Bayesian Variant Calling for Cell-Free DNA Sequencing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for detecting cancer-indicative signals in circulating tumor DNA from blood samples face challenges due to low sequencing depth and errors introduced during sample preparation and sequencing, leading to unreliable variant calling and false positives.
Innovation Solution
A processing system that uses Bayesian hierarchical models to filter out false positives by determining the likelihood of true alternate frequencies in cell-free and genomic nucleic acid samples, incorporating noise models and filtering processes to identify artifacts and enhance sensitivity and specificity in cancer prediction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional variant calling methods are used on cell-free DNA sequencing data, then variant detection can be performed, but the sequencing depth is insufficient and false positives increase due to low tumor DNA fraction
Solution Approach 1:
The patent segments the variant calling process into multiple independent stages: (1) candidate variant identification from sequencing data, (2) artifact probability calculation using separate artifact models for different error sources, (3) true variant probability calculation using population frequency models, and (4) final variant classification by comparing probabilities. This segmentation allows each stage to address specific error sources independently, improving overall accuracy while reducing false positives from any single error source.
Solution Approach 2:
The patent introduces probability scores as intermediary metrics between raw sequencing data and final variant calls. Specifically, it calculates artifact probabilities and true variant probabilities as intermediate values that mediate the decision-making process. These probability intermediaries allow the system to quantify uncertainty from different sources (sequencing errors, artifacts, low tumor fraction) and combine them systematically to reach reliable variant calls.
2Measurement precision
If sequencing depth is increased to improve tumor DNA detection, then sensitivity improves, but computational resources and time required for analysis increase significantly
Solution Approach 1:
The patent performs preliminary actions by pre-calculating and storing artifact probability models and population frequency models before actual variant calling. These pre-computed models are stored as reference data that can be quickly applied during analysis. This preliminary preparation reduces the computational burden during actual variant calling, allowing high sensitivity analysis without proportional increases in processing time.
Solution Approach 2:
The patent changes parameters by working with probability scores rather than raw read counts, and by using log-odds ratios to transform multiplication operations into addition operations. This parameter transformation simplifies the computational complexity of combining multiple probability factors, enabling efficient analysis even with high sequencing depth data.
3Reliability
If artifact filtering is applied to reduce false positives, then specificity improves, but true positives may be incorrectly filtered out reducing sensitivity
Solution Approach 1:
The patent applies local quality by calculating artifact probabilities and true variant probabilities separately for each candidate variant based on its specific characteristics (position, sequence context, observed artifacts). Rather than applying a uniform filtering threshold to all variants, the system evaluates each variant's likelihood of being an artifact versus a true variant using locally-specific probability calculations, preserving true positives while filtering artifacts.
Solution Approach 2:
The patent implements feedback by using population frequency information and artifact patterns observed in normal samples to inform the filtering decision for each variant. The system continuously refines its probability estimates by comparing individual variant characteristics against population-based expectations and artifact models, allowing dynamic adjustment of filtering stringency based on the specific evidence for each variant.
Data Source
AI summary
A processing system uses a Bayesian inference based model for targeted sequencing or variant calling. In an embodiment, the processing system generates candidate variants of a cell free nucleic acid sample. The processing system determines likelihoods of true alternate frequencies for each of the candidate variants in the cell free nucleic acid sample and in a corresponding genomic nucleic acid sample. The processing system filters or scores the candidate variants by the model using at least the likelihoods of true alternate frequencies. The processing system outputs the filtered candidate variants, which may be used to generate features for a predictive cancer or disease model.