It can happend in 2 reasons:

1) different algorythms are used in alignment generation - ClustalOmega for Uni-prot and COBALT is more complicated constraint based alignment.

2) different defaults in these two software (for example gap opening and extension penalties etc.)

More information see there

http://www.ncbi.nlm.nih.gov/tools/cobalt/cobalt.cgi?CMD=Doc

http://www.ncbi.nlm.nih.gov/pubmed/17332019

http://omictools.com/cobalt-s6072.html

http://a-little-book-of-r-for-bioinformatics.readthedocs.org/en/latest/src/chapter4.html

http://www.ebi.ac.uk/training/online/course/uniprot-exploring-protein-sequence-and-functional/how-use-uniprot-tools/aligning-multiple-pro

http://toolkit.tuebingen.mpg.de/clustalw/help_params

I hope this would be helpful for you

It can happend in 2 reasons:

1) different algorythms are used in alignment generation - ClustalOmega for Uni-prot and COBALT is more complicated constraint based alignment.

2) different defaults in these two software (for example gap opening and extension penalties etc.)

More information see there

http://www.ncbi.nlm.nih.gov/tools/cobalt/cobalt.cgi?CMD=Doc

http://www.ncbi.nlm.nih.gov/pubmed/17332019

http://omictools.com/cobalt-s6072.html

http://a-little-book-of-r-for-bioinformatics.readthedocs.org/en/latest/src/chapter4.html

http://www.ebi.ac.uk/training/online/course/uniprot-exploring-protein-sequence-and-functional/how-use-uniprot-tools/aligning-multiple-pro

http://toolkit.tuebingen.mpg.de/clustalw/help_params

I hope this would be helpful for you

In short - The issue is due to the fact that when we align two sequences, we do not know the right answer! I.e. exactly what substitutions/deletions have occurred is "hidden" from us. We address this by making an "educated guess" as to how the real alignment is.

Different alignment approaches "guess" in different ways, hence the answer (guess) will often be different.

Hope it makes sense!

Good point by Leon: but I'd go even further. When we align to sequences there is no RIGHT answer, because nature does not align sequences at all. The whole alignment process is an artificial, computational algorithm with no direct relationship to experiment (though very useful in bioinformatics).

Different alignment algorithms may not only work in different ways but also have different definitions of the best match between sequences, so their answers may be often different.

Very nice and helpful answers! Just to expand a little bit one of the points raised. The output of an alignment depends on the algorithm we use. In the easiest case of aligning two sequences we give  "x" positive points everytime that two aminoacids coincide and "y" positive points when there are conservative changes, then "a" negative points when there's a non conservative change, "b" negatives when there is a gap (gap-penalty), and "c" negatives for extending the gap. The values for a,b,c,x and y are defined by the programmer. The algorithm usually makes different alignments and presents the one that gives the highest total value... but as Bartosz said, nature does not align sequences

The proteins you are aligning are completely unrelated to eachother, so any alignment you are seeing is more the result of random matches. Random effects do tend to be very different between algorithms. In general, when you align sequences that are very similar, the difference between the alignment done with different algorithms will be minor. For more distant sequences, the evalutionary path between the sequences becomes more muddied and more difficult to predict, so different algorithms will start giving different results. If you align two sequences with no evalutionary path between them, well the results are just a random effect and you will likely see no to little correlation between the results from different algorithms. 

For tRNA synthetase this is not really unexpected. The catalytic domain should be very well preserved, but the binding sites will not be preserved. As we are not seeing the whole alignment, I suspect the part in the screenshot does not contain the catalytic domain (but I'm not that familiar with the structure of tRNA synthetases).

Thank you very much for the feedback. The tips were really helpful in improving my understanding regarding the concept behind these alignment tools. I am indeed very new in this field. A lot more to learn from experience researchers. 

Though my examples I used here (cysteinyl-tRNA synthetase vs methionyl-tRNA synthetase) were chosen mainly because I was just trying to compare my protein against all class I tRNA synthetase. Furthermore, CysRS and MetRS is highly homologous throughout the class I structure. In my study, I am trying to determine the path of evolutionary preserved domain of certain site (eg: catalytic site) and relate it to its structural conformation and physiological function. To actually compare all those proteins with my protein as Katleen said would make things "muddier". Probably I need to reduce my scope.

Thanks again for your useful insights. Really appreciate them! 

If you have different variants of your protein (i.e. different 'versions', which are different from each other due to deletions/insertions/substitutions) each with at associated phenotype, e.g. catalytic efficiency, the you could use the SigniSite server (see link) to identify amino acid residues significantly associated with either high or low catalytic efficiency...

http://www.cbs.dtu.dk/services/SigniSite/

There are different objectives of each alignment tool, hence, results will come differently as default matrices and gap opening and closing costs are different for each one. Details you will find in Mount"s Bioinformatics book and several recent reviews.