Data Sources, Licensing & Citations

The Rareytec platform is built on openly-licensed software and data. This page lists every external component we use, its license, and the citation it requires.

The Rareytec platform integrates only open-licensed software (MIT, BSD, Apache-2.0, PSF) and openly-licensed or public-domain data (CC-BY-4.0, U.S. public domain). The bulk of the property values served are Rareytec’s own predictions over public chemical identities. A first-class “source” field carries provenance on every fact, and any individual source can be switched off for a given delivery. Machine-learning–derived parameters are only used when their training data is itself unrestricted, because an ML output inherits the license of the data it was trained on. An installation ships with a clean open baseline preinstalled — the CHAOS σ-profiles, the NIST/TRC ThermoML experimental data, and Rareytec’s own estimations — so it is useful on day one. Any further databases an operator is entitled to (e.g. the Dortmund Data Bank, DIPPR, Reaxys) are connected on an in-house installation and accessed live under that operator’s own license; their data is never copied into, or redistributed by, the platform.
Provenance trap. A machine-learning output inherits the license of its training data. Any artifact trained on restricted data (e.g. the Dortmund Data Bank) or on restricted calculation results (e.g. ADF) is excluded, regardless of how the artifact itself is distributed — a free download of the artifact does not launder the provenance of what it was trained on.

Software components

ComponentLicenseUse
Python PSF License Runtime language
RDKit BSD-3-Clause Cheminformatics: structure parsing, InChIKey, formula, 2D depiction
ugropy MIT UNIFAC / Dortmund group fragmentation from SMILES (ILP set-cover assignment)
thermo / chemicals / fluids MIT Build-time verified source for published UNIFAC parameters and independent validation oracle (Caleb Bell)
NumPy BSD-3-Clause Numerics
Flask / Jinja2 / Werkzeug BSD-3-Clause Web framework + templating + auth hashing
Bootstrap MIT User-interface layout
Plotly.js MIT σ-profile and VLE charts
PubChemPy MIT Chemical-name resolution via PubChem

Component & property data

CHAOS database
CC-BY-4.0 — included with attribution

σ-profiles and σ-surfaces (COSMO-SAC cavity segments), dipole moments and molecular descriptors for the 53,078-component registry. Bundled (preinstalled) with the platform.

Cite: Jirasek et al., “CHAOS — A Consistent Large-scale Database for Sigma-Profiles and Other Molecular Descriptors,” arXiv:2511.19002; Zenodo doi:10.5281/zenodo.17691924. Licensed under CC-BY-4.0 (creativecommons.org/licenses/by/4.0/). Modified by Rareytec: indexed and reformatted into the component registry, with σ-profiles and cavity segments extracted for property prediction and 3-D visualisation. link

NIST / TRC ThermoML Archive
U.S. public domain (NIST) — included with attribution

Experimental thermophysical data (VLE, LLE, SLE, excess enthalpy, activity coefficients, density, viscosity, …) shown alongside predictions and used for validation. Bundled (preinstalled) with the platform; every data set keeps its original literature citation and DOI. The bundled copy is the complete free public release (v2020-09-30), covering the 2003–2019 calendar years — the last freely distributed ThermoML bulk archive. Later data is available only under your own license (see “Importing your own licensed data” below).

Cite: Data from the NIST/TRC ThermoML Archive (doi:10.18434/mds2-2422), a work of the U.S. Government not subject to copyright in the United States and distributed by NIST for reuse. Archive/format: Chirico, Frenkel, Diky et al., “ThermoML: An XML-Based Approach for Storage and Exchange of Experimental and Critically Evaluated Thermophysical and Thermochemical Property Data,” J. Chem. Eng. Data. The underlying measurements are credited to their respective authors through the per-data-set citation shown with each set. link

PubChem
Public domain (U.S. NLM / NCBI)

Chemical identity — resolving names to structures.

Cite: Kim et al., “PubChem 2023 update,” Nucleic Acids Res. 2023, 51, D1373. link

Model parameters

UNIFAC (original)
Open literature

Activity-coefficient model — original parameter set.

