pith. sign in

arxiv: 2605.29698 · v1 · pith:NDOTVGCHnew · submitted 2026-05-28 · 💻 cs.LG · physics.chem-ph

A Systematic Evaluation of Molecular Mixture Behavior Prediction

Roel J. Leenhouts , Nathan K. Morgan , William Green , Jan G. Rittig , Florence H. Vermeire This is my paper

Pith reviewed 2026-06-29 08:54 UTC · model grok-4.3

classification 💻 cs.LG physics.chem-ph
keywords molecular mixturesproperty predictionmachine learningevaluation frameworknon-ideal behaviordata splitsmixture datasets
0
0 comments X

The pith

Machine learning models for molecular mixtures can show good overall accuracy while failing to capture non-ideal interactions between components.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper develops an evaluation method that breaks down prediction errors in mixture properties into contributions from individual pure compounds and from deviations caused by mixing. It demonstrates that high accuracy on absolute values does not guarantee correct modeling of the extra effects from intermolecular interactions. When models are tested on mixtures involving molecules absent from training data, their ability to predict these non-ideal effects declines markedly. The work emphasizes the need for evaluation methods that go beyond total error to assess true mixture behavior.

Core claim

The authors claim that absolute accuracy metrics in mixture property prediction can obscure poor performance on non-ideal components, with substantial drops under molecule-based splits that prevent leakage, identifying transfer to unseen molecules as the main challenge.

What carries the argument

A decomposition framework using ideal-mixture baselines and excess-property metrics to separate pure-compound errors from interaction errors, paired with leakage-aware data splits.

If this is right

  • Evaluations of mixture models must include checks on excess properties to verify recovery of non-ideal behavior.
  • Performance on strict molecule splits provides a better indicator of generalization than random splits.
  • Curated matched datasets of pure and mixture properties enable more reliable benchmarking.
  • Models should be assessed for their ability to predict deviations from ideal mixing separately from pure component properties.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Explicit modeling of interaction terms could improve recovery of non-ideal effects beyond current end-to-end approaches.
  • Similar decomposition techniques might help in evaluating predictions for other complex systems like solutions or alloys.
  • The findings imply that larger and more diverse mixture datasets will be necessary to improve transfer performance.

Load-bearing premise

Ideal mixture baselines and excess property calculations isolate non-ideal errors without interference from dataset biases or model assumptions.

What would settle it

Compare model predictions of excess properties on mixtures of entirely new molecules against experimental data to see if the non-ideal component error is low.

Figures

Figures reproduced from arXiv: 2605.29698 by Florence H. Vermeire, Jan G. Rittig, Nathan K. Morgan, Roel J. Leenhouts, William Green.

Figure 1
Figure 1. Figure 1: Left: target and predicted mixture-property curves, with shaded regions indicating excess [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the four model comparison axes: component featurization (orange), interaction [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Differences in absolute, excess, and ideal RMSE from pure-to-mixture to molecule splits [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Absolute and excess MAE vs. the ideal-mixture reference baseline under the mixture split. [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Absolute and excess RMSE learning curves under the mixture split for [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Absolute and excess RMSE for temperature-context variants. [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Distribution of mixture-property values and molecular weights across evaluation tasks. [PITH_FULL_IMAGE:figures/full_fig_p016_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Distribution of number of components per mixture data point across evaluation datasets (6+ [PITH_FULL_IMAGE:figures/full_fig_p016_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Distribution of excess-property values across mixture datasets, where excess values are [PITH_FULL_IMAGE:figures/full_fig_p017_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Differences in absolute, excess, and ideal RMSE from pure-to-mixture to molecule splits [PITH_FULL_IMAGE:figures/full_fig_p021_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Absolute MAE vs. the ideal-mixture reference baseline under the mixture split, for all [PITH_FULL_IMAGE:figures/full_fig_p023_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Excess MAE vs. the ideal-mixture reference baseline under the mixture split, for all [PITH_FULL_IMAGE:figures/full_fig_p023_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Dataset-wise absolute-RMSE learning curves under the mixture split for [PITH_FULL_IMAGE:figures/full_fig_p024_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Dataset-wise excess-RMSE learning curves under the mixture split for [PITH_FULL_IMAGE:figures/full_fig_p024_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Fold-00 train/test temperature distributions for the solubility and viscosity mixture [PITH_FULL_IMAGE:figures/full_fig_p025_15.png] view at source ↗
read the original abstract

