Lineage Tracing - Usage Guide

Overview

Lineage tracing reads heritable marks (CRISPR/Cas9 scars, static expressed barcodes, or somatic mtDNA mutations) across single cells and asks which cells share which marks to recover an ontogenetic phylogeny or clonal grouping. The central idea: transcriptomic state does not fully predict fate (Weinreb 2020), so lineage information is orthogonal to expression and reconstruction is phylogenetics on error-prone scars where homoplasy and dropout dominate the result.

Prerequisites

pip install cassiopeia-lineage cospar scanpy

Quick Start

Tell your AI agent what you want to do: - "Build a lineage tree from my CRISPR scar character matrix" - "Compare solvers and tell me how robust the topology is" - "Group clonally related cells from mtDNA mutations" - "Recover early fate bias from sparse clonal barcodes"

Example Prompts

Tree Reconstruction

"Reconstruct a lineage tree with Cassiopeia and collapse mutationless edges" "My data has heavy homoplasy and dropout - which solver should I use?" "Run a panel of solvers and compare them with Robinson-Foulds and triplets-correct"

Clonal Dynamics and State

"Integrate clones with transcriptomic state using CoSpar" "Compute early fate bias for monocyte vs neutrophil from clonal data" "Which clones expanded over time and what is their signature?"

Mitochondrial and Native Tracing

"Use mtDNA mutations to group clonally related cells in this human sample" "Why is my mtDNA giving clonal blobs instead of a deep tree?"

Lineage vs State

"Does transcriptomic state predict fate in my system, or is lineage adding new information?" "Validate my state-based trajectory against the lineage tree"

What the Agent Will Do

1. Choose the recording assay by the question (deep topology, clonal state->fate, or retrospective native tissue)
2. Build a character matrix keeping missing (-1) distinct from unedited (0)
3. Assess missingness, informativeness, and barcode-collision risk before reconstruction
4. Reconstruct with a solver matched to scale and to homoplasy/dropout severity
5. Weight indels by formation probability and run a panel of solvers for robustness
6. Compare trees with Robinson-Foulds and depth-stratified triplets-correct
7. Integrate clone with state (CoSpar) to recover hidden fate bias and test state->fate

Tips

• State does not predict fate - Weinreb 2020 showed sisters in the same state diverge; lineage data is orthogonal to expression, not redundant.
• Missing is not unedited - keep the -1 missing state distinct from 0; collapsing them is the most consequential preprocessing error because heritable dropout deletes whole clades and biases topology.
• Homoplasy breaks parsimony - non-uniform Cas9 indels mean unrelated cells share frequent scars; weight indels by probability and use Startle's star-homoplasy model under heavy convergence.
• Run a solver panel - VanillaGreedy/Hybrid/ILP/NJ rarely agree everywhere; agreement is the practical certainty signal. HybridSolver is the scalable default.
• Deep splits are least certain - early near-root splits rest on the fewest, most-overwritten characters yet matter most; report branch support and trust leaf structure more.
• mtDNA gives clonal grouping, not deep trees - low mutation rate, hotspot homoplasy, heteroplasmy drift, and selection limit it; blacklist NUMTs, RNA-edit, and hotspot positions.
• Prospective vs retrospective matters - only prospective barcodes installed before the process can test state->fate; retrospective mtDNA recovers ancestry but cannot establish the preceding state.
• Static barcodes are flat clones - LARRY/CellTag give clonal membership, not division-order topology; use library complexity far above founder number to avoid collisions.
• Integrate clone with state - CoSpar recovers fate bias from sparse clones rather than assuming the manifold encodes fate, but does not build a phylogeny.

Related Skills

single-cell/trajectory-inference - state-based pseudotime/velocity that lineage data tests and corrects single-cell/preprocessing - QC, doublet handling, and normalization upstream of barcode and clone calls single-cell/clustering - cell-type labels annotated onto tree leaves and clones phylogenetics/modern-tree-inference - general phylogenetic inference, parsimony vs ML, and branch support