I thought that Needleman-Wunsch is the best approach to align sequences. However, I read that it is impossible to evaluate the statistical significance of the alignment if you do global alignment. So I would like to ask, what is the point of global alignment if we cannot evaluate statistical significance?
It's best to flip the question on its head and ask yourself instead, "Do we care about the statistical significance of an individual alignment?" The answer is "yes" if you're in a context like homology modeling, where you're aligning two protein sequences. The answer, however, is a very definite "no" in the context of NGS, where you just want the most likely alignment and a vague assessment of its likelihood of being wrong (for filtering purposes).