SNP

Base Calling

• Minimum Read depth
• Based on Phred scores
• 1% error rate
• Alignment (Trade off between accuracy and read depth)
• Recalibration of Pherd scores
  • Essential
  • Phred score of Q should be = 10 to the power Q by 10 or less. This is done by alignning with the reference with the known SNPs

Homo and Heterozgous SNPs in a diploid

• Homozygous -> If an SNP (different than ref) base is counted across the read depth to be more than 80%
• Hetorozygous -> If an SNP (different than ref) base is counted across the read depth to be less than 80%
• Sequence/alignment Error -> If an SNP based is counted to be less than 10%
• This is true of the depth is minimum of 20x

Accuracy in SNP calling

• Accuracy can be improved from single(Ref vs one sample) to multi samples (Ref vs several samles).
  • However false positives (SNP call) would also increase with more number of samples
• Possible accuracy by read depth based SNP calling is 85%
• Possible accuracy by LD (linkage disequilibrium) is >95%
  • Possible only when multi samples are used
  • Software that uses LD for SNP calling is Beagle, IMPUTE2, QCall, MaCH

Plan for SNP calling

• Assumptions
  • Multiple Genotypes instead of Ref vs One
  • Right combination (contrasting genotype types for specific type ) vs ref.
  • LD based SNP calling
  • Cross check the SNPs against all the 18 genotypes vs contrasting types
• Filtering
  • Use of LD (Software would estimate this)
  • HapMap data (Go-through, how haplotypic frequency will help in filtering the best SNPs )
  • Deviations from HWE (Estimate the allelic frequencies when HD is estimated and figure out how to estimate the variant SNPs to filter them off)