How lauffenburger-lab/lembas works
Project type and purpose
• Type: A scientific software pipeline / open‐source research tool for genome‐scale simulation of intracellular signaling.
• Purpose: Ingest signaling data, reconstruct prior‐knowledge networks, train recurrent neural nets constrained by that network, simulate perturbations (e.g., knockouts), and generate publication‐quality figures.
File structure
• Top‐level boundaries:
– Demo/: Minimal example pipeline.
– Model/: Core Python modeling code, data, pretrained models, scripts to run cross‐validation, ODE simulation, toy networks, etc.
– “Network Construction”/: Scripts (Python & MATLAB) to build the knowledge‐embedded network from OmniPath.
– “TF activities…” and “TF activities Ligand Screen”/: Pre‐ and postprocessing code (MATLAB, R, Python) to estimate transcription‐factor activities and convert into ASN format.
– “Viability data”/: Extraction scripts for CCLE viability data.
– “complexity tests”/: MATLAB performance profiling tests.
• Configuration & orchestration:
– downloadLiteratureData.py: data‐acquisition entry point.
– runSlurm*.sh: cluster orchestration on SLURM for large‐scale training.
• Supporting files: requirements implied in README for Python (PyTorch, numpy, pandas, scipy, matplotlib, seaborn), MATLAB, R (dorothea, DESeq2, limma).
README for architecture insights
• Four main stages described: data download, TF activity estimation, network reconstruction, model training & simulation to produce figures.
• Dependencies are clearly partitioned (Python vs. MATLAB vs. R).
• Figure‐by‐figure breakdown reveals which scripts belong in which stage.
Main components & their interactions
a. Data Ingestion
– downloadLiteratureData.py
– External sources: OmniPath API, public transcriptomics repositories, CCLE.
b. TF Activity Estimation
– MATLAB scripts: parseCondition.m, joinData.m
– R scripts: macrophage.R, run.R (DESeq2, limma, dorothea)
– Python converters: convertToASNFormat.py, constructASNLigandMatrix.py
c. Network Reconstruction
– Python: extractRL.py, extractPKN.py, trimKeggModel.py, trimLigandModel.py, trimMacrophageModel.py
– MATLAB: drawNetwork.m, drawSubNetwork.m
– Outputs: TSV network model files in “Network Construction/model/”
d. Model Implementation & Core Library
– Python modules: Model/bionetwork.py, activationFunctions.py, calculateODEvalues.py, ODEsimulation.py
– Utility: plotting.py
e. Model Training & Evaluation
– Python training scripts: ligandScreenCrossValidation.py, macrophageNetCrossValidation.py, viabilityNet.py, synthNet…
– Cross‐validation & sensitivity: ligandScreenCrossValidationResults.py, macrophageNetCrossValidationResults.py, viabilityNetResults.py
– Slurm scripts: runSlurmMacrophage.sh, runSlurmSynthScreen.sh, etc.
– Pretrained model artifacts: .pt files under Model/CVligandScreen, Model/CVmacrophage, Model/synthNetScreen, etc.
f. Simulation & Analysis
– Toy network experiments: toyNetRecurrent.py, toyNetNFKB.py, pipeNet.py
– ODE fits & forward/backward solves: fitODE.py, displayODEvalues.py, showActivationFunction.py
– Knockout & sensitivity analyses: ligandScreenKO.py, macrophageNetKO.py, synthNetKO.py
g. Visualization & Figure Generation
– Scripts in Model/figures and Demo/src/plotting.py
– Demo: Demo/main.py uses a small network example to illustrate pipeline end‐to‐end.
h. Execution Environment & Orchestration
– Python 3.7 + listed libraries
– MATLAB R2020a for network viz & complexity tests
– R 4.0.3 + dorothea, DESeq2, limma
– SLURM for cluster batch runs