Machine learning for molecular property prediction has focused largely on pure compounds, even though many practical applications depend on mixtures with intermolecular interactions. Recent work has expanded the availability of mixture datasets, but evaluation still focuses mainly on absolute accuracy. However, absolute errors in mixtures conflate pure-component contributions with deviations from ideal mixing. We propose an evaluation framework that decomposes mixture-property error into pure-compound and interaction (non-ideal) components. The framework combines leakage-aware split protocols, ideal-mixture baselines, and excess-property metrics. To support reproducible benchmarking, we curate seven matched pure and mixture physicochemical property datasets. Across multiple mixture-property tasks and model families, we find that strong absolute accuracy can mask poor recovery of non-ideal mixture behavior, and that performance drops substantially under strict molecule splits. These results identify transfer to unseen molecules as a central challenge in molecular mixture machine learning and motivate evaluation beyond absolute accuracy alone.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

0 major / 2 minor

Summary. The manuscript proposes an evaluation framework for machine learning models predicting physicochemical properties of molecular mixtures. The framework decomposes prediction error into pure-compound and non-ideal interaction (excess) components via ideal-mixture baselines and excess-property metrics, combined with leakage-aware splits. The authors curate seven matched pure-compound and mixture datasets and evaluate multiple model families, reporting that high absolute accuracy often masks poor recovery of non-ideal behavior while performance drops substantially under strict molecule-based splits. The central conclusion is that transfer to unseen molecules remains a key challenge in molecular mixture machine learning.

Significance. If the decomposition is shown to be free of confounding, the work would be significant for establishing a reproducible benchmarking standard that moves the field beyond absolute-error metrics toward isolating intermolecular interaction effects. The curation of matched datasets and the explicit comparison of absolute vs. excess metrics provide concrete evidence that current models struggle with generalization, which could guide future method development in a practically relevant domain.

minor comments (2)
  1. The abstract states the decomposition but does not include the explicit equation; adding it (or a reference to the methods section) would improve immediate clarity for readers.
  2. Figure or table captions should explicitly state the number of molecules in each strict split to allow quick assessment of the generalization gap magnitude.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive assessment of our work and the recommendation for minor revision. The report does not enumerate any specific major comments requiring point-by-point rebuttal.

Circularity Check

0 steps flagged

No circularity: empirical evaluation with independent baselines and metrics

full rationale

The paper is an evaluation study that curates datasets, applies leakage-aware splits, computes ideal-mixture baselines from pure-component data, and uses excess-property metrics to isolate non-ideal contributions. All reported findings (absolute accuracy masking non-ideal recovery, performance drop under molecule splits) are direct empirical observations from running existing model families on these datasets. No derivation, prediction, or uniqueness claim reduces by construction to fitted parameters, self-citations, or ansatzes; the decomposition is a definitional accounting identity using standard thermodynamic excess functions, not a self-referential result. Self-citations, if present, are not load-bearing for the central claims.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review yields no explicit free parameters, axioms, or invented entities beyond standard ML evaluation practices; the framework itself introduces no new physical postulates.

pith-pipeline@v0.9.1-grok · 5698 in / 1011 out tokens · 23298 ms · 2026-06-29T08:54:18.042338+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

69 extracted references · 58 canonical work pages · 3 internal anchors

  1. [1]

    Muhieddine, Shekhar K

    Mohamad H. Muhieddine, Shekhar K. Viswanath, Alan Armstrong, Amparo Galindo, and Claire S. Adjiman. Model-based solvent selection for the synthesis and crystallisation of pharmaceutical compounds. Chemical Engineering Science, 264:118125, 2022. DOI: 10.1016/j.ces.2022.118125

    work page doi:10.1016/j.ces.2022.118125 2022

  2. [2]

    Huber, Sara Iborra, and Avelino Corma

    George W. Huber, Sara Iborra, and Avelino Corma. Synthesis of Transportation Fuels from Biomass: Chem- istry, Catalysts, and Engineering.Chemical Reviews, 106(9):4044–4098, 2006. DOI: 10.1021/cr068360d

    work page doi:10.1021/cr068360d 2006

  3. [3]

    Patel, Absar Lakdawala, Sajan Chourasia, and Rajesh N

    Paresh D. Patel, Absar Lakdawala, Sajan Chourasia, and Rajesh N. Patel. Bio fuels for compression ignition engine: A review on engine performance, emission and life cycle analysis.Renewable and Sustainable Energy Reviews, 65:24–43, 2016. DOI: 10.1016/j.rser.2016.06.010

    work page doi:10.1016/j.rser.2016.06.010 2016

  4. [4]

    Roger A. Sheldon. The E factor 25 years on: the rise of green chemistry and sustainability.Green Chem., 19:18–43, 2017. DOI: 10.1039/C6GC02157C

    work page doi:10.1039/c6gc02157c 2017

  5. [5]

