melafit 0.6.0

Python package for high-precision circadian melatonin profile analysis

melafit

Python package for high-precision circadian melatonin profile analysis. Features a variety of baseline cosine functions for curve fitting (Van Someren & Nagtegaal, 2007) and a robust cost function for superior convergence, even with sparse data (Gabel et al., 2017).

Overview

melafit is a Python package designed for high-precision analysis of 24-hour melatonin secretion. While standard cosinor or harmonic analyses fail to capture the physiological nuances of the melatonin "wave," melafit implements several baseline cosine functions including bimodal, skewed and bimodal-skewed modifications. This approach accounts for the characteristic baseline, asymmetry and dual peaks often seen in high-resolution circadian melatonin data.

Furthermore, the library utilizes a specialized cost function developed to overcome common optimization hurdles (trivial all-zero solutions), ensuring stable convergence even when working with sparse or incomplete time series.

Its extensible architecture allows users to define custom waveform functions, cost functions, and result classes, enabling adaptation to diverse experimental paradigms beyond the built-in models.

Key Features

• Bimodal Waveform Fitting: Implementation of the Van Someren & Nagtegaal (2007) model for superior physiological accuracy.
• Optimized Convergence: Leverages the robust cost function described in Gabel et al. (2017) to ensure reliable fits across diverse datasets.
• Sparse Data Support: Capable of reconstructing full profiles and estimating circadian phase from limited data points, as well as determining dim light melatonin onset (DLMO) with partial data.
• Research-Ready: Direct derivation of Amplitude, DLMOn, DLMOff, Midpoint, Area and COG markers from continuous, fitted waveforms.

Installation

melafit is available on PyPI and can be installed with pip. However, installing directly into your system Python environment without a virtual environment is strongly discouraged, as it may cause conflicts with other packages. The recommended approach is to use Miniforge as the package and environment manager to create a dedicated virtual environment, as described below. As a community-maintained tool that defaults to the conda-forge channel, Miniforge carries no commercial licensing restrictions.

Standard installation

Download file melafit.yml to a directory of your choice (YOUR-DIRECTORY). Navigate to the directory, create and activate the conda environment:

cd YOUR-DIRECTORY
conda env create -f melafit.yml
conda activate melafit

The environment configuration file melafit.yml explicitly uses conda-forge as the sole package channel and excludes all other channels, ensuring reproducibility, avoiding potential conflicts between packages from different channels, and eliminating any commercial licensing concerns. melafit itself is installed from PyPI via pip as part of the environment setup. This will create a fully functional analysis environment, including all supporting packages (numpy, scipy, pandas, openpyxl and matplotlib).

Developer installation

Click to expand

If you intend to follow the development closely or contribute to the package, clone the repository first to a dedicated directory YOUR-DIRECTORY. Navigate to it and clone the repository as follows:

cd YOUR-DIRECTORY
git clone https://github.com/vitaliy-ch25/melafit.git
cd melafit

Then create and activate the conda environment using the developer configuration file melafit-dev.yml, which installs melafit directly from the cloned directory in editable mode:

conda env create -f melafit-dev.yml
conda activate melafit

With an editable install, any changes to the source code in the cloned directory take effect immediately without reinstalling the package.

Updating

Standard update

Download the latest melafit.yml to YOUR-DIRECTORY. Navigate to it and run the update command as follows:

cd YOUR-DIRECTORY
conda env update -f melafit.yml --prune

This updates both the dependencies and melafit itself to the latest released version.

Developer update

Click to expand

Navigate to the cloned repository directory and pull the latest version from the main branch:

cd YOUR-DIRECTORY/melafit
git pull

The editable install picks up the changes immediately. If dependencies in melafit-dev.yml have changed, also run:

conda env update -f melafit-dev.yml --prune

This updates both the dependencies and the melafit package itself to the latest version.

