Relaxing the Infinite Sites Assumption in Tumor Phylogeny

This direction is split into other directions: infinite sites assumption. Its last version stays readable and citable here.

Version 1 of 173 papers117 conceptsfrozen 2026-10-04

For decades, the bedrock of computational cancer biology has been a simplifying "shortcut" known as the Infinite Sites Assumption (ISA). The ISA posits that the genome is so vast that a specific mutation will only ever occur once in a tumor’s history and, once acquired, will never be lost. This assumption allows researchers to treat tumor evolution as a "perfect phylogeny"—a clean, branching tree where every branch is defined by the permanent gain of a new genetic trait.

However, as single-cell sequencing has matured, it has become clear that the ISA is frequently violated. Tumors are not just accumulating mutations; they are actively deleting them through large-scale chromosomal rearrangements and occasionally hitting the same genetic "hotspots" multiple times. A growing body of research is now focused on moving beyond the ISA to build models that account for mutation loss, back mutations, and parallel evolution.

The Eraser Problem: Mutation Loss via Deletion

The most common violation of the ISA in cancer is not a mutation "reverting" to its original state, but rather the physical deletion of the genomic segment that held the mutation. In solid tumors, copy-number alterations (CNAs) are rampant. If a cell acquires a point mutation (SNV) and its descendant later suffers a large-scale chromosomal deletion in that same region, the mutation is effectively "erased" from that lineage.

Traditional models like SCITE, developed by Jack Kuipers, Katharina Jahn, and Niko Beerenwinkel, initially provided a probabilistic way to handle the noise of single-cell data while adhering to the ISA. However, researchers soon realized that ignoring deletions leads to "impossible" trees or the misidentification of ancestral traits.

To solve this, newer frameworks like Phertilizer, developed by Leah L. Weber and Mohammed El-Kebir, explicitly model the loss of mutations. By treating chromosomal deletions as "erasers," these models can distinguish between a cell that never had a mutation and one that lost it, leading to a more accurate reconstruction of the tumor's "family tree."

Joint Inference: SNVs and CNAs Together

The intuition behind the most successful recent methods is that you cannot accurately track point mutations (SNVs) without also tracking the "canvas" they are written on (the copy number).

Methods like B-SCITE and DelSIEVE (by Senbai Kang and colleagues) perform joint inference. Instead of calling mutations first and building a tree second, they model the likelihood of the raw sequencing data given a tree that allows for both SNV gains and subsequent deletions.

DelSIEVE, for instance, uses an expanded genotype state space. It doesn't just ask "is the mutation present or absent?" but "is the mutation present, absent, or was the entire allele deleted?" This allows the model to distinguish between "allelic dropout" (a technical sequencing error) and a true biological deletion, a distinction that Jack Kuipers and others have shown is critical for avoiding "back-mutation" artifacts in phylogenetic trees.

Recurrence and the Finite Sites Model

While deletions are the primary cause of mutation loss, the ISA is also challenged by recurrent mutations—where the same genomic site mutates independently in two different branches of the tree (parallel evolution) or a mutated site reverts to its wild-type state (back mutation).

In A statistical test on single-cell data reveals widespread recurrent mutations in tumor evolution, Jack Kuipers and Niko Beerenwinkel demonstrated that these events are not just theoretical; they are widespread across human cancers. Rapid cell division and high selective pressure make it statistically likely that "hotspot" genes will be hit multiple times.

To accommodate this, researchers have moved toward finite-sites models. Tools like SIEVE relax the ISA to allow for multiple transitions at the same genomic location. This is particularly important in high-mutation environments, such as those found in mitochondrial DNA (mt-SCITE) or in species with extremely high mutation rates.

Irreversibility as a Structural Constraint

Interestingly, while some fields are struggling with the loss of mutations, others are leveraging irreversibility as a powerful signal for accuracy. In lineage tracing using CRISPR-Cas9 barcodes, mutations are strictly one-way: once a segment of DNA is deleted by the Cas9 enzyme, it cannot be "un-deleted."

Jean Feng and Amy D. Willis developed GAPML to exploit this. Because these deletions are irreversible, they provide a "one-way street" that constrains the possible shapes of the evolutionary tree. Similarly, TRIBAL, developed by Leah L. Weber and Mohammed El-Kebir, uses these irreversible genomic deletions to reconstruct B-cell clonal lineages, ensuring that the proposed trees are biologically possible.

Summary: From Perfect Phylogeny to Dollo Parsimony

The shift in the field can be summarized as a move from Perfect Phylogeny (mutations are gained once, never lost) to Dollo Parsimony (mutations are gained once, but can be lost many times).

By incorporating the biological reality of copy-number variations and finite-site mutations, researchers like Jack Kuipers, Niko Beerenwinkel, and Mohammed El-Kebir are creating a more nuanced and accurate map of cancer's evolutionary journey. These models don't just tell us what mutations a tumor has; they tell us the complex story of how those mutations were acquired, competed for dominance, and were sometimes lost along the way.

Papers in this version

Showing 60 of 73 papers.

History & citationVersion 1 · 2026-10-04

Split into other directions · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_relaxing_the_infinite_sites_assumption_i_64388e79,
  title = {Relaxing the Infinite Sites Assumption in Tumor Phylogeny},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/relaxing-the-infinite-sites-assumption-in-tumor-phylogeny/rdr_64388e795ddf4ab78dd6e4555c39be51},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_relaxing_the_infinite_sites_assumption_i_64388e79_v1,
  title = {Relaxing the Infinite Sites Assumption in Tumor Phylogeny},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/relaxing-the-infinite-sites-assumption-in-tumor-phylogeny/rdr_64388e795ddf4ab78dd6e4555c39be51/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}