I am working on a cancer sample exome data. I have used bowtie2 for alignment, samtools mpileup to call variations and snpEff for annotation. I am going to use vcfutils.pl varFilter to filter out false positive variations. I would like to know about the optimal value for the following filtering criterias :
Read Depth (DP)
Number of reads showing variation (DP4)
P value (PV4)
HWE P value (HWE)
Mapping quality (MQ)
Is there are any other important criterias to be considered to filter out false positives or any other potential tools?
So, there is not "the one" answer to your question, it depends on your data!...
What is your overall read depth? In exome sequence data sets, target regions are covered very unequally, you have to keep that in mind. The minimum read depth also depends on the required frequency of variants. If you desire five reads at minimum and call variants from 20% frequency on, one single read having a mutation (or read error?!) out of five would be enough to be called...
The variant frequency also depends on the purity of your sample, as heterozygous variants would be expected to have ~50% variant frequency.
0.05 is the most usual threshold for p-values, but how is it computed? Keep in mind that mutations called with low evidence (low read depth and variant frequency) usually depend much more on p-value computation as one read more or less may result in dramatic p-value changes!
The mapping quality tells you how "sure" you can be that the read is correctly mapped in a logarithmic manner and you should also add a filtering on base quality.
Thank you for your answer. Yes, you are right and I am aware of those details you mentioned. I have 98% of target covered with 5X, so I don't think I have any problem with coverage. I guess this issue comes with frequency filter only. Instead I thought of setting minimum threshold. This is what I have planned now, I am going to filter variations based on DP4 info in VCF format. It gives four values as follows,
1. Number of HQ reads showing same as reference base in forward strand
2. Number of HQ reads showing same as reference base in reverse strand
3. Number of HQ reads showing alternate base in forward strand
4. Number of HQ reads showing alternate base in reverse strand
I am planning to filter variations which has atleast 3 reads supporting a variation coming from both strands. This way, I can filter out strand biased variations, variations from Low Quality reads and few reads to support a variation. I know that still I will have few false positives but I don't want risk of losing true positives.
- Badges
- Science method