Getting Started

Code examples and some dummy data demonstrating melatonin profile curve fitting with this package are included in ./examples/ and ./data/. Copy sample scripts and datasets to your working directory and start from there. If you have performed the steps above as described, your script will 'see' all the required packages from any location. Simply make sure to use the virtual environment melafit you created.

Minimal example — fit a single participant and compute area/COG

Click to expand

"""
Determine phase markers from the full melatonin profile of one participant 
via fitting the BSBCF function.
"""

import os
import matplotlib.pyplot as plt
from matplotlib import dates
import melafit as mf

# Read full profile data from Excel spreadsheet
data = mf.read_data("./data/dummy_data_full.xlsx")

# Prepare results directory and collector
result_path = "./results/one_fit/"
os.makedirs(result_path, exist_ok=True)
collector = mf.ResultsCollector()

participant = 1

# Prepare data for the participant
p_data = mf.prepare_part_data(data, participant)

# Fit curve and compute resampled waveform
res = mf.fit(p_data.Timestamp, p_data.Mel, mf.bsbcf)
resampled_t = mf.gen_time_range(p_data.Timestamp, step="1min")
resampled_f = mf.bsbcf(t=resampled_t, p=res)

# Compute area and COG
ac = mf.area_cog(resampled_t, resampled_f)

# Collect all results for this participant
meta = mf.SessionInfo(p_data)
collector.add(meta, res, ac)

# Print summary
print(meta)
print(res)
print(ac)

# Visualize results
title_str = (f"{meta}, {ac}, R²={res.r2:.3f}")

plt.close("all")
plt.figure(figsize=(12, 5))
plt.scatter(p_data.Timestamp, p_data.Mel, c='b')
plt.plot(resampled_t, resampled_f, 'g')
plt.xlabel("Time, hh:mm")
plt.gca().xaxis.set_major_formatter(dates.DateFormatter('%H:%M'))
plt.ylabel("Concentration, pg/ml")
plt.title(title_str)
plt.legend(["Melatonin data", "BSBCF curve"])
plt.savefig(result_path + f"mel_data_{participant}_BSBCF.png")

# Show plot for 1 second
plt.pause(1.0)

# Save results to Excel file
collector.save(result_path, "results_one_fit_BSBCF.xlsx")

Running this code contained in example_one_fit.py produces the following text output in the terminal

Participant=1, 2026-03-19 12:00 – 2026-03-20 12:00
Fitted function: BSBCF, parameters: phi=0.108, b=1.650, H=65.766, c=0.235, v=0.185, m=0.198, R²=0.9978
Area=16.970, COG=02:08

and the following figure is displayed with a fitted BSBCF waveform and the data it was fitted to. Besides that, the results are stored to an Excel table results_one_fit_BSBCF.xlsx under ./results/one_fit/.

Data preparation

Follow the Excel table format and column naming conventions as in ./data/:

• Participant for study participant ID
• Date for dates of the respective samples
• Time for sample timestamps
• Mel for melatonin level values

Scientific Foundations

If you use melafit in your research, please cite the following foundational publications:

Human-Readable

1. Van Someren, E. J., & Nagtegaal, E. (2007). Improving melatonin circadian phase estimates. Sleep Medicine, 8(6), 590-601.
2. Gabel, V., et al. (2017). Differential impact in young and older individuals of blue-enriched white light on circadian physiology and alertness during sustained wakefulness. Scientific Reports, 7, 7620.

BibTeX

Click to expand

@article{vansomeren2007,
  title={Improving melatonin circadian phase estimates},
  author={Van Someren, Eus JW and Nagtegaal, Elsbeth},
  journal={Sleep Medicine},
  volume={8},
  number={6},
  pages={590--601},
  year={2007},
  publisher={Elsevier}
}

@article{gabel2017,
  title={Differential impact in young and older individuals of
         blue-enriched white light on circadian physiology and alertness
         during sustained wakefulness},
  author={Gabel, Virginie and Reichert, Carolin F and Maire, Micheline
          and Schmidt, Christina and Schlangen, Luc JM
          and Kolodyazhniy, Vitaliy and Garbazza, Corrado
          and Cajochen, Christian and Viola, Antoine U},
  journal={Scientific Reports},
  volume={7},
  pages={7620},
  year={2017},
  publisher={Nature Publishing Group}
}

