Methods and systems for detecting genetic variants

What is this patent about?

’063 is related to the field of genetic analysis, specifically the detection of rare genetic variants and copy number variations (CNV) in cell-free DNA (cfDNA) samples. In liquid biopsy applications, such as cancer monitoring, the ability to accurately quantify DNA fragments is often hampered by the loss of molecules during library preparation and sequencing, as well as errors introduced during PCR amplification.

The underlying idea behind ’063 is that the number of original DNA molecules in a sample can be more accurately estimated by tracking the recovery of complementary strands from individual double-stranded fragments. By using duplex tagging to distinguish the Watson and Crick strands, the system can identify which molecules were recovered as pairs, which were recovered as single strands, and—crucially—statistically infer the number of molecules that were entirely lost during the process.

The claims of ’063 focus on a method for classifying sequence data by non-uniquely tagging double-stranded cfDNA with a significant molar excess of barcoded adapters. The process involves grouping mapped sequencing reads into families based on their molecular barcodes and genomic start/stop positions to generate consensus sequences. These consensus sequences are then categorized as either paired (representing both strands of the original duplex) or unpaired (representing only one strand).

In practice, the invention utilizes a library of adapters where the number of unique barcodes is intentionally smaller than the number of DNA fragments, relying on the combination of the barcode and the fragment endpoints to uniquely identify parent molecules. This high-efficiency ligation ensures that a significant portion of the sample is tagged at both ends, allowing the bioinformatics pipeline to collapse redundant reads into high-fidelity consensus sequences while filtering out artifacts that appear on only one strand.

This approach differentiates itself from prior methods by moving beyond simple error correction to address the problem of stochastic sampling loss. By quantifying the ratio of paired to unpaired strands at specific genetic loci, the method corrects for local sequencing biases and missing data. This statistical inference provides a more robust foundation for detecting minute changes in copy number and identifying rare somatic mutations with a specificity exceeding 99.9%.