Reason for that effort was the detection of an error in one of the entries
in BLOSUM100 (a missing minus, see diff).
To guarantee reproducibility (and check all matrices for errors),
the blosum matrix files from NCBI at
ftp://ftp.ncbi.nlm.nih.gov/blast/matrices/ have been downloaded and
the generate_blosum_data.py script spits out the required content
in subst_weight_matrix.cc.