What is local maximum in sequence alignments ?

Question

[TJay]

Hi,

I noticed in many research papers they said that Clustal package has a problem of local maximum. And also local alignment,progressive alignments also faced that. So can any one explain what is actually local maximum means?

Answer 1

[Jean-Karim Heriche]

When optimizing a cost function or score, a local optimum (maximum or minimum) is a solution which is the best amongst several nearby solutions in the parameter space. It is called local because it is the best solution in a local region of the parameter space. This is different from the global optimum which is the best solution of all possible ones over the whole parameter space. Many algorithms are only guaranteed to find a local optimum but depending on the data, the solution they find can be the global optimum. In many cases, a local optimum is also a good enough solution for practical purposes. In the case of Clustal, this means the algorithm is guaranteed to find a local maximum of the scoring function it uses so that the resulting multiple alignment may not be the best possible.

[TJay]

Thanks. Got the idea

Answer 2

[Manu Prestat]

Often, heuristic algorithms are driven by an optimization function, e.g. looking for the minimization of the penalty score for an alignment. If the algorithm does not guaranty the "global" minimum score, it means that the minimum found can be a "local" minimum. Thus, the best solution (in this case the best alignment) is not guarantied by the heuristic.

Answer 3

[natasha.sernova]

"The input for ClustalW is limited to a maximum of 500 sequences or to a 1MB file".

http://www.ebi.ac.uk/Tools/msa/clustalw2/help/faq.html

Is there a limit on the number of sequences or the size of the file that I submit to ClustalW?

"The input for ClustalW is limited to a maximum of 500 sequences or to a 1MB file (whichever is smaller). When the input file or the number of sequences is large, ClustalW can run for days and in some cases may not finish at all. If you plan to input large amounts of data/sequences you should use the email results option. For batch runs of ClustalW see our web services tools: http://www.ebi.ac.uk/Tools/webservices/