Ylaine/2009-8-25

From ZhangLabWiki
Jump to navigation Jump to search

Improving Clustering[edit]

  • Re-cluster rejects from first round
[error, groups] = snpCluster(nodb_match,[nodb_mismatch; nodb_miss]);
[reject_match, reject_miss] = filterSNPs(nodb_match, [nodb_mismatch; nodb_miss], groups);
[error, groups2] = snpCluster(reject_match(:,1:3), reject_miss(:,1:3));
    • Best combination so far is quality, coverage, and 10*heterozygosity, with 6.7% error: 49% MD, 3.01% FP
    • Rejects 2890 SNPs
  • Minimize over weights on heterozygosity: local minimum around 14, becomes erratic about 14.5. Lowest error rate is 6.64% (rejects 2888 SNPs)
' No dbSNP Clustering
Match 1k 536 416
Miss 1k 2821 86
Mismatch 1k 73 40
SNPs Called 20761 17873
False Call rate 0.139395983 0.00704974
  • 20x decrease in error

File:Cluster vs thresh.jpg

    • These are similar to last week's but now an accurate call is defined as being in dbSNP OR matching 1K Genomes

File:Aug17 percent acc.jpg File:Aug17 percent acc closeup.jpg

False Negatives[edit]

  • Same procedure as on 7/30
../Scripts/snpCovered2.pl gold.xaa orig.pileup > gold.coveredaa&