 Solution

Use the number of alignment columns, including gap columns, as the denominator. A column contributes to sequence identity when its two nongap letters are equal, to sequence similarity when its two nongap letters have a strictly positive BLOSUM62 score, and to the gap fraction when either letter is a gap. A zero substitution score is not counted as similar.

The implementation in part (a) computes each numerator directly during its final pass through the aligned strings:

for name, a in alignments.items():
    print(name, a["alignment_length"], a["identities"],
          a["similarities"], a["gaps"])
    print(a["identity_percent"], a["similarity_percent"], a["gap_percent"])

Identical columns also count as similar for these data, so these percentages are overlapping summaries rather than three categories summing to 100%.