Cite: Hansen, Rasmussen, Fredenslund, Schiller, Gmehling, Ind. Eng. Chem. Res. 1991, 30, 2352–2355. Fredenslund, Jones, Prausnitz, AIChE J. 1975, 21, 1086.

Modified UNIFAC (Dortmund)
Published (free companion download)

Activity-coefficient model — published modified-UNIFAC (Dortmund) parameter matrix.

Cite: Constantinescu & Gmehling, “Further Development of Modified UNIFAC (Dortmund): Revision and Extension 6,” J. Chem. Eng. Data 2016, 61, 2738. Also in Gmehling, Kolbe, Kleiber, Rarey, “Chemical Thermodynamics for Process Simulation,” free companion site chemthermo.ddbst.com.

NIST-modified UNIFAC
NIST (published, public)

Activity-coefficient model — NIST critically-evaluated parameter set (89 main groups, 984 interactions).

Cite: Kang, Diky, Frenkel, “New modified UNIFAC parameters using critically evaluated phase equilibrium data,” Fluid Phase Equilibria 388 (2015) 128–141; doi:10.1016/j.fluid.2014.12.042.

UNIFAC 2.0
Open (free ancillary data)

Activity-coefficient model — machine-learning completion of the published original-UNIFAC parameter table.

Cite: Hayer, Wendel, Mandt, Hasse, Jirasek, “Advancing thermodynamic group-contribution methods by machine learning: UNIFAC 2.0,” Chemical Engineering Journal 504 (2025) 158667; arXiv:2408.05220.

Group fragmentation (ugropy)
MIT

SMILES → UNIFAC / Dortmund subgroup assignment.

Cite: ugropy: an optimal functional-group identification package, Ind. Eng. Chem. Res. 2025; doi:10.1021/acs.iecr.5c02552.

Importable under your own license

These are not bundled and never redistributed by Rareytec. If your organisation already holds a license, they can be connected on your own installation and read live under that license — their data never enters our database or a delivered build.

NIST ThermoData Engine (TDE)
Importable — requires your own NIST TDE license

If you or your organisation owns a TDE license, its experimental SOURCE archive (compounds, references, pure & binary measurements) can be imported as a private data source on your installation (user or company layer). Import feasibility has been verified — TDE's database is technically readable (older versions are a Microsoft Access/Jet file; newer ones via the Access database engine).

Import feasibility verified.

DIPPR (AIChE / BYU)
Importable — requires your own DIPPR license

If you or your organisation is entitled to DIPPR, its pure-component constants and temperature-dependent correlations can be connected as a private source under your license (user or company layer).

Not yet tested — import from the DIPPR distribution format has not been verified.

Dortmund Data Bank (DDB)
Importable — requires your own DDB / DDBST license

If you or your organisation holds a DDB license, its experimental phase-equilibrium, excess-property and pure-component data can be connected as a private source under your license (user or company layer) — read live on your installation and shown alongside our predictions.

Not yet tested — import from the DDB distribution format has not been verified, and connection is subject to the terms of your DDB license.

Searching external sources for your project

The platform can assist you in finding data for the specific mixture and submixtures of YOUR project in external, freely accessible data sources. This is task-scoped research assistance, not bulk harvesting: queries are tied to a concrete project need, results are shown with their source, and anything you import lands only in that project’s workspace with full provenance (source, link, retrieval date, citation). The platform never copies external databases into its own shipped data.
Availability lookups (metadata only) — Dortmund Data Bank

The free DDBST online search reports WHAT data exist for a system (data banks, set/point counts, temperature and pressure ranges) without revealing any values. DDBST states explicitly that “no copyright is reserved for the search results,” so the platform may run these lookups automatically and cache the availability tables. Value ranges DDBST masks as “(hidden)” are shown as such — the platform never tries to infer values from availability metadata. To obtain the data itself: request a quote from DDBST, or use your own DDB license (export the system as PPDS in the Dortmund Data Bank application and import the file here).

