Difference between revisions of "Task 2 (MSUD)"

From Bioinformatikpedia
(→‎Discussion)
(→‎Discussion)
Line 43: Line 43:
   
 
=== Discussion ===
  
=== Discussion ===
−
*E-value distribution:
+
* E-value distribution:
−
** Very few hits were found with very low E-values -> hits with high statistical significance
 +
** Very few hits were found with very low E-values. These hits show high statistical significance.
−
** Because different databases were used for BLAST/PSI-BLAST and HHBLits, hhblits has found hits with larger range of e-value -> higher density for hhblits at high E-value
 +
** Because that different databases were used for BLAST/PSI-BLAST and HHBLits, hhblits has a set of hits with larger range of e-value.
−
*** E-value distribution of PSI-BLAST shift to low E-value side with more iterations -> better search result?
 +
*** E-value distribution of PSI-BLAST shift to low E-value side with more iterations. Although the hits are statistically more significant, but the biological significance should be tested. If more iterations were used the shift could be even larger, so the overlap between statistical hits and biological homologs must be evaluated. A proper number of iterations should be selected.
  +
 
* Identity distribution
  
* Identity distribution
−
** Results show that BLAST depend mostly on sequence identity -> possible lose of patterns with low sequence identity but high biological similarity
 +
** Results show that BLAST depends mostly on sequence identity. Homologs with low sequence identity but high biological similarity could be lost.
  +
 
* Intersection of hits
  
* Intersection of hits
  +
** PSI-BLAST with 2 iterations has bigger intersection with BLAST.
−
** HHBlits was not comparable to other methods due to different sequence database
  
−
** PSI-BLAST with 2 iterations has bigger intersection with BLAST
 +
** Two PSI-BLAST run with 2 iterations and different E-value cutoffs have very similar set of hits.
−
** two PSI-BLAST run with 2 iterations and different E-value cutoffs have very similar set of hits
 +
** PSI-BLAST with 10 iterations has smaller intersection with BLAST.
  +
** Two PSI-BLAST runs with 10 iterations and different E-value cutoffs share the fewest common hits. The explanation could be, the E-value cutoff may have higher influence than the number of iterations.
−
** PSI-BLAST with 10 iterations has less intersection with BLAST
  
  +
−
** two PSI-BLAST run with 10 iterations and different E-value cutoffs share the fewest common hits -> E-value cutoff may have higher influence after more iterations
  
 
* SCOP of hit sequences
  
* SCOP of hit sequences
  +
** Both BLAST and PSI-BLAST find the right fold class for BCKDHA.
−
** PDB sequence required -> no evaluation for HHBlits
  
  +
** PSI-BLAST finds more hits in the fold class that describes the query protein best. Most hits have c.36 which is for Thiamin diphosphate-binding fold. This fold classification is just the main binding function of BCKDHA.
−
** Both BLAST and PSI-BLAST find the right fold class for query protein
  
−
** PSI-BLAST generally find more hits in the fold class that describes the query protein best (e.g. DLD protein, c.3 is FAD/NAD(P)-binding domain)
 +
** PSI-BLAST also find hits in more fold classes which may describe biological similarities of domains and motives between hits and query protein.
  +
−
** PSI-BLAST also find hits in more fold classes which may describe domains of query protein
  
 
* Gene Ontology of hit proteins
  
* Gene Ontology of hit proteins
 
** Top-5 GO terms in hits of PSI-BLAST with different iterations are more conserved. They also have similar ranking of frequency.
 
** Top-5 GO terms in hits of PSI-BLAST with different iterations are more conserved. They also have similar ranking of frequency.
−
** PSI-BLAST finds out hits with more GO terms -> It may be more sensitive to functional patterns in sequence
 +
** PSI-BLAST finds out hits with more GO terms. It may be more sensitive to functional patterns in sequence.
   
 
== Multiple sequence alignments ==
  
== Multiple sequence alignments ==

Revision as of 17:15, 9 August 2013

Sequence searches

Lab journal

Results

We have performed sequence search experiments for all of the 4 subunits of BCKDC. In this page, we mainly describe and discuss the results for the subunit BCKDHA. Results and discussions for other 3 subunits are covered in this page: Additional Results.

Old version

The query sequences for the 4 subunits of BCKDC locate at /mnt/home/student/weish/master-practical-2013/task01/.

