Configuration Reference

This page documents the config.toml configuration file that controls all aspects of the PhosKinTime pipeline.

Location and Loading

The configuration file is config.toml at the project root. It is loaded by two modules:

config_loader.py — the root-level loader. Provides load(mode, section) (LRU-cached) and load_config_toml(). Used by the local pipeline and as the backbone for networkmodel.
networkmodel/config.py — loads config.toml via config_loader.load_config_toml() and exports constants for the global model pipeline.

Technical debt: Both modules load overlapping configuration surfaces from config.toml. They should be unified so networkmodel/config.py entirely delegates to config_loader.py. See the TODO comment in config_loader.py.

Section: `[paths]`

Global project paths. Used by all pipeline stages.

Key	Type	Default	Description
`data_dir`	string	`"data"`	Input data directory (relative to project root)
`results_dir`	string	`"results_model"`	Output results directory
`logs_dir`	string	`"results_model/logs"`	Log files directory
`ode_data_dir`	string	`"data"`	ODE model data directory

Tip: Use distinct results_dir / logs_dir for each optimization mode to avoid overwriting results. For example, use results_kinopt_local for kinopt --mode local.

Section: `[tfopt]`

Controls the TF → mRNA optimization. Consumed by tfopt.local and tfopt.evol.

Key	Type	Default	Description
`input1`	string	`"input1.csv"`	mRNA expression data file
`input3`	string	`"input3.csv"`	TF activity / Rout-Limma data
`input4`	string	`"input4.csv"`	TF-gene network file
`out_file`	string	`"tfopt_results.xlsx"`	Output Excel file
`time_points`	list[float]	`[4,8,15,30,60,120,240,480,960]`	Time grid (minutes)
`lower_bound`	float	`-4.0`	Lower bound for TF coefficients
`upper_bound`	float	`4.0`	Upper bound for TF coefficients
`loss_type`	int	`5`	Loss function (see below)

Loss type codes:

Code	Name
0	MSE
1	MAE
2	Soft L1
3	Cauchy
4	Arctan
5	Elastic Net
6	Tikhonov

Mode-specific overrides in [tfopt.modes.local] and [tfopt.modes.evol] take precedence.

Section: `[kinopt]`

Controls kinase → phosphorylation optimization. Consumed by kinopt.local and kinopt.evol.

Key fields mirror [tfopt] (input files, bounds, loss type, time points). Mode overrides in [kinopt.modes.local] and [kinopt.modes.evol].

Section: `[ode]`

Controls the per-protein ODE fitting pipeline (models/, paramest/, sensitivity/, steady/).

`[ode.bounds]`

Parameter bounds for ODE model fitting.

`[ode.bootstrap]`

Bootstrap sample count for confidence intervals.

`[ode.time]`

Time point grid for ODE integration.

`[ode.fit]`

Fit settings including loss type and composite weighting.

`[ode.sensitivity.morris]`

Morris sensitivity analysis settings (trajectories, levels).

`[ode.plot]`

Plotting options.

`[ode.inputs]` / `[ode.output]`

Input file names and output directory.

Section: `[networkmodel]`

Controls the global network-scale pipeline (networkmodel/). Consumed by networkmodel/config.py.

Core metadata

Key	Type	Example	Description
`app_name`	string	`"Phoskintime-Global"`	Application display name
`version`	string	`"0.4.0"`	Package version
`output_dir`	string	`"results_model_global_..."`	Output directory
`cores`	int	`80`	CPU cores to use (0 = all available)
`seed`	int	`42`	Random seed

Input files

Key	Type	Description
`kinase_net`	string	Kinase-substrate network (`input2.csv`)
`tf_net`	string	TF-gene network (`input4.csv`)
`ms`	string	Mass-spec protein data (`input1.csv`)
`rna`	string	RNA/transcriptomics data (`input3.csv`)
`phospho`	string	Phospho data (can be same as `ms`)
`kinopt`	string	Previous kinopt results (Excel)
`tfopt`	string	Previous tfopt results (Excel)