    David J. C. Constable, Conchita Jimenez-Gonzalez, and Richard K. Henderson. Perspective on Solvent Use in the Pharmaceutical Industry.Organic Process Research & Development, 11(1):133–137, 2007. DOI: 10.1021/op060170h

    work page doi:10.1021/op060170h 2007

  6. [6]

    Clarke, Wei-Chien Tu, Oliver Levers, Andreas Bröhl, and Jason P

    Coby J. Clarke, Wei-Chien Tu, Oliver Levers, Andreas Bröhl, and Jason P. Hallett. Green and Sustainable Solvents in Chemical Processes.Chemical Reviews, 118(2):747–800, 2018. DOI: 10.1021/acs.chemrev.7b00571

    work page doi:10.1021/acs.chemrev.7b00571 2018

  7. [7]

    Towards the prediction of drug solubility in binary solvent mixtures at various temperatures using machine learning.Research Square, 2024

    Zeqing Bao, Gary Tom, Austin Cheng, Alán Aspuru-Guzik, and Christine Allen. Towards the prediction of drug solubility in binary solvent mixtures at various temperatures using machine learning.Research Square, 2024. DOI: 10.21203/rs.3.rs-4170106/v1

    work page doi:10.21203/rs.3.rs-4170106/v1 2024

  8. [8]

    Cho, and H.C

    Bhupendra Singh Chauhan, Ram Kripal Singh, H.M. Cho, and H.C. Lim. Practice of diesel fuel blends using alternative fuels: A review.Renewable and Sustainable Energy Reviews, 59:1358–1368, 2016. DOI: 10.1016/j.rser.2016.01.062

    work page doi:10.1016/j.rser.2016.01.062 2016

  9. [9]

    G. M. Wilson and C. H. Deal. Activity Coefficients and Molecular Structure. Activity Coefficients in Changing Environments-Solutions of Groups.Industrial & Engineering Chemistry Fundamentals, 1(1): 20–23, 1962. DOI: 10.1021/i160001a003

    work page doi:10.1021/i160001a003 1962

  10. [10]

    Jones, and John M

    Aage Fredenslund, Russell L. Jones, and John M. Prausnitz. Group-contribution estimation of activity coef- ficients in nonideal liquid mixtures.AIChE Journal, 21(6):1086–1099, 1975. DOI: 10.1002/aic.690210607

    work page doi:10.1002/aic.690210607 1975

  11. [11]

    Conductor-like screening model for real solvents: A new approach to the quantitative calculation of solvation phenomena.The Journal of Physical Chemistry, 99(7):2224–2235, 1995

    Andreas Klamt. Conductor-like screening model for real solvents: A new approach to the quantitative calculation of solvation phenomena.The Journal of Physical Chemistry, 99(7):2224–2235, 1995. DOI: 10.1021/j100007a062. 10

    work page doi:10.1021/j100007a062 1995

  12. [12]

    HANNA: hard-constraint neural network for consistent activity coefficient prediction.Chemical Science, 15(47): 19777–19786, 2024

    Thomas Specht, Mayank Nagda, Sophie Fellenz, Stephan Mandt, Hans Hasse, and Fabian Jirasek. HANNA: hard-constraint neural network for consistent activity coefficient prediction.Chemical Science, 15(47): 19777–19786, 2024. DOI: 10.1039/d4sc05115g

    work page doi:10.1039/d4sc05115g 2024

  13. [13]

    Vermeire and William H

    Florence H. Vermeire and William H. Green. Transfer learning for solvation free energies: From quantum chemistry to experiments.Chemical Engineering Journal, 418:129307, 2021. DOI: 10.1016/j.cej.2021.129307

    work page doi:10.1016/j.cej.2021.129307 2021

  14. [14]

    Rittig and Alexander Mitsos

    Jan G. Rittig and Alexander Mitsos. Thermodynamics-consistent graph neural networks.Chemical Science, 15:18504–18512, 2024. DOI: 10.1039/D4SC04554H

    work page doi:10.1039/d4sc04554h 2024

  15. [15]

    Feinberg, Joseph Gomes, Caleb Geniesse, Aneesh S

    Zhenqin Wu, Bharath Ramsundar, Evan N. Feinberg, Joseph Gomes, Caleb Geniesse, Aneesh S. Pappu, Karl Leswing, and Vijay Pande. MoleculeNet: a benchmark for molecular machine learning.Chemical Science, 9:513–530, 2018. DOI: 10.1039/C7SC02664A

    work page doi:10.1039/c7sc02664a 2018

  16. [16]

    Therapeutics data commons: Machine learning datasets and tasks for drug discovery and development.arXiv preprint arXiv:2102.09548, 2021

    Kexin Huang, Tianfan Fu, Wenhao Gao, Yue Zhao, Yusuf H Roohani, Jure Leskovec, Connor W. Coley, Cao Xiao, Jimeng Sun, and Marinka Zitnik. Therapeutics Data Commons: Machine Learning Datasets and Tasks for Drug Discovery and Development. 2021. URLhttps://arxiv.org/abs/2102.09548

    work page arXiv 2021

  17. [17]

    doi: 10.1038/sdata.2014.22

    Raghunathan Ramakrishnan, Pavlo O. Dral, Matthias Rupp, and O. Anatole von Lilienfeld. Quantum chemistry structures and properties of 134 kilo molecules.Scientific Data, 1(1):140022, 2014. DOI: 10.1038/sdata.2014.22

    work page doi:10.1038/sdata.2014.22 2014

  18. [18]

    Segler, and Alain C

    Nathan Brown, Marco Fiscato, Marwin H.S. Segler, and Alain C. Vaucher. GuacaMol: Benchmarking Models for de Novo Molecular Design.Journal of Chemical Information and Modeling, 59(3):1096–1108,

  19. [19]

    DOI: 10.1021/acs.jcim.8b00839

    work page doi:10.1021/acs.jcim.8b00839

  20. [20]

    Open Graph Benchmark: Datasets for Machine Learning on Graphs

    Weihua Hu, Matthias Fey, Marinka Zitnik, Yuxiao Dong, Hongyu Ren, Bowen Liu, Michele Catasta, and Jure Leskovec. Open Graph Benchmark: Datasets for Machine Learning on Graphs. In H. Larochelle, M. Ranzato, R. Hadsell, M.F. Balcan, and H. Lin, editors,Ad- vances in Neural Information Processing Systems, volume 33, pages 22118–22133. Curran Asso- ciates, In...

    2020

  21. [21]

    Sample Efficiency Matters: A Benchmark for Practical Molecular Optimization

    Wenhao Gao, Tianfan Fu, Jimeng Sun, and Connor Coley. Sample Efficiency Matters: A Benchmark for Practical Molecular Optimization. In S. Koyejo, S. Mohamed, A. Agarwal, D. Belgrave, K. Cho, and A. Oh, editors,Advances in Neural Information Processing Systems, volume 35, pages 21342–21357. Cur- ran Associates, Inc., 2022. URL https://proceedings.neurips.cc...

    work page arXiv 2022

  22. [22]

    Leenhouts, Nathan Morgan, Emad Al Ibrahim, William H

    Roel J. Leenhouts, Nathan Morgan, Emad Al Ibrahim, William H. Green, and Florence H. Vermeire. Pooling solvent mixtures for solvation free energy predictions.Chemical Engineering Journal, 513: 162232, 2025. DOI: 10.1016/j.cej.2025.162232

    work page doi:10.1016/j.cej.2025.162232 2025

  23. [23]

    Chew, Mohammad Atif Faiz Afzal, Zachary Kaplan, Eric M

    Alex K. Chew, Mohammad Atif Faiz Afzal, Zachary Kaplan, Eric M. Collins, Suraj Gattani, Mayank Misra, Anand Chandrasekaran, Karl Leswing, and Mathew D. Halls. Leveraging high-throughput molecular simulations and machine learning for the design of chemical mixtures.npj Computational Materials, 11 (1):72, 2025. DOI: 10.1038/s41524-025-01552-2

    work page doi:10.1038/s41524-025-01552-2 2025

  24. [24]

    Leenhouts, Tara Larsson, Sebastian Verhelst, and Florence H

    Roel J. Leenhouts, Tara Larsson, Sebastian Verhelst, and Florence H. Vermeire. Property prediction of fuel mixtures using pooled graph neural networks.Fuel, 381:133218, 2025. DOI: 10.1016/j.fuel.2024.133218

    work page doi:10.1016/j.fuel.2024.133218 2025

  25. [25]

    Machine learning for predicting the viscosity of binary liquid mixtures.Chemical Engineering Journal, 464:142454, 2023

    Camille Bilodeau, Andrei Kazakov, Sukrit Mukhopadhyay, Jillian Emerson, Tom Kalantar, Chris Muzny, and Klavs Jensen. Machine learning for predicting the viscosity of binary liquid mixtures.Chemical Engineering Journal, 464:142454, 2023. DOI: 10.1016/j.cej.2023.142454

    work page doi:10.1016/j.cej.2023.142454 2023

  26. [26]

    Mani Sarathy

    Nursulu Kuzhagaliyeva, Samuel Horváth, John Williams, Andre Nicolle, and S. Mani Sarathy. Artificial intelligence-driven design of fuel mixtures.Communications Chemistry, 5(1), 2022. DOI: 10.1038/s42004- 022-00722-3

    work page doi:10.1038/s42004- 2022

  27. [27]

    MixtureSolDB, dataset of solubility values for organic compounds in binary mixtures of solvents at various temperatures.ChemRxiv, 2025

    Dmitry Malikov, Lev Krasnov, Marina Kiseleva, Elizaveta Meshcheriakova, Fedor Kuznetsov, Vladimir Elistratov, Matvei Vasiyarov, Sergei Tatarin, and Stanislav Bezzubov. MixtureSolDB, dataset of solubility values for organic compounds in binary mixtures of solvents at various temperatures.ChemRxiv, 2025. DOI: 10.26434/chemrxiv-2025-m51v8

    work page doi:10.26434/chemrxiv-2025-m51v8 2025

  28. [28]

    Mejía-Mendoza, Seyed Mohamad Moosavi, and Benjamin Manuel Sanchez

    Ella Miray Rajaonson, Mahyar Rajabi Kochi, Luis M. Mejía-Mendoza, Seyed Mohamad Moosavi, and Benjamin Manuel Sanchez. CheMixHub: Datasets and Benchmarks for Chemical Mixture Property Prediction. InThe Thirty-ninth Annual Conference on Neural Information Processing Systems Datasets and Benchmarks Track, 2025. URLhttps://openreview.net/forum?id=8HUnx0rJNq. 11

    2025

  29. [29]

    J. M. Prausnitz, R. N. Lichtenthaler, and Edmundo Gomes de Azevedo.Molecular thermodynamics of fluid-phase equilibria third edition. Prentice Hall, 1999

    1999

  30. [30]

    Henri Renon and J. M. Prausnitz. Local compositions in thermodynamic excess functions for liquid mixtures.AIChE Journal, 14(1):135–144, 1968. DOI: 10.1002/aic.690140124

    work page doi:10.1002/aic.690140124 1968

  31. [31]

    Martins, Márcio J

    Rosana J. Martins, Márcio J. E. de M. Cardoso, and Oswaldo E. Barcia. Excess Gibbs Free Energy Model for Calculating the Viscosity of Binary Liquid Mixtures.Industrial & Engineering Chemistry Research, 39 (3):849–854, 2000. DOI: 10.1021/ie990398b

    work page doi:10.1021/ie990398b 2000

  32. [32]

    Machine learning for predicting thermodynamic proper- ties of pure fluids and their mixtures.Energy, 188:116091, 2019

    Yuanbin Liu, Weixiang Hong, and Bingyang Cao. Machine learning for predicting thermodynamic proper- ties of pure fluids and their mixtures.Energy, 188:116091, 2019. DOI: 10.1016/j.energy.2019.116091

    work page doi:10.1016/j.energy.2019.116091 2019

  33. [33]

    A systematic study of key elements underlying molecular property prediction.Nature Communications, 14(1), 2023

    Jianyuan Deng, Zhibo Yang, Hehe Wang, Iwao Ojima, Dimitris Samaras, and Fusheng Wang. A systematic study of key elements underlying molecular property prediction.Nature Communications, 14(1), 2023. DOI: 10.1038/s41467-023-41948-6

    work page doi:10.1038/s41467-023-41948-6 2023

  34. [34]

    Blumenthal, and Olga V

    Roman Joeres, David B. Blumenthal, and Olga V . Kalinina. Data splitting to avoid information leakage with DataSAIL.Nature Communications, 16(1), 2025. DOI: 10.1038/s41467-025-58606-8

    work page doi:10.1038/s41467-025-58606-8 2025

  35. [35]

    Schweidtmann, Jan G

    Artur M. Schweidtmann, Jan G. Rittig, Jana M. Weber, Martin Grohe, Manuel Dahmen, Kai Leonhard, and Alexander Mitsos. Physical pooling functions in graph neural networks for molecular property prediction. Computers & Chemical Engineering, 172:108202, 2023. DOI: 10.1016/j.compchemeng.2023.108202

    work page doi:10.1016/j.compchemeng.2023.108202 2023

  36. [36]

    Dobbelaere, István Lengyel, Christian V

    Maarten R. Dobbelaere, István Lengyel, Christian V . Stevens, and Kevin M. Van Geem. Geometric deep learning for molecular property predictions with chemical accuracy across chemical space.Journal of Cheminformatics, 16(1), 2024. DOI: 10.1186/s13321-024-00895-0

    work page doi:10.1186/s13321-024-00895-0 2024

  37. [37]

    McGill, Florence H

    Esther Heid, Charles J. McGill, Florence H. Vermeire, and William H. Green. Characterizing Uncertainty in Machine Learning for Chemistry.Journal of Chemical Information and Modeling, 63(13):4012–4029,

  38. [38]

    DOI: 10.1021/acs.jcim.3c00373

    work page doi:10.1021/acs.jcim.3c00373

  39. [39]

    Van Lehn, and Victor M

    Shiyi Qin, Shengli Jiang, Jianping Li, Prasanna Balaprakash, Reid C. Van Lehn, and Victor M. Zavala. Capturing molecular interactions in graph neural networks: a case study in multi-component phase equilibrium.Digital Discovery, 2:138–151, 2023. DOI: 10.1039/D2DD00045H

    work page doi:10.1039/d2dd00045h 2023

  40. [40]

    Di Wu, Zutao Zhu, Jun Zhang, Huaqiang Wen, Saimeng Jin, and Weifeng Shen. An Interpretable Solute–Solvent Interactive Attention Module Intensified Graph-Learning Architecture toward Enhancing the Prediction Accuracy of an Infinite Dilution Activity Coefficient.Industrial & Engineering Chemistry Research, 63(19):8741–8750, 2024. DOI: 10.1021/acs.iecr.4c00107

    work page doi:10.1021/acs.iecr.4c00107 2024

  41. [41]

    A smile is all you need: predicting limiting activity coefficients from SMILES with natural language processing.Digital Discovery, 1:859–869, 2022

    Benedikt Winter, Clemens Winter, Johannes Schilling, and André Bardow. A smile is all you need: predicting limiting activity coefficients from SMILES with natural language processing.Digital Discovery, 1:859–869, 2022. DOI: 10.1039/D2DD00058J

    work page doi:10.1039/d2dd00058j 2022

  42. [42]

    Molecular machine learning in chemical process design.Current Opinion in Chemical Engineering, 52:101239, 2026

    Jan G Rittig, Manuel Dahmen, Martin Grohe, Philippe Schwaller, and Alexander Mitsos. Molecular machine learning in chemical process design.Current Opinion in Chemical Engineering, 52:101239, 2026

    2026

  43. [43]

    Rittig, Kobi C

    Jan G. Rittig, Kobi C. Felton, Alexei A. Lapkin, and Alexander Mitsos. Gibbs–Duhem-informed neu- ral networks for binary activity coefficient prediction.Digital Discovery, 2:1752–1767, 2023. DOI: 10.1039/D3DD00103B

    work page doi:10.1039/d3dd00103b 2023

  44. [44]

    Rondinelli, and Wei Chen

    Hengrui Zhang, Tianxing Lai, Jie Chen, Arumugam Manthiram, James M. Rondinelli, and Wei Chen. Learning Molecular Mixture Property Using Chemistry-Aware Graph Neural Network.PRX Energy, 3: 023006, 2024. DOI: 10.1103/PRXEnergy.3.023006

    work page doi:10.1103/prxenergy.3.023006 2024

  45. [45]

    Enis Leblebici

    Ulderico Di Caprio, Florence Vermeire, Tom Van Gerven, and M. Enis Leblebici. Physics-informed machine learning predicting CO2 capture performances of organic mixtures.Chemical Engineering and Processing - Process Intensification, 216:110410, 2025. DOI: 10.1016/j.cep.2025.110410

    work page doi:10.1016/j.cep.2025.110410 2025

  46. [46]

    Rittig, Elie Akanny, Sandip Bhattacharya, Christina Kohlmann, and Alexander Mitsos

    Christoforos Brozos, Jan G. Rittig, Elie Akanny, Sandip Bhattacharya, Christina Kohlmann, and Alexander Mitsos. Predicting the temperature-dependent CMC of surfactant mixtures with graph neural networks. Computers & Chemical Engineering, 198:109085, 2025. DOI: 10.1016/j.compchemeng.2025.109085

    work page doi:10.1016/j.compchemeng.2025.109085 2025

  47. [47]

    Deep Neural Networks for Multicomponent Molecular Systems.ACS Omega, 5(33): 21042–21053, 2020

    Kyohei Hanaoka. Deep Neural Networks for Multicomponent Molecular Systems.ACS Omega, 5(33): 21042–21053, 2020. DOI: 10.1021/acsomega.0c02599

    work page doi:10.1021/acsomega.0c02599 2020

  48. [48]

    BP-Kelley/descriptastorus: Descriptor computation(chemistry) and (optional) storage for Machine Learning, 2026

    Bp-Kelley. BP-Kelley/descriptastorus: Descriptor computation(chemistry) and (optional) storage for Machine Learning, 2026. URL https://github.com/bp-kelley/descriptastorus. Accessed: May 4, 2026. 12

    2026

  49. [49]

    Mordred: a molecular descriptor calculator.Journal of Cheminformatics, 10(1):4, 2018

    Hirotomo Moriwaki, Yu-Shi Tian, Norihito Kawashita, and Tatsuya Takagi. Mordred: a molecular descriptor calculator.Journal of Cheminformatics, 10(1):4, 2018. DOI: 10.1186/s13321-018-0258-y

    work page doi:10.1186/s13321-018-0258-y 2018

  50. [50]

    Extended-connectivity fingerprints.Journal of Chemical Information and Modeling, 50(5):742–754, 2010

    David Rogers and Mathew Hahn. Extended-Connectivity Fingerprints.Journal of Chemical Information and Modeling, 50(5):742–754, 2010. DOI: 10.1021/ci100050t

    work page doi:10.1021/ci100050t 2010

  51. [51]

    Translation be- tween Molecules and Natural Language

    Carl Edwards, Tuan Lai, Kevin Ros, Garrett Honke, Kyunghyun Cho, and Heng Ji. Translation be- tween Molecules and Natural Language. In Yoav Goldberg, Zornitsa Kozareva, and Yue Zhang, editors, Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 375–413, Abu Dhabi, United Arab Emirates, 2022. Association for Comput...

    work page doi:10.18653/v1/2022.emnlp-main.26 2022

  52. [52]

    Mol-BERT: An Effective Molecular Representation with BERT for Molecular Property Prediction.Wireless Communications and Mobile Computing, 2021(1):7181815, 2021

    Juncai Li and Xiaofei Jiang. Mol-BERT: An Effective Molecular Representation with BERT for Molecular Property Prediction.Wireless Communications and Mobile Computing, 2021(1):7181815, 2021. DOI: 10.1155/2021/7181815

    work page doi:10.1155/2021/7181815 2021

  53. [53]

    Analyzing learned molecular Zhanget al.| AIBuildAI-2 9 representations for property prediction.Journal of Chemical Information and Modeling, 59 (8):3370–3388, 2019

    Kevin Yang, Kyle Swanson, Wengong Jin, Connor Coley, Philipp Eiden, Hua Gao, Angel Guzman-Perez, Timothy Hopper, Brian Kelley, Miriam Mathea, Andrew Palmer, V olker Settels, Tommi Jaakkola, Klavs Jensen, and Regina Barzilay. Analyzing Learned Molecular Representations for Property Prediction. Journal of Chemical Information and Modeling, 59(8):3370–3388, ...

    work page doi:10.1021/acs.jcim.9b00237 2019

  54. [54]

    Schoenholz, Patrick F

    Justin Gilmer, Samuel S. Schoenholz, Patrick F. Riley, Oriol Vinyals, and George E. Dahl. Neural Message Passing for Quantum Chemistry. In Doina Precup and Yee Whye Teh, editors,Proceedings of the 34th International Conference on Machine Learning, volume 70 ofProceedings of Machine Learning Research, pages 1263–1272. PMLR, 2017. URLhttps://proceedings.mlr...

    2017

  55. [55]

    Greenman, Yunsie Chung, Shih-Cheng Li, David E

    Esther Heid, Kevin P. Greenman, Yunsie Chung, Shih-Cheng Li, David E. Graff, Florence H. Vermeire, Haoyang Wu, William H. Green, and Charles J. McGill. Chemprop: A Machine Learning Package for Chemical Property Prediction.Journal of Chemical Information and Modeling, 64(1):9–17, 2024. DOI: 10.1021/acs.jcim.3c01250

    work page doi:10.1021/acs.jcim.3c01250 2024

  56. [56]

    Fabian Jirasek and Hans Hasse. Combining Machine Learning with Physical Knowledge in Thermodynamic Modeling of Fluid Mixtures.Annual Review of Chemical and Biomolecular Engineering, 14(V olume 14, 2023):31–51, 2023. DOI: 10.1146/annurev-chembioeng-092220-025342

    work page doi:10.1146/annurev-chembioeng-092220-025342 2023

  57. [57]

    Gibbs–Helmholtz graph neural network: capturing the temperature dependency of activity coefficients at infinite dilution.Digital Discovery, 2:781–798, 2023

    Edgar Ivan Sanchez Medina, Steffen Linke, Martin Stoll, and Kai Sundmacher. Gibbs–Helmholtz graph neural network: capturing the temperature dependency of activity coefficients at infinite dilution.Digital Discovery, 2:781–798, 2023. DOI: 10.1039/D2DD00142J

    work page doi:10.1039/d2dd00142j 2023

  58. [58]

    BigSolDB 2.0, dataset of solubility values for organic compounds in different solvents at various tempera- tures.Scientific Data, 12(1):1236, 2025

    Lev Krasnov, Dmitry Malikov, Marina Kiseleva, Sergei Tatarin, Sergey Sosnin, and Stanislav Bezzubov. BigSolDB 2.0, dataset of solubility values for organic compounds in different solvents at various tempera- tures.Scientific Data, 12(1):1236, 2025. DOI: 10.1038/s41597-025-05559-8

    work page doi:10.1038/s41597-025-05559-8 2025

  59. [59]

    Machine learning for fuel property predictions: A multi-task and transfer learning approach.SAE Technical Paper Series, 1, 2023

    Tara Larsson, Florence Vermeire, and Sebastian Verhelst. Machine learning for fuel property predictions: A multi-task and transfer learning approach.SAE Technical Paper Series, 1, 2023. DOI: 10.4271/2023-01- 0337

    work page doi:10.4271/2023-01- 2023

  60. [60]

    Graff, Nathan K

    David E. Graff, Nathan K. Morgan, Jackson W. Burns, Anna C. Doner, Brian Li, Shih-Cheng Li, Joel Manu, Angiras Menon, Hao-Wei Pang, Haoyang Wu, Akshat Shirish Zalte, Jonathan W. Zheng, Connor W. Coley, William H. Green, and Kevin P. Greenman. Chemprop v2: An Efficient, Modular Machine Learning Package for Chemical Property Prediction.Journal of Chemical I...

  61. [61]

    DOI: 10.1021/acs.jcim.5c02332

    work page doi:10.1021/acs.jcim.5c02332

  62. [62]

    Kovary, Desmond Gilmour, thibaultvarin r, Jackson Burns, Julien St-Laurent, t, DomInvivo, Saurav Maheshkar, and rbyrne momatx

    Emmanuel Noutahi, Cas Wognum, Hadrien Mary, Honoré Hounwanou, Kyle M. Kovary, Desmond Gilmour, thibaultvarin r, Jackson Burns, Julien St-Laurent, t, DomInvivo, Saurav Maheshkar, and rbyrne momatx. datamol-io/molfeat: 0.9.4, 2023. URLhttps://doi.org/10.5281/zenodo.8373019

    work page doi:10.5281/zenodo.8373019 2023

  63. [63]

    RDKit: Open-source cheminformatics., 2026

    Greg Landrum. RDKit: Open-source cheminformatics., 2026. URL https://www.rdkit.org/. Ac- cessed: May 4, 2026

    2026

  64. [64]

    Deep Sets

    Manzil Zaheer, Satwik Kottur, Siamak Ravanbakhsh, Barnabas Poczos, Ruslan Salakhutdinov, and Alexander Smola. Deep Sets. InAdvances in Neural Information Processing Sys- tems 30, 2017. URL https://proceedings.neurips.cc/paper_files/paper/2017/file/ f22e4747da1aa27e363d86d40ff442fe-Paper.pdf

    2017

  65. [65]

    Attentive Pooling Networks

    Cicero dos Santos, Bing Xiang, and Bowen Zhou. Attentive Pooling Networks.arXiv preprint arXiv:1602.03609, 2016. URLhttps://arxiv.org/abs/1602.03609. 13

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  66. [66]

    Order Matters: Sequence to sequence for sets

    Oriol Vinyals, Samy Bengio, and Manjunath Kudlur. Order Matters: Sequence to Sequence for Sets.arXiv preprint arXiv:1511.06391, 2016. URLhttps://arxiv.org/abs/1511.06391

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  67. [67]

    Xgboost: A scalable tree boosting system

    Tianqi Chen and Carlos Guestrin. XGBoost: A Scalable Tree Boosting System. InProceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD ’16, page 785–794, New York, NY , USA, 2016. Association for Computing Machinery. DOI: 10.1145/2939672.2939785

    work page doi:10.1145/2939672.2939785 2016

  68. [68]

    Massively Parallel Hyperparameter Tuning.CoRR, abs/1810.05934, 2018

    Liam Li, Kevin Jamieson, Afshin Rostamizadeh, Ekaterina Gonina, Moritz Hardt, Benjamin Recht, and Ameet Talwalkar. Massively Parallel Hyperparameter Tuning.CoRR, abs/1810.05934, 2018. URL http://arxiv.org/abs/1810.05934

    work page arXiv 2018

  69. [69]

    Tune: A Research Platform for Distributed Model Selection and Training

    Richard Liaw, Eric Liang, Robert Nishihara, Philipp Moritz, Joseph E Gonzalez, and Ion Stoica. Tune: A Research Platform for Distributed Model Selection and Training.arXiv preprint arXiv:1807.05118, 2018. A Technical Appendices and Supplementary Material This appendix collects the implementation and data details that support the main evaluation. It first ...

    work page internal anchor Pith review Pith/arXiv arXiv 2018