Biopsy-aware phylogenetic reconstruction from longitudinal single-cell copy number profiles

Abstract

Longitudinal single-cell copy-number profiles provide information about cancer ancestry. Reconstructions that treat all observations as terminal leaves and disregard collection order cannot represent sampled ancestors or use their temporal relationships. We introduce reconstruction methods in the Cancer Tree Biopsy Framework (CTBF) that distinguish a copy-number profile from its separate observations across biopsies. Biopsy-guided reconstruction connects successive samples, defers ambiguous parent assignments, and completes the unresolved ancestry. Partial trees leave unresolved ancestral profiles unspecified. Closed-state trees assign every ancestor an observed profile. For comparison, pooled methods combine the profiles from all biopsies without using collection order. Across 100 independently simulated histories, the biopsy-guided partial methods attained higher mean rooted generalized Robinson--Foulds similarity than pooled neighbor joining and topology-only MEDICC2, using both complete and observed-state-restricted true trees. Under a fixed computational budget, they also attained higher mean similarity than NestedBD-Long at the two evaluated evolutionary depths. Convergence of its Markov chain Monte Carlo runs was not established within that budget. Classical neighbor joining and anticentral top completion gave closely similar partial-tree scores. Closed-state performance depended on evolutionary depth and the evaluation endpoint. An application to three longitudinal KTN152 biopsies demonstrated genome-scale feasibility on patient-derived integer copy-number estimates.