Predicting clone genotypes from tumor bulk sequencing of multiple samples.

Area of Science:

• Oncology
• Bioinformatics
• Computational Biology

Background:

• Bulk sequencing of tumors reveals significant genomic heterogeneity within patients.
• Numerous computational methods exist for inferring tumor cell population (clone) genotypes from bulk sequencing data.
• The accuracy of these methods for estimating clone counts and genotypes is not well-established.

Purpose of the Study:

• To assess the performance of nine computational methods for inferring tumor clonal genotypes.
• To compare the accuracy of existing methods with a new method, CloneFinder.
• To identify limitations and challenges in current computational approaches for clonal deconvolution.

Main Methods:

• Performance evaluation of eight established methods and one novel method (CloneFinder).
• Analysis using computer-simulated datasets with varying clonal evolutionary relationships.
• Assessment of clone genotype inference accuracy and clone count estimation.

Main Results:

• CloneFinder, LICHeE, CITUP, and cloneHD achieved low error rates (<5% per clone) for evolutionarily related clones.
• Methods performed poorly on datasets with mixtures of clones from different lineages.
• Clone count estimation varied, with underestimation by cloneHD and overestimation by PhyloWGS.
• BayClone2, Canopy, and Clomial required prior knowledge of clone number; AncesTree and Canopy had limited output.

Conclusions:

• Deconvoluting tumor clone genotypes from single nucleotide variant (SNV) frequency differences remains challenging.
• Current computational methods show variable performance, particularly with complex clonal architectures.
• There is a need for improved computational methods and robust software for accurate clone genotype inference.