If there is no associated publication on melafit yet, please cite the package directly using the following reference:

Kolodyazhniy, V., & Cajochen, C. (2026). melafit: Python package for 
high-precision circadian melatonin profile analysis (Version x.y.z). 
[Computer software]. GitHub: https://github.com/vitaliy-ch25/melafit. 
DOI: https://doi.org/10.5281/zenodo.20573831

Authors

• Vitaliy Kolodyazhniy – Lead Developer
• Christian Cajochen – Scientific Lead

Revision History

Click to expand

v0.6.0 - Resampling function

• New function resample_t for resampling input data to the given time axis

v0.5.3 - Bugfixes

• Fixed nan handling in abs_threshold() and func_defaults()
• Added respective unit tests
• Fixed related identifiers in .zenodo.json

v0.5.2 - Zenodo DOI and citation metadata

• Zenodo DOI registered; DOI badge added to README
• CITATION.cff added with full author metadata, ORCID identifiers and concept DOI for consistent citation across all releases
• .zenodo.json added for automatic Zenodo metadata population on each future release, including author names, affiliations, ORCID identifiers and related publication DOIs
• Package title updated to "melafit: Python package for high-precision circadian melatonin profile analysis" across CITATION.cff, .zenodo.json and the suggested citation reference in README
• Package citation reference in README updated to use the permanent DOI
• Corrects the PyPI publishing failure in v0.5.1 caused by a versioning error in _version.py; v0.5.1 was released on GitHub and Zenodo only
• Fixed a rendering issue inside the collapsible sections of the README on GitHub Pages
• Improved the Overview section of the README and added a github link to the recommended reference
• No changes to package functionality

v0.5.1 - Documentation and example fix

• Fixed a rendering issue in the README where content following collapsible sections was truncated on GitHub Pages
• In example_one_fit.py, plt.pause(1.0) instead of plt.waitforbuttonpress() to avoid issues with the Spyder IDE when inline figures are configured
• No changes to package functionality

v0.5.0 - API improvements and bugfixes

• New dlmo() function in markers.py computes DLMO from the rising slope only, returning a DLMOResult with dlmo phase and threshold; raises ValueError with a descriptive message (threshold value and data range) if the waveform never crosses the threshold
• New DLMOResult dataclass with __str__() and to_dict() (timing as HH:MM string); accepted by ResultsCollector.add() alongside SessionInfo and FitResult
• example_dlmo.py updated to use dlmo() and restricted to bcf/sbcf (bimodal functions are not appropriate for onset-only partial data); gen_time_range() call uses full_day=False
• Module docstrings added to all three example scripts
• melafit/__init__.py: corrected description of the midpoint marker
• example_one_fit.py: collector.add() now includes the fit result (res)
• Unit tests added for DLMOResult, dlmo(), and ResultsCollector integration with DLMOResult
• day_profile() gains an interp parameter (default None) for optional interpolation of empty bins before averaging, useful for sparse raw data
• All marker functions (dlmo, midpoint, area_cog) now accept a binsize parameter and apply interp='linear' by default, so sparse inputs are interpolated automatically before phase/marker extraction
• amplitude() and area_cog() switch to np.nanmin/np.nanmax so input NaNs (e.g. empty Excel cells) are ignored rather than propagated
• Unit tests added for interior NaN handling (TestInteriorNaNHandling)
• Fixed a bug in day_profile() that caused an error when the input data had a sampling period other than 1 minute; the function now correctly handles raw data and fitted curves at any temporal resolution. Combined with the new interp parameter and binsize support in the marker functions, it is now also possible to extract phase markers directly from raw data without any prior curve fitting or interpolation
• Fitting and marker functions (fit(), amplitude(), dlmo(), midpoint(), area_cog(), day_profile()) now accept pd.Series for value arguments and pd.Series/pd.DatetimeIndex for time arguments
• gen_time_range() switched from np.arange to integer step counting to avoid off-by-one errors from floating-point rounding
• area_cog(): baseline fallback to nanmin is now computed from the resampled waveform (after day_profile()) rather than the raw input
• area_cog() and midpoint() docstrings note the 24 h coverage assumption
• Unit tests added: TestSeriesInput (all new input combinations) and TestMarkersFromRawData (all four markers applied to sparse participant data without curve fitting)

