Background

Personalized Medicine promises to use genomic data to design treatment tailored to genetic make-up. However, all sequencing technologies are subject to a certain amount of error, resulting in either the “calling” of a variant that the sample DNA does not possess (false positive), or failing to identify one that it does (false negative).

Technology Overview

Variant calling involves separation of a signal from noise. Some noise is random, but much of it is genomic-context dependent. Analysis of patterns of variation across all the reads from multiple reference samples (e.g. >1000 genomes) provides a profile of the noise to be expected at that locus. Variant calling can be substantially improved by comparing the variants observed at each locus (from a test or query sample) against this noise pattern to determine whether what is seen is entirely consistent with background noise, or represents a ‘signal’ - i.e. a true variant. The invention involves a protocol for scanning query data (sets of reads, that are typically stored in a single ‘BAM’ file) against a reference genome sequence set (, ).

Benefits

• Improved variant calling from 63% using a leading company’s tech to 100% using proposed method
• Reduces false positive rate by providing statistical significance for accurate calls
• Reduces false negative rate by not having to filter out all call at a given locus, so called “black listing”

Applications

• Sequencing accuracy
• Sequencing sensitivity
• Personalized medicine
• Discovery research

CHLA Case No. 2017-013