ORION-B & DAOISM Related Free Software

The ORION-B and DAOISM projects develop open-source software to turn rich molecular-line observations into quantitative physical and chemical information. The packages below cover Bayesian inference, neural-network emulation, information analysis, multilayer radiative-transfer fitting, and practical tools for handling radio-astronomy data. Icons provide direct access to package installation, documentation, and source code when available.

python   BEETROOTS (BayEsian infErence with spaTial Regularization of nOisy multi-line ObservaTion mapS) — Pierre Palud   pypi-beetroots   docs-beetroots   repo-beetroots

BEETROOTS reconstructs maps of physical conditions such as temperature and gas density from multispectral cubes, for example maps of molecular-line integrated intensities. It combines Bayesian inference with spatial regularization so that neighbouring pixels help constrain one another, which is especially useful in low signal-to-noise regions.

The package supports realistic noise models combining additive Gaussian noise, multiplicative lognormal noise, and upper limits. It can use NNBMA emulators to accelerate expensive forward models, quantify uncertainties through Bayesian sampling, and test whether the adopted physical model and noise model can reproduce the observations. The same framework can also be run in optimization mode when faster point estimates are sufficient.

The documentation provides worked examples explaining the inputs and outputs and how to apply BEETROOTS to new observation maps and astrophysical models.

Main reference:   paper   Palud et al. (2025)

Method paper:   paper   Palud et al. (2023)

python   InfoVar (Informative Variables) — Lucas Einig   pypi-infovar   docs-infovar   repo-infovar

InfoVar quantifies how informative candidate observables are about physical quantities of interest. It can compare individual variables and combinations of variables, helping identify which measurements carry the most useful information for an inference problem.

Developed in the context of molecular-line diagnostics, it can be used to rank spectral tracers and to study how the information content changes across parameter space. This makes it useful both for interpreting existing observations and for selecting observables when designing future studies.

Main reference:   paper   Einig, Palud et al. (2024)

python   NNBMA (Neural Network-Based Model Approximation) — Lucas Einig, Pierre Palud   pypi-nnbma   docs-nnbma   repo-nnbma

NNBMA builds neural-network emulators of numerical models. It was originally developed to approximate the Meudon PDR code, reducing expensive model evaluations to millisecond-scale predictions while retaining enough accuracy for large inference problems.

The package provides a NumPy-like API built on PyTorch, without requiring users to work directly with deep-learning frameworks. It includes multilayer-perceptron architectures adapted to model emulation, methods for handling anomaly-prone training data, efficient derivatives of model outputs with respect to inputs, and optimized evaluation when only a subset of outputs is needed.

Main reference:   paper   Palud, Einig et al. (2023)

python   TOAST (mulTilayer clOud for simultAneouS analysis of mulTi-molecular lines) — Léontine Ségal   repo-toast

TOAST is a radiative-transfer-based code for inferring the physical, chemical, and kinematic properties of molecular gas when conditions vary along the line of sight. It simultaneously fits multiple molecular lines with a configurable multilayer cloud model, making it possible to represent structures such as a dense inner region surrounded by a more translucent envelope.

The code derives maps of fitted properties pixel by pixel over a field of view, supports multiprocessing, and provides confidence intervals on the estimated parameters. Users can customize the molecular lines, number of layers, abundance constraints, and kinematic assumptions. TOAST relies on the non-LTE radiative-transfer code RADEX.

The method was developed and tested on ORION-B observations of CO and HCO+ isotopologues toward the Horsehead nebula, where a three-layer model reproduces spectral asymmetries and separates the contributions of translucent and dense gas along the line of sight.

Main reference:   paper   Ségal et al. (2024)

python   RANCH (Radio AstroNomy Cubes Handler for data science) — Lucas Einig   pypi-ranch   docs-ranch   repo-ranch

RANCH provides a Python API for loading, manipulating, and preparing radio-astronomy FITS cubes for data-science workflows. It is intended to simplify the repetitive cube-handling steps that sit between calibrated observations and statistical or machine-learning analyses.

python   IRAM 30-meter EMIR observations informativity — Lucas Einig   docs-infobs   repo-infobs

This package estimates how informative spectral-line observations are for constraining physical conditions. It can simulate observations with the IRAM 30-meter telescope and EMIR receiver, while allowing other instrumental configurations to be explored as well.

Line intensities are predicted with a neural-network emulator of the Meudon PDR code, making it practical to compare the information carried by thousands of potential transitions over large model grids.

Main reference:   paper   Einig, Palud et al. (2024)

python   BeehiveNet — Lucas Einig   docs-beehivenet   repo-beehivenet

BeehiveNet defines, trains, and combines several specialist artificial neural networks. Instead of relying on a single global network, it allows different expert networks to focus on different physical regimes, with the aim of making inference more robust and interpretable.

It was developed for astrophysical parameter estimation, but the architecture is generic enough for other regression problems in which the relation between observables and targets changes across distinct regimes.

python   LaTeX ISM emission lines — Lucas Einig, Pierre Palud   repo-latexlines

This code implements a conversion of emission lines formatted according to the convention of the Meudon PDR code as well as a number of user-friendly methods for manipulating these lines.