Ylaine/2009-7-9
Jump to navigation
Jump to search
SNP Filtering[edit]
- Wrote 'extractSNPlines.pl' to get pileup lines by line number (instead of grep-ing for position)
./extractSNPlines.pl notwo.pileup snoOrigLocs.txt > snp.pileup
- Discovered that pileup does not output SNP calls. Modify 'heterozygousSNPfilter.pl' to extract heterozygous SNPs 'cns.final.snp', put call into hash keyed by position. Then iterate through lines of snp.pileup and count occurrences of each called nucleotide.
Bayesian SNP Calling[edit]
- We want to determine if a SNP called as heterozygous is actually homozygous (both alleles same as reference)
- Assume binomial discribution of nucleotides
Calculating False Positives[edit]
- dbSNP is a possibly biased database, while the 1000genome SNPs are unbiased (based on sequencing data, not just looking at known SNP locations).
- 1000genome SNP list may be smaller than list of true SNPs.
- If there are no false positives, then the proportion of our SNPs in common with dbSNPs should be independent of presence in dbSNP.
- Table and FP calculation edited 7/15/09
1K Genome | In dbSNP | Percent | Not in dbSNP | Percent |
Match | 6300 | 85.40% | 171 | 17.12% |
Miss | 648 | 8.78% | 807 | 80.78% |
Mismatch | 429 | 5.82% | 21 | 2.10% |
Total | 7377 | 999 | ||
- Calculate false positive rate: if misses should only account for 5.82% of SNPs, then there should be 58.10 misses for the 999 SNPs not in dbSNP. This means the rest of the SNPs are false positives, for a total FP rate of 8.59%
Yesterday's SNP Frequency Measurements[edit]
- Possible over-representation of heterozygous SNPs: much more common among presumed false positives than direct hits.
' | In dbSNP | Not in dbSNP |
Homozygous: | 55.33% | 20.72% |
Heterozygous: | 44.67% | 79.28% |
Transition: | 76.48% | 67.17% |
Transversion: | 23.52% | 32.83% |
Total: | 7377 | 999 |