SureTypeSC

SureTypeSC applies a two-stage machine learning approach to produce high-confidence genotypes from single-cell whole-genome amplified (WGA) DNA profiled on Illumina SNP bead arrays, mitigating WGA-induced noise for single-cell genotyping and downstream analyses.

Key Features:

• Large SNP resource: Uses a dataset of 28.7 million SNPs typed at high confidence from WGA DNA derived from 104 single cells across two Coriell repository cell lines.
• Array platform: Operates on data generated with Illumina SNP bead array technology.
• Reference genotype establishment: Leverages trio information (mother-father-proband) from multiple technical replicates of bulk DNA to establish high-quality reference genotypes on the SNP array.
• Two-stage machine learning: Implements a two-stage algorithm that employs Random Forest and Gaussian Mixture models to predict genotypes and filter noise.
• Genotype confidence scoring: Quantifies per-genotype confidence using Bayesian statistics.
• WGA noise mitigation: Distinguishes true genetic variants from artifacts introduced by whole-genome amplification while retaining the majority of high-quality SNPs.
• Implementation: Implemented in Python.

Scientific Applications:

• De novo mutation detection: Enables identification of de novo variants in single cells with improved confidence despite WGA artifacts.
• Linkage analysis: Supports linkage studies that require accurate single-cell genotype calls.
• Lineage tracing: Facilitates reconstruction of cell lineages by providing high-confidence SNP calls from individual cells.
• High-confidence single-cell genotyping: Provides robust genotype calls for single-cell genomics studies that require mitigation of WGA-induced errors.

Methodology:

Generates SNP calls from Illumina SNP bead arrays on WGA single-cell DNA; establishes array reference genotypes using trio information and multiple technical replicates of bulk DNA; applies a two-stage machine learning pipeline using Random Forest and Gaussian Mixture models for genotype prediction; and uses Bayesian statistics to quantify per-genotype confidence; implemented in Python.