Multiple Alignment with Profile HMMs

Profile HMMs can help us to obtain an approximate solution to the multiple alignment problem. Given n sequences $S^{(1)},\ldots,S^{(n)}$, consider the following cases:

1.

If the profile HMM $\mathcal{P}$ is known, the following procedure can be applied:

• Align each sequence S⁽ⁱ⁾ to the profile separately.

• Accumulate the obtained alignments to a multiple alignment.

2.

If the profile HMM $\mathcal{P}$ is not known, one can use the following technique in order to obtain an HMM profile from the given sequences:

• Choose a length L for the profile HMM and initialize the transition and emission probabilities.

• Train the model using the Baum-Welch algorithm, on all the training sequences.

• Obtain the multiple alignment from the resulting profile HMM, as in the previous case.

One can use an extension of the above approach to identify similar patterns in a given set of sequences by using the profile HMM for local alignment (see figure 6.6). Next, we present another approach to this problem, which tackles the problem from a totally different perspective.