Strand bias in variant calls

[wangzkai]

Most, if not all, strand bias filters of SNV calls require that a variant to be observed on reads aligned to both directions. While I understand the rationale behind this is to avoid strand specific sequencing errors (such as the famous CCGG tetramer), I wonder whether one should also check whether the variant allele is supported by DNA fragments originated from both strands of the original DNA molecule? The reason I think there is a difference here is because the strandness of a read is not always the same as the strandness of the DNA fragment it was derived from. For Illumina PE libraries, my understanding is that the strandness of the first read should correspond to the strandness of the DNA fragment. Are there existing tools/callers that look at this type of strandness and strand bias filters?

Thoughts and comments are welcomed. Thanks!