Pith. sign in

REVIEW 3 major objections 5 minor 63 references

Pure Component Property Estimation Framework Using Explainable Machine Learning Methods

T0 review · 3 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

Pith's one-line read This paper claims that a connectivity-matrix molecular representation, pooled by random forest and selected by adjusted $R^2$, predicts four pure-component properties more accurately than group-contribution features.

desk verdict Solid feature-generation work undermined by best-fold reporting; the quantitative claims are not supported as written. read the letter →

arxiv 2505.09783 v1 pith:2LNYCL4B submitted 2025-05-14 stat.AP cs.LG

classification stat.APcs.LG
keywords purecomponentpropertyestimationconnectivitymatrixgroupcontributionrandomforestfeatureselectionadjustedR2Shapleyvaluesmolecularrepresentationthermodynamicproperties
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tries to establish that a molecular representation based on connectivity matrices—matrices recording each atom's identity and bond order—can replace hand-built group contributions as the input to machine-learned property models. It claims that, after random-forest feature ranking and adjusted $R^2$ selection, the resulting sparse feature vectors support more accurate prediction of normal boiling point, liquid molar volume, critical temperature, and critical pressure. The practical payoff would be a systematic, interpretable way to estimate properties of new and complex molecules without predefined functional-group libraries. On the paper's comparisons, the connectivity-matrix models reduce test-set RMSE by up to 83.8% against the group-contribution baselines and shrink the feature count from 13,316 to about 100 with no loss of accuracy.

What carries the argument

The machinery is the connectivity matrix: a symmetric matrix whose diagonal entries are atomic numbers and whose off-diagonal entries are bond orders (single, double, aromatic, triple as 1, 2, 1.5, 3). Taking submatrices of radius one to four around each atom, the framework classifies each submatrix by its determinant and eigenvalue set, labels each distinct class, and counts how often each class appears; those counts are the sparse molecular feature vector. Random-forest impurity reduction ranks the features, and the adjusted $R^2$ of the resulting ANN model decides how many top-ranked first- and higher-order features to keep. Shapley values then assign each retained feature a marginal contribution to the prediction, converting the counts back into statements about which structural units raise or lower a property.

What would settle it

Take the reported 20-fold test results for the GPR model in Tables S4–S7 and replace the best-fold RMSE used in Tables 8–11 with the mean or median over all folds. If the connectivity-matrix model's mean test RMSE for critical temperature is not below the group-contribution baseline's mean, the claimed 83.8% reduction disappears.

Watch

Extended reading notes

Core claim

The central claim is that the connectivity-matrix representation captures the local chemical environment of every atom—first, second, third, and fourth bonding shells—by extracting submatrices, classifying them by determinant and eigenvalues, and counting their frequencies, so that the one-to-one mapping between molecular structure and predicted property is stronger than the mapping produced by group-contribution features. With the same 424-feature budget, the paper finds that both a Gaussian process regressor and an artificial neural network trained on these features beat the corresponding group-contribution models on the test set for all four target properties; for critical temperature the test RMSE drops by 83.8%. It also reports that random-forest feature pooling plus adjusted $R^2$ selection reduces the full feature set from 13,316 to 100 features without degrading prediction, and that Shapley-value analysis ties the leading features to chemically meaningful structures such as benzene-ring carbons and chain methylene groups. The paper therefore claims not just a better fit but a more interpretable and more transferable route from molecular structure to pure-component properties.

Load-bearing premise

The load-bearing assumption is that the cross-validation fold with the lowest test error, rather than an average fold, represents the model's performance; the paper applies this rule consistently, while the supplementary tables show test errors varying widely across folds for the same model.

