Is a PAM log odds matrix suitable for determining probability?

Question

[pp425x]

I am trying to analyze some mutations with codons changes where for example GTG changes to GCG (Val -> Ala) and I was wondering if there were an effective method for determining the odds of that amino acid changing to another. Would the PAM10 log-odds be useful for this?

I've done some searching online and came across the following bit of information:

Where the log-odds ratio > 0, the occurrence is more often than by random chance

Where the log-odds ratio < 0, the occurrence is less often than by random chance

Based on my example of (Val -> Ala), the corresponding amino acid letters would be A and V and in the PAM10 matrix the log odds score is -5. So based on that would I be able to conclude that it is unlikely that GTG changing to GCG was by chance? In other words it was not random. Would this make sense or is this not a good approach? I'm just not sure if I can use this model or if it is primarily intended for large sequences rather than a single codon.

Answer 1

[Mensur Dlakic]

Your interpretation of the substitution matrix meaning is correct. According to PAM10 matrix, Val -> Ala is unlikely to occur.

However, you should understand the meaning of substitution matrices, which is explained here. Generally speaking, PAM matrices with low numbers are for very related sequences - PAM10 would be in that category. BLOSUM matrices are the opposite - for them low numbers mean less related sequences. That point is highlighted here. Depending on which matrix you choose as representative for your proteins, the likelihood of Val -> Ala will change.

BLOSUM62 is by far the most popular matrix for the purposes of sequence comparisons that detect relationships between proteins. Val -> Ala has 0 in the BLOSUM62 matrix, which means that the substitution is neutral.