HRD Detection via Shallow WGS and Deep Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for detecting homologous recombination deficiency (HRD) in cancer patients require high sequencing depth and matched normal tissue samples, limiting their effectiveness and accessibility, especially for patients with low tumor proportion or low sequencing depth tumor samples.
Innovation Solution
A method using whole genome sequencing (WGS) data from tumor samples to generate genome copy number profiles and chromosome instability scores via a deep learning model, which identifies a risk score for HRD without the need for normal tissue samples, allowing for the detection of HRD at lower sequencing depths and in samples with low tumor proportion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep sequencing approach is used to identify HRD markers, then measurement precision is improved, but device complexity and cost increase
Solution Approach 1:
The patent changes the parameter of sequencing depth from high (500x) to low (0.1x-1x), fundamentally altering the technical approach. Instead of relying on deep sequencing to detect HRD markers, the invention uses shallow whole-genome sequencing combined with a machine learning model that analyzes copy number variations and chromosomal instability patterns, achieving accurate HRD detection without the need for high sequencing depth
Solution Approach 2:
The patent replaces the mechanical/physical approach of deep sequencing with a computational approach using machine learning. The system substitutes the need for high-depth physical sequencing with an algorithmic analysis of shallow sequencing data, using trained models to predict HRD status based on copy number profiles and chromosomal instability scores
2Measurement precision
If matched normal tissue samples are required for HRD detection, then measurement precision is improved, but ease of operation deteriorates
Solution Approach 1:
The patent extracts and removes the requirement for matched normal tissue samples from the HRD detection process. The invention is designed to work with tumor-only samples by using machine learning models that can identify HRD-related copy number variations and chromosomal instability patterns without needing to compare against normal tissue, thereby simplifying the sampling process
Solution Approach 2:
The system enables the tumor sample to serve itself by using the tumor's own genomic characteristics (copy number variations, chromosomal instability patterns) detected through shallow whole-genome sequencing. The machine learning model analyzes these intrinsic features to predict HRD status, making the assay self-sufficient without requiring external normal tissue controls
3Measurement precision
If high tumor fraction is required for accurate HRD detection, then measurement precision is improved, but adaptability deteriorates
Solution Approach 1:
The patent changes the parameter of tumor fraction requirement from high (>25%) to very low (can work with <5% tumor content). The shallow whole-genome sequencing approach combined with machine learning analysis of copy number variations allows the system to detect HRD patterns even when tumor cells represent a small fraction of the sampled material, greatly expanding applicability to heterogeneous samples
Data Source
AI summary
The present disclosure provides methods and compositions, e.g., kits, for detecting homologous recombination deficiency in a cancer patient. In certain embodiments, the methods disclosed are based on genomic copy number summarization and deep learning model to detect homologous recombination deficiency from extremely low coverage whole genome sequencing data.