Unless you are certain about gene sequences and need very precise data, there is a workaround: assume that ratio of synonymous sites to nonsynonymous per a genomic region, say of 1000 nucleotides, is constant, then Ka/Ks is approximately your synonymous/nonsynonymous SNP data multiplied by that constant. That constant is the same between genes in first approximation and you can rank genes using synonymous/nonsynonymous SNP as if it is Ka/Ks. Now you need to "normalize" by looking at genes you have a neutral selection for sure. Instead of normalization you can test if ratio for a particular gene or set of genes is different from that "norm" that corresponds to neutral selection using chi-squared test or similar statistic. Data might need to be filtered a bit to remove outliers. This is simplified aproach, but other more precise approaches can be used too.