conserved (14/21)||Proline and Aspartic acid|
|C (Disulfide bond)||58 (connected with 104)||non-conserved (17/21)||non-conserved (15/21)||non-conserved (17/21)||non-conserved (17/21)||non-conserved (16/21)||no dominated substitution|
|C (Disulfide bond)||104 (connected with 58)||non-conserved (14/21)||non-conserved (14/21)||non-conserved (14/21)||non-conserved (14/21)||non-conserved (15/21)||no dominated substitution|
|C (Disulfide bond)||277 (connected with 328)||conserved (20/21)||non-conserved (16/21)||non-conserved (18/21)||non-conserved (19/21)||non-conserved (17/21)||Serine|
|C (Disulfide bond)||328 (connected with 277)||non-conserved (19/21)||non-conserved (19/21)||non-conserved (19/21)||non-conserved (19/21)||non-conserved (19/21)||Argenine and Glutamic acid|
|C (Disulfide bond)||505 (connected with 522)||non-conserved (17/21)||non-conserved (17/21)||non-conserved (16/21)||non-conserved (16/21)||non-conserved (16/21)||Tyrosine and Glutamine|
|C (Disulfide bond)||522 (connected with 505)||non-conserved (18/21)||non-conserved (16/21)||non-conserved (18/21)||non-conserved (18/21)||non-conserved (16/21)||no dominated substitution|
As you can seen in the table above, only the active site is completely conserved. But the other positions are also well conserved. Some substitutions, which could be found here, probably do not damage the protein that much.
Back to [Tay-Sachs Disease]