v0.4.1 - Bugfix and documentation polish

• ResultsCollector.save() now appends .xlsx only when the supplied filename does not already end with .xlsx, preventing double-extension filenames such as results.xlsx.xlsx
• Unit test test_save_filename_with_xlsx_extension added to verify that a filename passed with .xlsx already present is written without modification
• Docstrings improved in markers.py and results.py
• README: collapsible sections added for BibTeX citation block, Revision History, Developer installation, and Developer update; section captions and markup refined; recommended package/environment manager updated to Miniforge
• midpoint() now raises ValueError with a descriptive message (data range and threshold value included) when the threshold is never crossed, replacing a silent IndexError
• area_cog() now raises ValueError with a descriptive message when the baseline is never crossed from below, or when the area under the curve is zero
• prepare_part_data() issues warnings.warn() instead of print() when correcting a duplicate timestamp, so the message integrates with standard Python warning filters
• Unused imports removed from markers.py (phase_to_string) and utils.py (os, scipy.optimize)
• Unit tests added: test_threshold_never_crossed_raises (TestMidpoint), test_baseline_never_crossed_raises and test_zero_area_raises (TestAreaCog)
• pyproject.toml: keywords, classifiers, and Documentation URL added

v0.4.0 - Improved API and examples

• AmplitudeResult now includes a baseline field (waveform minimum)
• ResultsCollector records and Excel output now include the baseline column
• __str__() added to all AnalysisRecord subclasses for human-readable print() output:
  • SessionInfo: participant name and session date/time range
  • AmplitudeResult: amplitude and baseline values
  • MidpointResult: DLMOn, DLMOff and Midpoint times
  • AreaCogResult: area and COG time
  • FitResult: function name, parameters and R²
  • AnalysisRecord base: generic fallback derived from to_dict()
• Example scripts updated to use print(meta), print(res), print(mid, ac) directly via the new __str__ representations
• Unit tests extended to cover the new baseline field and ResultsCollector column
• All example scripts simplified to import melafit as mf (single top-level import replaces multiple from melafit.xxx import ... lines)
• New minimal getting-started example example_one_fit.py: single-participant fit with bsbcf, area_cog, result collection, plot and Excel export
• os.makedirs(result_path, exist_ok=True) added to example_dlmo.py and example_full_profile.py so result directories are created automatically
• README: collapsible getting-started example with output figure added to the Getting Started section
• Version is now managed in a single source of truth: melafit/_version.py contains __version__; pyproject.toml uses dynamic = ["version"] with [tool.setuptools.dynamic] pointing to melafit._version.__version__; melafit/__init__.py imports and re-exports __version__ from ._version
• Fixed citation author in module docstring: "Ruf et al. (1992)" → "Ruf (1992)"

