Joint Mutation Calling and Phylogenetic Inference

20 concepts18 papers1 proposal

To map the evolutionary history of tumors or understand somatic mosaicism in healthy tissues, we must analyze genomes at the resolution of individual cells. However, a single human cell contains a mere 6 picograms of DNA. Because modern sequencing platforms require nanograms of input material, this DNA must undergo massive Whole Genome Amplification (WGA).

This chemical amplification introduces severe technical noise that mimics biological variation. Polymerase errors generate false-positive mutations (Alexej Abyzov et al., One thousand somatic SNVs per skin fibroblast cell set baseline of mosaic mutational load with patterns that suggest proliferative origin), while heavy amplification can create artificial chimeric reads that look identical to genuine gene fusions (Zijie Jin et al., Single-cell gene fusion detection by scFusion).

The most disruptive artifact is allelic dropout (ADO). ADO occurs when one of the two alleles at a heterozygous locus fails to amplify entirely (Jack Kuipers et al., Tree Inference for Single-Cell Data; Liang Chang et al., Single-cell whole-genome sequencing, haplotype analysis in prenatal diagnosis of monogenic diseases). If a cell carries a true somatic mutation on one chromosome, but that specific strand drops out during WGA, the sequencer reads only the wild-type allele. The cell is misclassified as unmutated, creating massive false-negative rates and extreme data sparsity (Tianyun Zhang et al., De novo identification of expressed cancer somatic mutations from single-cell RNA sequencing data).

Why Independent Variant Calling Fails

If we treat each cell as an isolated island and apply standard variant-calling algorithms independently, the results are highly inaccurate. Single-cell libraries lack the depth of bulk sequencing, meaning technical noise easily overwhelms the true biological signal. As demonstrated by Hossein Farahani et al. in Benchmarking bulk and single-cell variant-calling approaches on Chromium scRNA-seq and scATAC-seq libraries, independent cell calling suffers from high false discovery rates because it cannot distinguish non-biological artifacts from rare somatic mutations.

Furthermore, independent calling leaves us blind to the true history of the tissue. When a mutation is missing from a cell's sequence, we cannot determine if the mutation was genuinely lost due to a biological deletion or if it disappeared due to technical allelic dropout (Jack Kuipers et al., Single-cell sequencing data reveal widespread recurrence and loss of mutational hits in the life histories of tumors).

Shared Evolutionary History as a Statistical Regularizer

The core insight that makes single-cell variant calling work is that cells do not exist in a vacuum; they share an evolutionary lineage. Because mutations are passed down from mother to daughter cells, their distribution across a population follows a tree structure.

Instead of calling mutations first and building a tree second, modern frameworks jointly infer the phylogenetic tree and the cell genotypes. In this paradigm, the phylogenetic tree acts as a powerful statistical regularizer that "fills in the blanks" caused by allelic dropout.

Imagine a cluster of ten cells that share twenty rare somatic mutations, strongly indicating they belong to the same evolutionary subclone. If one cell is missing two of these mutations, an independent caller would classify it as a separate clone or a back-mutation. A joint probabilistic framework, however, evaluates the likelihood of a rare biological back-mutation versus a common technical dropout event. Because the tree topology heavily favors shared ancestry, the model correctly identifies the missing variants as false negatives caused by ADO (Jack Kuipers et al., SCIΦ: Single-Cell Mutation Identification via Phylogenetic Inference). This joint MCMC sampling strategy allows algorithms to reconstruct high-fidelity lineages despite messy data (Jack Kuipers et al., Single-cell mutation calling and phylogenetic tree reconstruction with loss and recurrence).

Without Tree Regularization (Independent Calling):
Cell 1: [Mut A, Mut B, Mut C]  --> Clone 1
Cell 2: [Mut A, Mut B, -----]  --> False New Subclone (due to ADO)
Cell 3: [Mut A, Mut B, Mut C]  --> Clone 1

With Tree Regularization (Joint Inference):
Cell 1, 2, and 3 share a common lineage node defined by [Mut A, Mut B, Mut C].
The missing value in Cell 2 is statistically corrected to a false negative.

