Probabilistic SNV Detection in Cell-Free DNA for Minimal Residual Disease
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for detecting somatic single nucleotide variants (SNVs) in cell-free DNA (cfDNA) face challenges such as low tumor fraction, low sequencing coverage, and high overlapping rates, which hinder accurate detection and estimation of tumor fraction.
Innovation Solution
A novel probabilistic method using a Bayesian-based probability framework to estimate the likelihood of genotypes from sequencing data, which includes steps to estimate the global tumor fraction, determine SNVs, filter candidates based on cfDNA properties, and re-estimate SNVs to improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing SNV detection methods (VarScan2, MuTect) are used on cfDNA samples, then they can detect somatic mutations in solid tumor samples, but they fail to identify somatic SNVs in cfDNA even with strong evidence due to low tumor fraction
Solution Approach 1:
The patent modifies the detection parameters by incorporating a tumor fraction parameter (θ) into the likelihood calculation. The genotype likelihood function is changed to account for the mixed normal-tumor composition of cfDNA, allowing accurate SNV detection even when tumor fraction is as low as 0.1%. This parameter change enables the method to distinguish true somatic variants from sequencing errors in low-tumor-fraction samples.
Solution Approach 2:
The patent performs preliminary estimation of tumor fraction (θ) and sequencing error rate (ε) before SNV detection. By estimating these parameters first using maximum likelihood estimation on the sequencing data, the method prepares the necessary contextual information to accurately interpret variant allele frequencies and distinguish true mutations from artifacts in low-tumor-fraction cfDNA samples.
2Measurement precision
If deep sequencing is used to capture tumor signals in cfDNA (Capp-Seq with iDES), then sensitive detection of cancer mutations is achieved, but the expensive cost limits sequencing to only a small panel of genomic regions
Solution Approach 1:
The patent changes the analytical approach by using a probabilistic model that incorporates tumor fraction estimation and sequencing error suppression. This allows the method to achieve high detection sensitivity without requiring ultra-deep sequencing of limited panels. The likelihood-based framework efficiently utilizes available sequencing data to detect SNVs across broader genomic regions at lower sequencing depths.
Solution Approach 2:
The patent replaces the mechanical approach of increasing sequencing depth (more cycles, higher cost) with a computational approach using probabilistic modeling and maximum likelihood estimation. This substitution allows the method to achieve equivalent or superior detection sensitivity through statistical analysis rather than brute-force sequencing, enabling broader genomic coverage at reduced cost.
3Extent of automation
If SiNVICT computational SNV caller uses Poisson function to model allelic distribution in cfDNA, then it provides probabilistic modeling, but the model is overly simplistic and fails to consider the impure nature of cfDNA, resulting in high number of false positives
Solution Approach 1:
The patent improves the probabilistic model by changing from a simple Poisson distribution to a more sophisticated likelihood function that explicitly models the mixed normal-tumor genotype composition. The new model incorporates parameters for tumor fraction (θ) and sequencing error rate (ε), and uses a product of probabilities across all reads covering a locus. This enhanced modeling significantly reduces false positives while maintaining computational automation.
Solution Approach 2:
The patent creates a composite probabilistic model that combines multiple components: the binomial distribution of allele counts, the tumor fraction parameter, the sequencing error rate, and the joint genotype probabilities. This composite approach captures the complex impure nature of cfDNA better than the simple Poisson model, achieving higher reliability in SNV detection.
4Device complexity
If existing methods do not consider overlapping read mates in cfDNA sequencing data, then they simplify the analysis, but they miss valuable information for detecting mutations in cfDNA
Solution Approach 1:
The patent merges information from overlapping read mates by incorporating all reads covering a genomic locus into the likelihood calculation. When read mates overlap, their information is combined multiplicatively in the probability model, increasing the effective coverage and statistical power. This merging of information from multiple reads improves mutation detection accuracy without significantly increasing analysis complexity.
Data Source
Figure 1
Figure 2
Figure 2
AI summary
The present disclosure provides a probabilistic model for accurate and sensitive somatic single nucleotide variant (SNV) detection in cell-free nucleic acid samples comprising a set of sequence data. A joint genotype may be determined for each locus in the set of sequence data, and germline mutations may be intrinsically removed. A set of filtrations can be applied to eliminate low quality somatic variant calls. Further, a global tumor cell-free deoxyribonucleic acid (cfDNA) fraction and overlapping read mates can be considered, thereby enabling accurate SNV detection and variant allele frequency estimation from samples with low tumor cfDNA fraction. A sensitive early detection of minimal residual disease (MRD) is designed by using the probabilistic model and the machine learning model for distinguishing true variants from sequencing errors.