How to not penalize a specific mismatch when aligning?

Question

[ervaulsy]

Hi, I'm trying to align bisulfite sequencing read file to a reference genome. I'm duplicating data analysis method in a paper. In the paper they are saying "... Following this three-letter alignment, we refined each alignment using the Smith–Waterman algorithm. We used a score matrix that does not penalize a mismatch between A in the read and G in the top strand of reference genome sequence" They used SOAP2 Aligner, I looked for options but I couldn't find one for this. How and where can i create a score matrix that doesn't penalize a specific mismatch (A and G)? Alignment rates were really low when I skipped this step.