Editorial extensions

If this is right

  • With the feature dimension fixed at 424, connectivity-matrix features outperform group-contribution features inside both GPR and ANN models for $T_b$, $L_{mv}$, $T_c$, and $P_c$ on the test set.
  • The framework needs no predefined functional-group library; it can in principle generate features for any molecule that can be written as SMILES.
  • Feature pooling cuts the full 13,316-feature vector to around 100 features for the ANN without losing predictive accuracy, which lowers computational cost and improves interpretability.
  • Shapley analysis links the most influential features to identifiable structural units such as aromatic carbons, chain methylenes, and C–H environments, with directions that match physical reasoning about intermolecular forces.
  • The paper proposes that the same approach can serve as the property-estimation layer for mixture reconstruction and process integration modeling.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the reported advantage is evaluated across all cross-validation folds rather than the single best fold, the size of the improvement may shrink; the supplementary fold-level tables provide the data to test this directly.
  • Because the feature semantics are chemically interpretable, the same representation could support transfer learning to properties with few experimental measurements.
  • The lower joint conditional entropy reported for the connectivity-matrix dataset suggests that the representation will help isomeric discrimination; a targeted test on isomer sets would make this explicit.
  • Pooling features separately per property leaves open the possibility of multi-task models that share a common feature basis across properties, and the Shapley overlap between $T_b$ and $T_c$ suggests such sharing could be efficient.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes an end-to-end framework for estimating pure-component properties (normal boiling point Tb, liquid molar volume Lmv, critical temperature Tc, and critical pressure Pc) from molecular structure. Molecular structures are converted to connectivity-matrix-based submatrix features, random forest is used for supervised feature ranking and pooling, adjusted R2 is used to select the number of retained features, and ANN and GPR models are trained and compared with group-contribution (GC) baselines. Shapley values are used to interpret the most influential molecular substructures. The central claim is that the connectivity-matrix representation, combined with RF pooling and adjusted-R2 selection, outperforms GC-based ML models, with test RMSE reductions up to 83.8% and a feature-count reduction from 13,316 to 100 without loss of accuracy.

Significance. If the central claim were established, the framework would be a useful, interpretable alternative to GC-based property estimation, particularly because the representation is generated automatically from SMILES and the SHAP analysis connects features to molecular substructures. The paper deserves credit for comparing against external GC baselines from the literature and for reporting full per-fold results in the supplementary tables. However, the headline quantitative claims are currently not supported: every reported comparison uses the single best cross-validation fold, and the abstract's feature-count claim is contradicted by the feature counts used in the results tables. These issues are load-bearing because the claimed superiority over GC models rests entirely on these numbers.

major comments (3)
  1. [§5.2, Tables 8–11 and Tables S4–S7] All headline comparisons select the cross-validation fold with the lowest test RMSE. Section 5.2 states 'We show the fold with the lowest RMSE on the test set among the six folds as the optimal one,' and Tables 8–11 apply the same rule to 20-fold GPR. This is not a minor reporting choice: Table S6 shows GPR-CM Tb test RMSE ranging from 20.24 K (fold 14) to 119.50 K (fold 17), with R2 from 0.39 to 0.94; Tables S4, S5, and S7 show similarly large spreads. Reporting the best order statistic as the model's performance, and using the test set to select that fold, makes the test set a model-selection set and biases the reported metrics optimistically. The claimed RMSE reductions of 39.7–83.8% in the abstract and conclusion are therefore not established; a fold-averaged comparison with mean and standard deviation is required before any superiority claim can be made.
  2. [Abstract; §5.2, Fig. 8 and Fig. 9] The abstract claims that the feature pooling method reduces the number of features from 13,316 to 100 without compromising accuracy, but all reported prediction results use 160 features for ANN-CM (Tables 2, 3, and 7) and 424 features for GPR-CM (Tables 8–11). The adjusted-R2 maximum at 100 features in Fig. 9 is not the configuration used in any of the reported predictions. This internal inconsistency directly undermines the 'feature reduction to 100' claim and needs to be resolved: either the abstract should be corrected to reflect the 160-feature model actually used, or the results should be recomputed for the 100-feature configuration.
  3. [§3.3.1, Eq. (8)] Adjusted R2 as defined in Eq. (8) is a degree-of-freedom correction for linear models where k is the number of estimated parameters. Applying it to ANN and GPR models with k set to the number of input features is not a valid complexity penalty, because the model complexity of an ANN with 500 hidden neurons or a GPR with optimized kernel hyperparameters is not equal to the number of input features. While this issue does not directly affect the reported best-fold comparisons because the final models use a fixed feature count, it calls into question the stated rationale for the adjusted-R2-based feature selection.
minor comments (5)
  1. [§3.1 and §3.3.2, Eq. (20)] Equation (20) is used twice: once for joint conditional entropy in §4.1 and once for the RBF kernel in §3.3.2. The equations should be renumbered to avoid ambiguity.
  2. [§5.3, Table 15] The text near Fig. 14 refers to 'Table 12' when presenting the correspondence between molecular structures and features, but the table is numbered Table 15 in the manuscript; the cross-reference should be corrected.
  3. [Table 9] The GPR-GC-Alshehri row reports an all-dataset RMSE of '0.00[61]' for Lmv, which appears to be a typographical error; the actual value should be stated.
  4. [§5.2, Eq. (21)] Eq. (21) defines the 'relative absolute error' (RAE), but the text refers to it as 'relative absolute percentage error' (RAPE); the terminology should be consistent.
  5. [Throughout] The manuscript would benefit from a data- and code-availability statement; none is currently provided, which makes it difficult to reproduce the reported feature-construction and model-training pipeline independently.

