PLASTRO

PLASTRO is a Python package for simulating and analyzing cellular plasticity in single-cell data. It provides comprehensive tools for studying how cells transition between different phenotypic states and how these transitions relate to lineage relationships.

Key Features

• Plasticity Simulation: Random walk plasticity and cluster-based transitions
• Lineage Tracing Integration: CRISPR-based lineage tracing simulation with Cassiopeia
• PLASTRO Score: Novel overlap-based metrics for quantifying cellular plasticity.
• Phylogenetic Analysis: Neighbor-joining tree construction from single-cell data
• Data Simulation: Generate realistic synthetic datasets with branching differentiation
• High Performance: Optimized overlap computation (10-100x speedup over naive methods)

Installation

Quick Install

PLASTRO requires pybind11 to be installed first for building essential dependencies:

pip install pybind11
pip install plastro

From TestPyPI (Latest Development Version)

pip install pybind11
pip install --index-url https://test.pypi.org/simple/ --extra-index-url https://pypi.org/simple/ plastro

Development Install

git clone https://github.com/dpeerlab/PLASTRO.git
cd PLASTRO
pip install pybind11
pip install -e .

Conda Install (coming soon)

conda install -c conda-forge plastro

Quick Start

Basic PLASTRO Score Computation

import plastro
import pandas as pd

# Load your single-cell data and lineage tracing data
character_matrix = pd.read_csv('character_matrix.csv', index_col=0)
adata = plastro.load_data('single_cell_data.h5ad')

# Compute Gini-based plasticity scores (recommended)
gini_scores = plastro.PLASTRO_score(
    character_matrix=character_matrix,
    ad=adata,
    flavor='gini',
    latent_space_key='X_dc'  # or 'X_pca', 'X_umap'
)

print(f"Mean Gini plasticity score: {gini_scores['Gini_Index'].mean():.3f}")

Generate Synthetic Data with Plasticity

# Create synthetic single-cell dataset
n_leaves = 8
sample_res = 50  
n_dim = 20

# Generate branching structure
sample_structure = plastro.create_random_binary_tree(n_leaves, sample_res)
full_simulated_ad = plastro.generate_ad(sample_structure, n_dim)

# Subset to terminal branches
ad = plastro.subset_to_terminal_branches(full_simulated_ad)

# Simulate lineage tracing
cass_tree = plastro.simulate_lineage_tracing(
    sim_ad=full_simulated_ad, 
    terminal_ad=ad,
    latent_space_key='X_dc'
)

Simulate Cellular Plasticity

# Random walk plasticity
plastic_cells = {'6': 0.3, '5': 0.2}  # 30% of cluster 6, 20% of cluster 5
walk_lengths = {'6': 500, '5': 1000}
plastic_walk_ad = plastro.random_walk_plasticity(
    full_simulated_ad, ad, plastic_cells, walk_lengths
)

# Cluster switch plasticity  
destination_clusters = {
    '11': {'destination': '7', 'proportion': 0.4},
    '6': {'destination': '10', 'proportion': 0.2}
}
plastic_leiden_ad = plastro.cluster_switch_plasticity(
    full_simulated_ad, ad, destination_clusters, column='leiden'
)

Documentation & Examples

Example Notebooks

Complete example workflows are available in docs/notebooks/:

1. Plasticity Simulation Example:

   • Generate synthetic single-cell data with branching differentiation
   • Simulate CRISPR-based lineage tracing
   • Apply random walk and cluster switch plasticity
   • Visualize phenotypic changes

2. PLASTRO Overlap Analysis:

   • In-depth analysis of lineage-phenotype relationships
   • Detailed explanation of overlap computation methods
   • Interpretation of PLASTRO scores

Key Documentation Sections

• Installation Guide: Detailed installation instructions
• API Reference: Complete function documentation
• Tutorials: Step-by-step guides

Core API

Main Functions

# PLASTRO Score Computation
plastro.PLASTRO_score(character_matrix, ad, flavor='gini')
plastro.PLASTRO_overlaps(character_matrix, ad, maximum_radius=500)