Optimizer settings

Key	Type	Default	Description
`optimizer`	string	`"pymoo"`	`"optuna"` or `"pymoo"`
`n_gen`	int	`1000`	Number of generations (Pymoo)
`pop`	int	`300`	Population size
`n_trials`	int	`1000`	Number of Optuna trials
`refine`	bool	`false`	Enable iterative refinement pass
`num_refinements`	int	`0`	Number of refinement iterations
`hyperparam_scan`	bool	`false`	Enable Optuna hyperparameter scan

Loss and regularization

Key	Type	Default	Description
`loss`	int	`0`	Loss type (0=MSE, 1=Huber, 2=Pseudo-Huber, ...)
`lambda_prior`	float	`0.1`	Weight for prior adherence
`lambda_protein`	float	`1.0`	Weight for protein fit error
`lambda_rna`	float	`1.0`	Weight for RNA fit error
`lambda_phospho`	float	`1.0`	Weight for phospho fit error

Data flags

Key	Type	Default	Description
`normalize_fc_steady`	bool	`false`	Normalize data to t=0
`use_initial_condition_from_data`	bool	`false`	Use data values as t=0 state
`scaling_method`	string	`"raw"`	Data scaling: `raw`, `fc_start`, `robust_fc`, `max_scale`, etc.

Weighting methods

Key	Options	Default
`weighting_method_protein`	`uniform`, `linear_early`, `exp_late`, etc.	`"uniform"`
`weighting_method_rna`	same	`"uniform"`

Sensitivity analysis

Key	Type	Default	Description
`sensitivity_analysis`	bool	`true`	Enable post-optimization sensitivity
`sensitivity_perturbation`	float	`0.05`	Perturbation fraction
`sensitivity_trajectories`	int	`100`	Morris trajectories
`sensitivity_levels`	int	`40`	Morris grid levels
`sensitivity_metric`	string	`"total_signal"`	Metric: `total_signal`, `mean`, `variance`, `l2_norm`

`[networkmodel.timepoints]`

Key	Description
`protein`	Time grid for protein data (minutes)
`phospho_protein`	Time grid for phospho data
`rna`	Time grid for RNA data

`[networkmodel.bounds]`

Kinetic parameter bounds for the global ODE. Each entry is [min, max].

Parameter	Meaning
`c_k`	Kinase activity multiplier
`A_i`	Basal mRNA transcription rate
`B_i`	mRNA degradation rate
`C_i`	Translation rate
`D_i`	Protein deactivation rate
`Dp_i`	Phospho-site dephosphorylation rate
`E_i`	Transcriptional efficacy
`tf_scale`	TF scaling factor

`[networkmodel.models]`

Key	Values
`available_models`	`["distributive", "sequential", "combinatorial", "saturation"]`
`default_model`	`"distributive"`

`[networkmodel.solver]`

Key	Default	Description
`absolute_tolerance`	`1e-8`	ODE absolute tolerance
`relative_tolerance`	`1e-8`	ODE relative tolerance
`max_timesteps`	`200000`	Maximum ODE integration steps
`use_custom_solver`	`false`	Use custom Numba-based solver instead of SciPy

Which module consumes which section

Section	Consumer
`[paths]`	`config_loader.py`, all subpackages
`[tfopt]`	`tfopt.local`, `tfopt.evol`
`[kinopt]`	`kinopt.local`, `kinopt.evol`
`[ode]`	`models/`, `paramest/`, `sensitivity/`, `steady/`
`[networkmodel]`	`networkmodel/config.py`, `networkmodel/runner.py`

Configuration Reference

Location and Loading

Section: [paths]

Section: [tfopt]

Section: [kinopt]

Section: [ode]

[ode.bounds]

[ode.bootstrap]

[ode.time]

[ode.fit]

[ode.sensitivity.morris]

[ode.plot]

[ode.inputs] / [ode.output]

Section: [networkmodel]