pykinbiont

Python interface for KinBiont.jl — a Julia package for model-based analysis of microbial kinetics data.

Requirements

• Python ≥ 3.11
• Julia ≥ 1.10 (installed separately — julialang.org)

Installation

pip install pykinbiont

Or with uv:

uv add pykinbiont

---

Two modes of operation

Mode 1 — Managed environment (default)

juliacall creates and manages its own isolated Julia environment with Kinbiont installed automatically. No local KinBiont.jl clone needed.

First run is slow (Julia downloads and installs Kinbiont and its dependencies). Subsequent runs are fast because the environment is cached.

import pykinbiont

result = pykinbiont.fitting.fitting_one_well_log_lin(data, "A1", "exp1")

---

Mode 2 — Existing local KinBiont.jl environment

If you already have KinBiont.jl installed locally with all dependencies resolved, you can point pykinbiont directly at that Julia project. This skips the managed environment and reuses what you already have.

One-time setup — add PythonCall to your KinBiont.jl project:

julia --project=/path/to/KinBiont.jl -e 'using Pkg; Pkg.add("PythonCall")'

Then configure pykinbiont to use that path (persisted across sessions):

import pykinbiont

pykinbiont.configure("/path/to/KinBiont.jl")  # run once

From then on, just import and use — no reinstallation:

import pykinbiont

result = pykinbiont.fitting.fitting_one_well_log_lin(data, "A1", "exp1")

You can also set the path via environment variable before launching Python, which avoids calling configure() entirely:

export JULIA_PROJECT=/path/to/KinBiont.jl

---

API

pykinbiont.configure(project_path)

Persist a local KinBiont.jl path for Mode 2. Saved to ~/.config/pykinbiont/config.json. Must be called before the first fitting or conversion call (i.e. before Julia starts).

pykinbiont.init(project_path=None)

Explicitly start Julia and load Kinbiont. Optional — all functions trigger this automatically on first use.

---

Conversion utilities — pykinbiont.convert

import numpy as np
import pandas as pd
import pykinbiont

# numpy array (2, N) or DataFrame → Julia Matrix{Float64}
jl_mat = pykinbiont.convert.to_julia_matrix(np_array)
jl_mat = pykinbiont.convert.to_julia_matrix(df)  # DataFrame with columns [time, OD]

# Julia array → numpy
arr = pykinbiont.convert.from_julia_array(jl_mat)

# Julia matrix → DataFrame
df = pykinbiont.convert.julia_matrix_to_dataframe(jl_mat, columns=["time", "OD"])

Input data layout for time-series: shape (2, N) where row 0 is time and row 1 is OD. A DataFrame is expected to have time in the first column and OD in the second.

---

Fitting — pykinbiont.fitting

fitting_one_well_log_lin

Log-linear fit of the exponential growth phase for a single growth curve.

import numpy as np
import pykinbiont

data = np.array([
    [0.0, 0.5, 1.0, ..., 7.0],   # time points
    [0.01, 0.012, 0.015, ..., 0.65],  # OD values
])

result = pykinbiont.fitting.fitting_one_well_log_lin(
    data,
    name_well="A1",
    label_exp="exp1",
    # optional parameters:
    type_of_smoothing="rolling_avg",  # "rolling_avg", "lowess", or "NO"
    pt_avg=7,                         # rolling average window (needs ≥ pt_avg points)
    pt_smoothing_derivative=7,        # growth rate estimation window
    pt_min_size_of_win=7,             # minimum exponential window size
    threshold_of_exp=0.9,             # quantile threshold for exp phase detection
)

Minimum number of data points required: pt_avg + pt_smoothing_derivative (default: 14). For small datasets reduce these parameters, e.g. pt_avg=3, pt_smoothing_derivative=3, pt_min_size_of_win=3.

Returns a LogLinResult dataclass:

Field	Type	Description
method	str	Always "Log-lin"
params	pd.Series	14 named fitting parameters (see below)
fit	pd.DataFrame	Columns time, log_fit over the exponential window
smoothed	pd.DataFrame	Columns time, OD — smoothed input curve
confidence_band	np.ndarray	95% confidence band over the fitted window

params fields:

Name	Description
label_exp	Experiment label
name_well	Well name
t_start_exp	Start time of exponential window
t_end_exp	End time of exponential window
t_max_gr	Time of maximum specific growth rate
gr_max	Maximum specific growth rate
growth_rate	Fitted growth rate (log-linear slope)
sigma_growth_rate	Standard error of growth rate
doubling_time	log(2) / growth_rate
doubling_time_lower_95	Doubling time lower 95% bound
doubling_time_upper_95	Doubling time upper 95% bound
intercept	Log-linear fit intercept
sigma_intercept	Standard error of intercept
pearson_r	Pearson correlation coefficient of the fit

All numeric fields are NaN if the exponential window could not be detected.