# Plasticity Simulation
plastro.random_walk_plasticity(full_ad, subset_ad, plastic_cells, walk_lengths)
plastro.cluster_switch_plasticity(full_ad, subset_ad, destination_clusters)

# Data Generation
plastro.create_random_binary_tree(n_leaves, sample_res)
plastro.generate_ad(sample_structure, n_dim)
plastro.simulate_lineage_tracing(sim_ad, terminal_ad)

# Distance Calculations  
plastro.euclidean_distance(coordinates)
plastro.cosine_distance(coordinates)
plastro.manhattan_distance(coordinates)
plastro.archetype_distance(data, archetypes)

# Phylogenetic Analysis
plastro.neighbor_joining(distance_matrix, outgroup=None)

Core Modules

• plastro.overlap: PLASTRO score computation and overlap analysis
• plastro.plasticity: Cellular plasticity simulation methods
• plastro.lineage_simulation: CRISPR-based lineage tracing simulation
• plastro.phenotype_simulation: Synthetic single-cell data generation
• plastro.phylo: Phylogenetic tree construction and analysis

Complete Example Workflow

import plastro
import pandas as pd

# 1. Generate synthetic data (or load your own)
sample_structure = plastro.create_random_binary_tree(n_leaves=8, sample_res=50)
full_simulated_ad = plastro.generate_ad(sample_structure, n_dim=20)
ad = plastro.subset_to_terminal_branches(full_simulated_ad)

# 2. Simulate lineage tracing
cass_tree = plastro.simulate_lineage_tracing(
    sim_ad=full_simulated_ad,
    terminal_ad=ad,
    latent_space_key='X_dc'
)
character_matrix = cass_tree.character_matrix

# 3. Simulate plasticity
plastic_cells = {'6': [cell1, cell2, cell3]}  # Specific cells to make plastic
walk_lengths = {'6': 500}
plastic_ad = plastro.random_walk_plasticity(
    full_simulated_ad, ad, plastic_cells, walk_lengths
)

# 4. Compute PLASTRO scores
original_scores = plastro.PLASTRO_score(
    character_matrix, ad, flavor='gini'
)
plastic_scores = plastro.PLASTRO_score(
    character_matrix, plastic_ad, flavor='gini'
)

# 5. Compare plasticity
print(f"Original mean Gini score: {original_scores['Gini_Index'].mean():.3f}")
print(f"Plastic mean Gini score: {plastic_scores['Gini_Index'].mean():.3f}")

Dependencies

Core requirements:

• Python ≥ 3.10
• pybind11 ≥ 2.6.0 (required for building graph-walker)
• graph-walker ≥ 1.0.6 (essential for random walk functionality)
• NumPy ≥ 1.20.0
• Pandas ≥ 1.3.0
• SciPy ≥ 1.7.0
• scikit-learn ≥ 1.0.0
• scikit-bio ≥ 0.5.7 (for robust neighbor-joining trees)
• NetworkX ≥ 2.6.0
• matplotlib ≥ 3.4.0
• scanpy ≥ 1.8.0
• anndata ≥ 0.8.0
• ete3 ≥ 3.1.2 (for phylogenetic tree manipulation)
• tqdm ≥ 4.60.0
• seaborn ≥ 0.11.0
• icecream ≥ 2.1.0

Optional dependencies:

• cassiopeia-lineage (for advanced lineage tracing simulation)
• igraph (for Leiden clustering in phenotype simulation)

Data Requirements

PLASTRO works with:

• AnnData objects containing single-cell data with dimensionality reduction coordinates
• Character matrices (pandas DataFrame) from CRISPR lineage tracing with cells as rows
• Distance matrices for lineage and phenotypic relationships
• Cluster annotations (leiden, louvain, etc.) for cluster-based plasticity

License

This project is licensed under the MIT License - see the LICENSE file for details.

Support

• Documentation: plastro.readthedocs.io
• Issues: GitHub Issues
• Discussions: GitHub Discussions

---

PLASTRO - Comprehensive analysis of cellular plasticity in single-cell data