Results for sequence search locate in the directory /mnt/home/student/weish/master-practical-2013/task02/01-seq-search/results. For BLAST and PSI-BLAST, statistics (such as E-value, probability and identity) are stored in *.tsv files. Detailed results are shown in xml files. For HHBlits, the *.hhr files contain information about statistics and hits.

Distributions of E-value and sequence identity

  • E-value and identity distribution for different sequence search methods

  • E-value distribution of sequence search methods. (Query sequence is RefSeq of BCKDHA)

  • Indentity distribution of sequence search methods. (Query sequence is RefSeq of BCKDHA)

Intersection of hits

  • Intersection of hits between different sequence search methods

  • Relative intersection of hits between BLAST and other sequence search methods.

  • Relative intersection between PSI-BLAST(iter. 2, E-value 0.002) and other sequence search methods.

  • Relative intersection between PSI-BLAST(iter. 2, E-value 10e-10) and other sequence search methods.

  • Relative intersection between PSI-BLAST(iter. 10, E-value 0.002) and other sequence search methods.

  • Relative intersection between PSI-BLAST(iter. 10, E-value 10e-10) and other sequence search methods.

  • Relative intersection between HHBlits and other sequence search methods.

Evaluation through structure and function

  • Distribution of SCOP folds

  • Distribution of SCOP fold in BLAST hits(only one classified PDB structure was found in SCOP)

  • Distribution of SCOP fold in hits of PSI-BLAST(iter. 2, E-value 0.002)

  • Top-5 common GO terms in hits with GO annotation

  • Top-5 common GO terms in BLAST hits

  • Top-5 common GO terms in hits of PSI-BLAST(iter. 2, E-value 0.002)

  • Top-5 common GO terms in hits of PSI-BLAST(iter. 10, E-value 10e-10)

Discussion

  • E-value distribution:
    • Very few hits were found with very low E-values. These hits show high statistical significance.
    • Because that different databases were used for BLAST/PSI-BLAST and HHBLits, hhblits has a set of hits with larger range of e-value.
      • E-value distribution of PSI-BLAST shift to low E-value side with more iterations. Although the hits are statistically more significant, but the biological significance should be tested. If more iterations were used the shift could be even larger, so the overlap between statistical hits and biological homologs must be evaluated. A proper number of iterations should be selected.
  • Identity distribution
    • Results show that BLAST depends mostly on sequence identity. Homologs with low sequence identity but high biological similarity could be lost.
  • Intersection of hits
    • PSI-BLAST with 2 iterations has bigger intersection with BLAST.
    • Two PSI-BLAST run with 2 iterations and different E-value cutoffs have very similar set of hits.
    • PSI-BLAST with 10 iterations has smaller intersection with BLAST.
    • Two PSI-BLAST runs with 10 iterations and different E-value cutoffs share the fewest common hits. The explanation could be, the E-value cutoff may have higher influence than the number of iterations.
  • SCOP of hit sequences
    • Both BLAST and PSI-BLAST find the right fold class for BCKDHA.
    • PSI-BLAST finds more hits in the fold class that describes the query protein best. Most hits have c.36 which is for Thiamin diphosphate-binding fold. This fold classification is just the main binding function of BCKDHA.
    • PSI-BLAST also find hits in more fold classes which may describe biological similarities of domains and motives between hits and query protein.
  • Gene Ontology of hit proteins
    • Top-5 GO terms in hits of PSI-BLAST with different iterations are more conserved. They also have similar ranking of frequency.
    • PSI-BLAST finds out hits with more GO terms. It may be more sensitive to functional patterns in sequence.

Multiple sequence alignments

Lab journal

Results

In the following sections the MSAs, visualised with Jalview, are shown.

