# Why PHMM is used For Protein Analysis

I am interested on how profile hidden markov model is used for protein analysis in bioinformatics. Can you guys explain why phmm is more preferable method comparing to simple MSA(Multiple Sequence Alignment) scoring or simple MSA probability scoring? For example we have three MSA sequences as based model and a target to be scored.

A B C B C A
A B B C C A
A B C B A A

### Target Sequence

A B B B C A

By MSA scoring we can count score for target sequence comparing to a profile MSA. For example above target sequence protein column will counted as match(A), match(B), mismatch(B-majorityMSA(C)), match(B-majorityMSA(B)), match(C-majorityMSA(C)), match(A). Just give score for match and mismatch and we can get total score for a target sequence comparing to a profile MSA sequences.

By MSA probability scoring we can count probability of each protein in each column so for example above
P(target_seq similar to MSA profile) = P(A in MSA sequences)*P(B in MSA sequences)*.... *P(A in MSA sequences) = 1*1*(1/3)*(2/3)*(2/3)*1 = 4/27

Isn't MSA scoring like above examples powerful enough?

Thanks a lot !!

• Please edit your question and tell us exactly what type of analysis you are thinking of. By MSA, I assume you mean multiple sequence alignment (or are you thinking of some sort of mass spectrometry?), but I don't see how the two are related. Are you thinking of HMMs of protein domains? Something else? Are you referring to finding protein homologs? Tools like PSI-BLAST or HMMER?
– terdon
Sep 27, 2017 at 15:33
• I've edited my question, yes it's all about protein sequence analysis. Sep 28, 2017 at 1:31

The simple probabilistic scoring scheme you describe to query a target sequence against a (protein) family represented as a multiple sequence alignment (MSA) is actually similar in essence to a Position Weight Matrix (PWM), although PWMs are more refined. PWMs are powerful rapid tools often used for motif analyses.

However, your approach (and PWMs) don't handle insertions or deletions (indels), and you can't easily add these into your model (for example, consider that 2 1bp indels should not score the same as 1 2bp indel, and indels could occur within your MSA as well as in your target sequence). You could try and add some arbitrary gap opening and extension penalties somehow (like BLAST) but then you'd abandon your probabilistic model! By contrast, pHMMs have a fully fledged probabilistic indel model with insertion and deletion states.

Finally, of practical note, there are some excellent implementations of pHMMs for protein sequence analysis, such as HMMER3 and HHSuite.

• I am sorry, still don't get it. What do you mean by 2 1bp indel and 1 2bp indel ? Can you give some illustration? Oct 1, 2017 at 6:33
• @RamandikaPranamulia 1 2bp deletion: ATG--AT and 2 1bp deletion AT-GA-T Oct 2, 2017 at 14:44

I recommend you to read this paper in which the applications of HMM in protein sequence alignments are discussed.