Free reference sources (rights reserved) — retrieval on your request

Sources that are free to consult but not openly licensed (the Korea Thermophysical Properties Data Bank KDB; the NIST Chemistry WebBook, whose Standard Reference Data is copyrighted under the U.S. SRD Act unlike the public-domain ThermoML archive) are searched and retrieved only when you ask, for your specific system, into your project only. Each user’s retrieval is a small, task-scoped extraction under the lawful-user provisions of database law; the platform keeps no shared cache and never pools such retrievals across users or into the shipped baseline. Every dataset carries the source’s requested citation.

Confidential projects

Any external query necessarily sends your component identities (name/CAS/structure) to the external source’s server. Projects can be flagged confidential — at creation or any time later — which disables all AUTOMATIC external queries; you can still query explicitly with a click.

Importing your own licensed data

The bundled experimental baseline is deliberately limited to freely redistributable data — for ThermoML that is the 2003–2019 public archive. If you or your organisation hold a license for data outside that window or from another source — later ThermoML content via NIST ThermoData Engine, the Dortmund Data Bank, DIPPR, in-house measurements, or a private journal-supporting-information collection — you may import it into your own user or company database. Such data is treated as being under your license: it stays private to your installation, powers your own predictions, regressions and simulator exports, and is never copied into the shipped baseline or redistributed by the platform.

How your data is stored

Data is kept in three separate databases by provenance and entitlement, so the redistributable baseline is never mixed with tenant- or user-licensed data.

1. Core baseline
Ships with every installation · read-mostly

The clean, redistributable baseline: Rareytec’s own Rarey–Nannoolal and COSMO-SAC predictions over the public component registry, the CHAOS σ-profiles, and the free public NIST/TRC ThermoML experimental archive (2003–2019). Everything here is open-licensed or public-domain.

2. Company database
One per deployment · shared within the organisation

Data the organisation is entitled to and chooses to share in-house: values imported under the company’s own licenses, and project data that has been reviewed and promoted (see below). Private to the deployment; never redistributed by Rareytec.

3. User / project workspace
Per user · per project

Where each user works: imported data files (IK-CAPE / DDB exports, ThermoML, Reaxys, PPDS, CHEMCAD .ccsim / .nf), manual entries, and regressed parameters — held per project and used for that user’s own calculations and exports.

Data promotion — propose & audit

Data only ever moves upward through a reviewed valve, never automatically. From a project a user can PROPOSE a component’s values for promotion — attesting they hold the right to share the data and attaching a citation/source; the proposal captures a snapshot of the values and enters a pending queue. A designated reviewer then AUDITS the proposal and approves or rejects it. On approval the data is published to the shared company database with a full audit trail: who proposed it, who reviewed it, when, the attested source, and the identifier of the minted company source. Every shared value therefore stays traceable to its origin and to the licensing attestation made when it was shared.

Propose Pending review Audited & published to the company bank — with a full who/when/source audit trail.

Rareytec’s own contribution

  • Rarey–Nannoolal group-contribution estimations (normal boiling temperature, vapor pressure, critical properties) — our own scientific work. Cite: Nannoolal, Rarey, Ramjugernath, Cordes, Fluid Phase Equilibria 226 (2004) 45; and subsequent papers.
  • COSMO-SAC mixture predictions computed from CHAOS σ-profiles.
  • All platform predictions and the curated registry schema are Rareytec’s own work.

Sources we deliberately do not use

Dortmund Data Bank (DDB), DETHERM, DIPPR, Reaxys

Proprietary, paywalled commercial databases whose licenses restrict USE — not merely redistribution. To keep the platform’s data provenance clean and unencumbered we do not use them for any purpose, including internal development or validation. DIPPR data, where a user is entitled to it, is accessed only per-user under that user’s own license and is never stored in our database.

Modified UNIFAC (Dortmund) — consortium-only complete matrix

We use the PUBLISHED modified-UNIFAC (Dortmund) parameter table (J. Chem. Eng. Data 2016 revision; free companion download of the Gmehling et al. textbook) — see Model parameters above. What we do NOT use is the consortium-only complete/current commercial matrix distributed by DDBST to members, nor the Dortmund Data Bank experimental databank.

Last reviewed 2026-07-08. If you believe a source is mis-attributed, contact Rareytec.