From:
`To:
`Subject:
`Date:
`Attachments:
`
`John West
`Richard Chen; Scott Kirk; Mark Pratt; Christian Haudenschild
`Slide for accuracy requirements walkthrough meeting 3-5pm today
`Tuesday, October 9, 2012 1:58:57 PM
`Accuracy mtg 9Oct2012 JW.pptx
`
`All,
`
`I put together one slide which summarizes my understanding of where we are headed on
`accuracy and how that should tie in to the product specifications we have for upcoming
`versions. Perhaps we can start our 3-5pm meeting by discussing this framework ?
`
`Thanks,
`
`John
`
`Personalis EX2096
`
`

`

`JW Accuracy Overview
`
`CHARACTERIZING ERRORS
`# by type, mechanism, medical application
`
`Methods to identify errors:
` A / B repeatability or vs Gold Std
` MIE (need SV, not just SNP)
` Single genome methods:
` Tri-allelic loci
` Odd allele ratios at het loci
`Low / no coverage (WGS & exomes)
`Coverage variability sample to sample
`Particularly for SV detection in exomes
`
`Define regions linked to mechanisms
`Download from UCSC browser (mostly done)
`Run alignment degeneracy alg on MiSeq :
`Read lengths
`Molecule lengths
`Coverage statistics
`To define low or variable coverage regions
`
`Pareto analysis
`# errors by type, error mechanism
`Focus this by medical application
`
`Re-Do this with modern data
` v3 seq chemistry, gel & PCR-free WGS
`v3 seq chem exomes
` CEPH1463, Venter, Diversity panel
` Data purchased, downloaded, seqc’d
`
`FIXING ERRORS
`
`Bioinformatic approaches:
` Better reference (NEMAR, SV version, hg20)
`Improve SV algorithms (many parts)
`Upstream-only base call around homopolymers
`LobSTR
`Use of array data (Genotyping and SV)
` Better alignment algs (3rd party & ours)
`
`Lab & bioinformatic approaches
` Pullout & integration with regular data
`Exome supplement
`Long read / molecule (SD’s, compressions)
`High GC
`Optimize operation for accuracy vs $
`Novel Assays (Pulsed field, phasing, etc)
`
`Content development
` Data downloaded, purchased, seqc’d in-house
`e.g. for BreakSeq, norms, GT ever seen ?
`MAF for SV/CNV/InDel, ethnic controls
`
`FLAGGING ERRORS
`
`Descriptors, not just Q-Scores:
`Why we think the Q is low
`
`More accurate Q-Scores
` Difficult sell due to competitive noise
`
`PRODUCT SPEC / FEATURE
`
`Reduced error rates
`Focused on medical content
`Quantitative Personalis vs Standard
`By method to ID errors
` by variant type &mechanism:
`SNP
`NEMAR hg20 vs Public hg19
`Coverage filled in
` Homopolymer region caller
`Alignment degen w long reads
` Optimize for accuracy vs $
`STR’s, InDel’s
`SV’s
`Qualitative too:
`e.g. SV zygocity & in exomes
`e.g. STR analyzed at all
`e.g. Reporting tri-allelic loci
`
`More content covered
` Finished genes in exomes
`Medical content outside exons
`Whole genome isn’t:
`Degenerate alignment (e.g. HLA)
`Low coverage
`
`Flags which guide the type of
`validation / follow-up
`
`© Personalis, Inc. All Rights Reserved. COMPANY CONFIDENTIAL.
`
`1
`
`Personalis EX2096
`
`