Nathaniel Linden-Santangeli

University of California San Diego 2024

Repository for: Increasing certainty in systems biology models using Bayesian multimodel inference

Authors: Nathaniel Linden-Santangeli, Jin Zhang, Boris Kramer, Padmini Rangamani

Date: June 2024

This repository provides all code and data that is necessary to reproduce all results shown in the manuscript.

Instructions to reproduce results and figures

Source code is located in the src/MAPK/ directory and results and data are located in the results/ directory.

The excel file source_data.xlsx contains all results reported in the figures and tables of the manuscript. Each associated script has code to generate the figure panels and tables in the manuscript and export the source data to the results/ directory.

Local Identifiability analysis

The Julia code to run identifiability analysis for each model is located in src/MAPK/identifiability/. Each script runs the analysis for every model independently.

Supplemental Figure S1 (Global sensitivity analysis)

1. Run src/MAPK/gsa/gsa_sampling.sh to generate samples of model outputs.
2. Run src/MAPK/gsa/gsa_analyze.py to perform Morris analysis and make plots.

Synthetic data generation and Keyes et al. eLife. 2020. data normalization.

All synthetic data generation and preprocessing occurs in src/MAPK/process_data.ipynb.

Figure 3 and Supplemental Figures S2, S3 (Keyes et al. eLife. 2020. data inference/MMI)

0. (optional to rerun SMC) Run src/MAPK/param_est/Keyes_CYTO_inference.sh and src/MAPK/param_est/Keyes_PM_inference.sh without the --skip_sample flag on each call. Rerunning with this flag enabled regenerates posterior predictive samples from previous SMC results
1. Run src/MAPK/param_est/Keyes_analyze.py and src/MAPK/param_est/Keyes_data_len_analyze.py to plot posterior predictive trajectories
2. Run src/MAPK/multimodel_inference/Keyes_MMI.ipynb to perform MMI, run error analysis, and make plots
3. Run src/MAPK/param_est/Keyes_data_params_compDiff_analyze.ipynb to plot marginal eCDFs

Supplemental Figures S4, S6 (Synthetic dose-response data)

Figure S4

0. (optional to rerun SMC) Run src/MAPK/param_est/HF96_DR_inference.sh without the --skip_sample flag on each call. Rerunning with this flag enabled regenerates posterior predictive samples from previous SMC results
1. Run src/MAPK/param_est/HF96_DR_analyze.py to plot posteriors
2. Run src/MAPK/multimodel_inference/HF96_DR_MMI.ipynb to perform MMI, run error analysis, and make plots

Figure S6

1. Run src/MAPK/multimodel_inference/HF95_DR_model_perturb.ipynb for panels A-E
2. Run src/MAPK/multimodel_inference/HF95_DR_model_combinatorics.ipynb for panels F,G

Figure 4 and Supplemental Figure S7 and S8 (Keyes et al. eLife. 2020. data shortening and quality reduction

data shortening

0. (optional to rerun SMC) Run src/MAPK/param_est/Keyes_CYTO_10min_inference.sh, src/MAPK/param_est/Keyes_CYTO_20min_inference.sh, src/MAPK/param_est/Keyes_CYTO_30min_inference.sh, src/MAPK/param_est/Keyes_PM-10min_inference.sh, src/MAPK/param_est/Keyes_PM-20min_inference.sh, and src/MAPK/param_est/Keyes_PM-30min_inference.sh without the --skip_sample flag on each call. Rerunning with this flag enabled regenerates posterior predictive samples from previous SMC results
1. Run src/MAPK/param_est/Keyes_data_len_analyze.py and src/MAPK/param_est/Keyes_analyze.py to plot posterior predictive trajectories
2. Run src/MAPK/multimodel_inference/Keyes_MMI.ipynb to perform MMI, run error analysis, and make plots

data quality reduction

1. [Requires SMC rerun--possible long runtime] Run src/MAPK/param_est/Keyes_data_quality_inference.py
2. Run src/MAPK/multimodel_inference/Keyes_data_quality_MMI.ipynb

Figure 5 and Supplemental Figure S10 (Keyes et al. eLife. 2020. Rap1/ERK negative feedback modifications)

0. (optional to rerun SMC) Run src/MAPK/param_est/Keyes_Rap1_negFeed_inference.sh without the --skip_sample flag on each call. Rerunning with this flag enabled regenerates posterior predictive samples from previous SMC results
1. Run src/MAPK/param_est/Keyes_Rap1_negFeed_analyze.py to plot posterior predictive trajectories
2. Run src/MAPK/multimodel_inference/Keyes_Rap1_negFeed_MMI.ipynb to perform MMI, run error analysis, and make plots

PSIS-LOO-CV vs. LOO-CV

The notebook src/MAPK/LOOCV_analyze.ipynb compares PSIS-LOO-CV to LOO-CV.

Dependencies

Julia

Julia version 1.10.0 is used for local identifiability analysis.

Packages

• StructuralIdentifiability.jl - v0.5.6
• DifferentialEquations.jl - v7.10.0

Python

Python version 3.11.6 is used for all other analysis.

Packages

• Jax - 0.4.25 (CPU computing)
• Jaxlib - 0.4.20 (CPU computing)
• numpy - 1.26.1
• pandas - 2.1.2
• matplotlib - 3.8.1
• seaborn - 0.13.0
• PyMC - 5.10.4
• numpyro - 0.13.2
• pytensor - 2.18.6
• ArViz - 0.16.1
• Preliz - 0.3.6
• Diffrax - 0.5.0
• Optimistix - 0.0.6
• Lineax - 0.0.4
• tqdm - 4.66.1
• func_timeout - 4.3.5