openpkflow.ivivc¶

In Vitro-In Vivo Correlation (Level A): deconvolution, convolution prediction, Levy plots, and FDA predictability assessment.

Public API¶

Symbol	Type	Description
`wagner_nelson(times, concentrations, ...)`	function	1-cmt Wagner-Nelson deconvolution; returns cumulative F_a
`loo_riegelman(times, concentrations, ...)`	function	2-cmt Loo-Riegelman deconvolution; requires kel, k12, k21
`convolution_predict(dissolution_times, dissolution_pct, ...)`	function	Numerical convolution of dissolution input with IV UIR
`levy_plot_data(in_vitro_times, in_vitro_fraction, in_vivo_fraction)`	function	Compute Levy plot regression data (slope, R2, residuals)
`ivivc_predictability(observed_cmax, predicted_cmax, observed_auc, predicted_auc)`	function	FDA %PE assessment; returns pass/fail per FDA 1997 criteria
`IVIVCResult`	dataclass	Comprehensive IVIVC result; `.summary()`, `.plot()`, `.report()`, `.to_dict()`
`IVIVCStudy`	class	High-level orchestrator: `.analyze()` returns `IVIVCResult`

Deconvolution methods¶

Method	Compartment	Required parameters
Wagner-Nelson	1-cmt	`kel` (or IV UIR data to estimate it)
Loo-Riegelman	2-cmt	`kel` (k10), `k12`, `k21`

Both methods return a cumulative fraction absorbed array (F_a) as numpy ndarray.

Predictability criteria (FDA 1997)¶

Criterion	Threshold
%PE Cmax	<= 15% per formulation
%PE AUCinf	<= 15% per formulation
Mean absolute %PE	<= 10% across formulations

All three must pass for overall IVIVC predictability.

IVIVCResult fields¶

Field	Type	Description
`method`	`str`	Deconvolution method used
`times`	`np.ndarray`	In vivo plasma sampling times
`concentrations`	`np.ndarray`	Observed plasma concentrations
`fa`	`np.ndarray`	Cumulative fraction absorbed
`levy_plot`	`dict`	Slope, intercept, R-squared, residuals
`ivt_times`	`np.ndarray`	In vitro dissolution time points
`ivt_fraction`	`np.ndarray`	Cumulative fraction dissolved
`predicted_times`	`np.ndarray`	Convolution-predicted plasma times
`predicted_concs`	`np.ndarray`	Convolution-predicted concentrations
`predictability`	`dict`	%PE_Cmax, %PE_AUC, passes flags, overall_pass

Report formats¶

.report("out.html") -- HTML with 4-panel figure .report("out.md") -- Markdown .report("out.pdf") -- PDF (requires openpkflow[reports]) .report("out.docx") -- Word (requires openpkflow[reports])

Notes¶

Wagner-Nelson is valid for 1-compartment kinetics only. If the drug shows distribution-phase behaviour, use Loo-Riegelman with the 2-cmt microconstants.
The convolution predictor uses numerical trapezoidal convolution with a uniform time grid spanning the UIR range. Input rate is computed from the derivative of the interpolated dissolution curve.
Levy plots filter to the 0.05-0.95 fraction range for robust regression, per FDA convention.