v0.3.0 - Cleaner API

• AnalysisResult renamed to AnalysisRecord
• AnalysisInfo renamed to SessionInfo; describes the data acquisition session only (participant, start, end); func and r2 removed
• SessionInfo constructor simplified: accepts a single p_data DataFrame (as returned by prepare_part_data()); participant, start and end are derived automatically
• FitResult.to_dict() now includes func (waveform function name) and r2 (R² goodness of fit), both computed automatically at the end of fit()
• r2 is no longer a field of SessionInfo; fit() computes and stores it in FitResult directly
• compute_wave() eliminated; replaced by the gen_time_range() + waveform-function pattern: gen_time_range() accepts a Timestamp series or explicit tmin/tmax bounds and a pandas offset string for step, and returns a time axis as float days since the Unix UTC epoch, which is then passed directly to the waveform function (e.g. bsbcf(t=gen_time_range(series, step="1min"), p=fit_result))
• New helper to_days() converts timestamps to float days since the Unix UTC epoch; timezone-naive input is treated as UTC, timezone-aware input is converted to UTC first
• New helper from_days() is the inverse of to_days(); returns a UTC-aware pd.DatetimeIndex
• day_profile(), midpoint() and area_cog() now accept a float days array (as returned by gen_time_range()) in addition to pd.DatetimeIndex
• fit() now accepts a datetime64 array or pandas Timestamp Series for time_fit; conversion via to_days() is automatic
• prepare_part_data() no longer adds a Timedays column to the returned DataFrame; time handling is done internally via to_days()
• prepare_part_data() returns an independent copy of the participant's data; mutations to the returned DataFrame do not affect the original; redundant Date and Time columns are dropped (both are combined in Timestamp)

v0.2.0 - Simplified API

• New module results.py with classes: AnalysisResult (abstract base), AnalysisInfo, AmplitudeResult, MidpointResult, AreaCogResult, FitResult and ResultsCollector for result management
• FitResult wraps scipy.optimize.OptimizeResult in its result field; fit() now returns FitResult
• FitResult can be passed directly to waveform functions and compute_wave
• amplitude(), midpoint() and area_cog() now return their respective result classes instead of tuples/floats
• to_dict() on all result classes: timing fields returned as HH:MM strings, other fields as native types
• AnalysisInfo.r2 defaults to NaN for convenience in DLMO workflows
• New string_to_phase() utility function in utils.py (inverse of phase_to_string)
• day_profile() accepts separate times and values parameters
• PARAM_NAMES renamed to BUILTIN_PARAM_NAMES
• Example scripts simplified via ResultsCollector and result classes
• Unit tests for all new classes and methods

v0.1.3

• Improved documentation
• Developer installation option

v0.1.2

• Improved documentation

v0.1.1 - First PyPI release

• Enhanced function fit() to support custom waveform functions with user-defined initial parameters and bounds
• Changed named parameter order in fit(): cost_f and cost_p are now the last two parameters
• Fixed returned type hints in func_defaults()
• Additional unit tests for new functionality
• Improved README
• Package registered in Python Package Index PyPI

v0.1.0 - First public release

• Dictionary support for waveform function parameters throughout the package: all functions accept both dict and np.ndarray for parameter input
• Named parameter constants: BCF_PARAM_NAMES, SBCF_PARAM_NAMES, BBCF_PARAM_NAMES, BSBCF_PARAM_NAMES and PARAM_NAMES lookup
• New utility functions params_to_array() and array_to_params() for conversion between array and named dictionary representations
• fit() now returns named parameter dictionary as res.p in addition to the standard scipy res.x array
• fit() now accepts cost_p dictionary for passing parameters to the cost function (e.g. {"eps": 1e-6})
• New utility function params_to_string() for human-readable parameter output
• Fixed area_cog(): baseline subtraction and bin size normalization
• Unit tests for all public functions in fitting, markers and utils

Initial revisions (v0.0.1 – v0.0.9)

• Full implementation of melatonin profile analysis as described in Gabel et al. (2017)
• Waveform functions: bcf, sbcf, bbcf, bsbcf
• Markers: amplitude, midpoint, DLMOn, DLMOff, area, cog
• Utilities: read_data, prepare_part_data, compute_wave, day_profile, abs_threshold, time_to_phase, phase_to_string, phase_diff
• Example scripts: example_dlmo.py (DLMO from partial data) and example_full_profile.py (full profile analysis)
• MIT license, packaging metadata and README

License

This project is licensed under the MIT License. See the LICENSE file for details.