Leveraging Raw Read Depths and Coverage

Early joint-inference models simplified data into binary matrices (0 for absent, 1 for present). However, binarization discards valuable statistical evidence. Modern frameworks operate directly on raw sequencing read counts and coverage depths (Senbai Kang et al., SIEVE: joint inference of single-nucleotide variants and cell phylogeny from single-cell DNA sequencing data).

By modeling the exact probability of observing a specific number of mutant versus wild-type reads, these tools distinguish between different types of missing data. For example, if a genomic site has zero reads, it represents an amplification failure. If it has twenty reads but they all show the wild-type allele, the model calculates the likelihood of a complete allelic dropout versus a true biological deletion. This allows frameworks to map complex genomic events, such as distinguishing true evolutionary deletions from uneven WGA coverage (Senbai Kang et al., DelSIEVE: cell phylogeny modeling of single nucleotide variants and deletions from single-cell DNA sequencing data).

Haplotype Phasing and Population Constraints

Another powerful mechanism to bypass ADO is to look at neighboring genetic variants on the same chromosome. Chromosomes are amplified as physical strands, meaning an amplification failure typically drops out an entire local haplotype block.

By leveraging known germline variations through read-phasing, algorithms can track exactly which parental chromosome was successfully sequenced. If a genomic region consistently yields reads from Haplotype A but none from Haplotype B, the model infers a localized dropout or a copy number alteration rather than a series of independent point mutations (Hazal Koptagel et al., Scuphr: A probabilistic framework for cell lineage tree reconstruction; Yifan Zhao et al., High-resolution detection of copy number alterations in single cells with HiScanner).

When individual cell data is too sparse to phase effectively, population-level genetics can step in. By utilizing linkage disequilibrium patterns from large reference panels, tools can statistically impute missing alleles caused by ADO, enabling accurate variant calling even from sparse single-cell transcriptomic or epigenomic data (Jinzhuang Dou et al., Single-nucleotide variant calling in single-cell sequencing data with Monopogen).

Pooling Information Across Cells and Modalities

Finally, when single-cell data is too noisy on its own, algorithms pool information across cells or integrate external data modalities to anchor the analysis.

One approach is to combine single-cell sequencing with robust bulk sequencing data. As shown by the authors of Integrative inference of subclonal tumour evolution from single-cell and bulk sequencing data, bulk data provides highly accurate variant allele frequencies across the entire tumor population. This population baseline constrains the search space for the single-cell phylogenetic tree, neutralizing individual-cell ADO errors.

Alternatively, cells can be grouped into clonal populations before final genotype assignment. Bayesian non-parametric clustering methods pool statistical power across similar cells, allowing the model to infer the true genotype of a clone collectively rather than struggling with the missing values of individual cells (Nico Borgsmüller et al., Bayesian non-parametric clustering of single-cell mutation profiles).

Key Foundational Work

Single-cell sequencing data reveal widespread recurrence and loss of mutational hits in the life histories of tumors (2017) — contributed 1 concept elements to this direction.
Tree Inference for Single-Cell Data (2016) — contributed 1 concept elements to this direction.
One thousand somatic SNVs per skin fibroblast cell set baseline of mosaic mutational load with patterns that suggest proliferative origin (2017) — contributed 1 concept elements to this direction.
History & citationVersion 2 · 2026-10-04

Current · Permanent link

  1. Version 2 (latest) · 2026-10-04 · 18 papers
  2. Version 1 · 2026-10-04 · 26 papers
Cite this direction

Latest version (a living page):

@misc{lacuna_allelic_dropout_c514018c,
  title = {allelic dropout},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/allelic-dropout/rdr_c514018c2ccf456e81fffcc3ce821b44},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-19},
}

This exact version (version 2, frozen text):

@misc{lacuna_allelic_dropout_c514018c_v2,
  title = {allelic dropout},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/allelic-dropout/rdr_c514018c2ccf456e81fffcc3ce821b44/v/2},
  note = {Version 2, frozen 2026-10-04. Accessed 2026-11-19},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake