How gifford-lab/prescient works
PRESCIENT is a generative model for single-cell time-series. The principal workflow prepares expression data, trains a stochastic neural model, then simulates trajectories; perturbation analysis compares simulations from altered and unaltered expression. The graph also retains documented legacy fate/interpolation analysis modules. Training and simulation relationships are supported by excerpts or the SOURCE INDEX; the README gives only a high-level product description, so unsampled capabilities and connections are not expanded.
Main entry points: model.py Generator and loss · __main__.py Command entry · perturbation_analysis.py Perturbation analysis · run.py Training loop
Components 15
Data preparation
- Data processor
prescient/commands/process_data.py - Prepared data
Model training
- Training command
prescient/commands/train_model.py - Training loop
prescient/train/run.py - Generator and loss
prescient/train/model.py - Training utilities
prescient/train/util.py - Model checkpoints
Simulation analysis
- Trajectory command
prescient/commands/simulate_trajectories.py - Trajectory simulator
prescient/simulate/sim.py - Perturbation analysis
prescient/commands/perturbation_analysis.py - Gene perturbation
prescient/perturb/pert.py
Legacy analysis
- Fate analysis
depr/veres.py - Interpolation analysis
depr/weinreb.py
Other components
- Researcher
- Command entry
prescient/__main__.py
Connections 24
- Researcher → Command entry: runs commands
prescient/__main__.py - Command entry → Data processor: dispatches to
- Command entry → Training command: dispatches to
- Command entry → Trajectory command: dispatches to
- Command entry → Perturbation analysis: dispatches to
- Researcher → Data processor: supplies expression
prescient/commands/process_data.py - Data processor → Prepared data: saves data
prescient/commands/process_data.py - Researcher → Training command: provides data
prescient/commands/train_model.py - Training command → Training loop: starts training
prescient/commands/train_model.py - Training loop → Generator and loss: optimizes
prescient/train/run.py - Training loop → Training utilities: uses helpers
prescient/train/run.py - Training loop → Model checkpoints: saves weights
prescient/train/run.py - Researcher → Trajectory command: requests simulation
prescient/commands/simulate_trajectories.py - Trajectory command → Generator and loss: loads model
prescient/commands/simulate_trajectories.py - Trajectory command → Trajectory simulator: invokes
prescient/commands/simulate_trajectories.py - Trajectory simulator → Generator and loss: steps model
prescient/simulate/sim.py - Researcher → Perturbation analysis: requests comparison
prescient/commands/perturbation_analysis.py - Perturbation analysis → Gene perturbation: transforms expression
prescient/commands/perturbation_analysis.py - Perturbation analysis → Generator and loss: loads model
prescient/commands/perturbation_analysis.py - Perturbation analysis → Trajectory simulator: runs both conditions
prescient/commands/perturbation_analysis.py - Researcher → Fate analysis: runs legacy analysis
depr/veres.py - Researcher → Interpolation analysis: runs legacy analysis
depr/weinreb.py - Fate analysis → Generator and loss: uses training package
depr/veres.py - Interpolation analysis → Generator and loss: uses training package
depr/weinreb.py