BCKDHA

Low sequence identity

Mafft: MSUD BCKDHA low seq ident mafft.png

Muscle: MSUD BCKDHA low seq ident muscle.png

T-Coffee: MSUD BCKDHA low seq ident tcoffee.png

Espresso: MSUD BCKDHA low seq ident espresso.png

High sequence identity

Mafft: MSUD BCKDHA high seq ident mafft.png

Muscle: MSUD BCKDHA high seq ident muscle.png

T-Coffee: MSUD BCKDHA high seq ident tcoffee.png

Espresso: MSUD BCKDHA high seq ident espresso.png

Whole range sequence identity

Mafft: MSUD BCKDHA whole range seq ident mafft.png

Muscle: MSUD BCKDHA whole range seq ident muscle.png

T-Coffee: MSUD BCKDHA whole range seq ident tcoffee.png

Espresso: MSUD BCKDHA whole range seq ident espresso.png

BCKDHB

Low sequence identity

Mafft: MSUD BCKDHB low seq ident mafft.png

Muscle: MSUD BCKDHB low seq ident muscle.png

T-Coffee: MSUD BCKDHB low seq ident tcoffee.png

Espresso: MSUD BCKDHB low seq ident espresso.png

High sequence identity

Mafft: MSUD BCKDHB high seq ident mafft.png

Muscle: MSUD BCKDHB high seq ident muscle.png

T-Coffee: MSUD BCKDHB high seq ident tcoffee.png

Espresso: MSUD BCKDHB high seq ident espresso.png

Whole range sequence identity

Mafft: MSUD BCKDHB whole range seq ident mafft.png

Muscle: MSUD BCKDHB whole range seq ident muscle.png

T-Coffee: MSUD BCKDHB whole range seq ident tcoffee.png

Espresso: MSUD BCKDHB whole range seq ident espresso.png

DBT

Low sequence identity

Mafft: MSUD DBT low seq ident mafft.png

Muscle: MSUD DBT low seq ident muscle.png

T-Coffee: MSUD DBT low seq ident tcoffee.png

High sequence identity

Mafft: MSUD DBT high seq ident mafft.png

Muscle: MSUD DBT high seq ident muscle.png

T-Coffee: MSUD DBT high seq ident tcoffee.png

Whole range sequence identity

Mafft: MSUD DBT whole range seq ident mafft.png

Muscle: MSUD DBT whole range seq ident muscle.png

T-Coffee: MSUD DBT whole range seq ident tcoffee.png

DLD

Low sequence identity

Mafft: MSUD DLD low seq ident mafft.png

Muscle: MSUD DLD low seq ident muscle.png

T-Coffee: MSUD DLD low seq ident tcoffee.png

High sequence identity

Mafft: MSUD DLD high seq ident mafft.png

Muscle: MSUD DLD high seq ident muscle.png

T-Coffee: MSUD DLD high seq ident tcoffee.png

Whole range sequence identity

Mafft: MSUD DLD whole range seq ident mafft.png

Muscle: MSUD DLD whole range seq ident muscle.png

T-Coffee: MSUD DLD whole range seq ident tcoffee.png

Discussion

For the datasets with high sequence identity the three MSA programs Mafft, Muscle and T-Coffee come to similar results and find almost the same conserved blocks. Sometimes T-Coffee arranges gaps differently than the others and so does not find as much conserved columns. Especially at the ends of the sequences, the results of the programs differ a little. This is due to different scoring schemes that are used in the programs.

For low sequence identity, the programs have problems to find the right alignment. They do not agree in the position of gaps and also sometimes find different conserved columns. They do not cope with low similarity and so one cannot really rely on these results. Here structural information, as it is used in Espresso (which belongs to T-Coffee), can help to find the right alignment: Espresso can align more residues than T-Coffee.

For whole range sequence identity the results are similar w. r. t. many and different gaps at the ends of the sequences, but the programs agree more in the conserved columns that they find.

The results of Muscle and Mafft seem more similar to each other than to those of T-Coffee. T-Coffee often treats the ends of the sequences, which have low sequence identity, differently than the others. It is striking that almost always the alignment of Muscle has the shortest length, especially in cases with low sequence identity. If an alignment is very long, this means there are many gaps and less aligned residues, this might be a sign of bad alignment quality.

Altogether, there appear regions with many conserved columns and those with many gaps. The conserved blocks or columns correspond to secondary structure elements and functionally important residues, respectively. Gaps in the alignment appear in regions where there are loops in the structure of the protein, so that insertions or deletions that occur during evolution do not alter the overall structure or function of the protein.

As criteria for a good alignment one could run different alignment algorithms like in this task and compare the results. If one of them finds more conserved columns, this might be better than another. Different programs can be better than others if different datasets are used, so it is always a good idea to try more than one algorithm and pick out the best result. Mafft is often a good choice because it generated relatively precise results but still is very fast.

Retrieved from "https://i12r-studfilesrv.informatik.tu-muenchen.de/wiki/index.php?title=Task_2_(MSUD)&oldid=34460"