Circularity Check

1 steps flagged · score 3.0 of 10

Best-fold, test-selected reporting forces the headline RMSE reductions; the core CM-vs-GC comparison remains a genuine out-of-sample result with independent content.

  1. fitted input called prediction [Section 5.2, Fig. 8, Tables 2–3 and 8–11; headline metrics quoted in the Abstract and Conclusion.]
    "We show the fold with the lowest RMSE on the test set among the six folds as the optimal one. ... The results shown Tables 8-11 correspond to the fold with the lowest test RMSE in the 20-fold cross-validation."

    The reported test-set results are selected by querying the test set itself: the reported fold is the one with the lowest test RMSE, and Fig. 8 likewise selects 160 features where 'R2 attains its maximum value' and 'RMSE reaches its minimum value' on the test set. The headline reductions (Tb 39.7%, Lmv 66.7%, Tc 83.8%, Pc 63.0%) are thus minima over folds, not typical performance: Table S6 shows GPR-CM Tb test RMSE spanning 20.24 to 119.50 K (R2 0.39 to 0.94) across the 20 folds, so the quoted 20.24 K is the lucky tail. Because within-fold predictions are out-of-sample and the paper trains its own GC baselines, the qualitative CM-over-GC ordering is independent; the circularity is confined to the magnitude of the claimed improvements.

full rationale

The central claim—CM-based features outperform GC-based features for Tb, Lmv, Tc, and Pc—rests on genuinely out-of-sample predictions: the paper trains its own ANN-GC and GPR-GC baselines under the same 95/5 splits and compares test-set RMSE/R2 with CM features (Tables 7–11), also importing reference numbers from Cao et al. [17] and Alshehri et al. [10,61]. No definitional circularity is present: CM features are computed from canonical SMILES via RDKit connectivity matrices, independently of the target property values, and no target is defined in terms of the model output. The connectivity-matrix representation is taken from the authors' own prior paper [35] (Jie Li co-authors both), but this self-citation is a method citation, not load-bearing evidence, because the present paper re-evaluates CM versus GC features with fresh held-out experiments. The SHAP analysis interprets the fitted model rather than deriving a prediction; the mechanistic statements (e.g., longer carbon chains raise Tb and lower Pc) are external chemical knowledge, so this is not circular. The one genuine circularity-adjacent step is the test-set-based selection protocol: the reported fold (and feature count) is chosen by the lowest test RMSE, making the headline 'up to 83.8%' claims statistically forced maxima rather than typical performance. This inflates the apparent improvements and conflicts with the paper's statement that the test set is 'used exclusively to evaluate the model's final performance,' but it does not make the core representation comparison circular. The abstract's 13,316-to-100 feature claim also mismatches the executed experiments (160 features for ANN-CM, 424 for GPR-CM), which is an internal-consistency problem rather than a circularity. Overall score 3: one partial fitted-input-called-prediction issue, with the central derivation otherwise self-contained.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

No new physical entities are invented. The central claim rests on a representational assumption inherited from the authors' prior connectivity matrix work, plus several evaluation choices: the feature count, ANN width, and the use of adjusted R2 as a complexity penalty. The most consequential premise is that the best cross-validation fold represents generalization, which the paper's own supplementary tables contradict.

free parameters (6)
  • Selected feature subset size = 160 (body) vs 100 (abstract)
    The feature count is chosen by maximizing adjusted R2 on the training and validation data, but the abstract claims 100 while all results tables use 160. Reported RMSE values depend on this choice.
  • ANN hidden layer width = 500 neurons
    Chosen by scanning 50 to 600 neurons and selecting the configuration with the lowest best fold test RMSE, as shown in Table 6.
  • ANN learning rate = 0.0001
    Set based on preliminary experiments in Section 4.3; it affects convergence and final predictions.
  • RF hyperparameters = max_depth=30, min_samples_leaf=2, min_samples_split=2, n_estimators=500
    Optimized by GridSearchCV on the training data; they control feature importance ranking and therefore which features are selected.
  • GPR kernel initial length scales = 0.5, 1, 2, 5
    Adopted from Alshehri et al. [10] for fair comparison and then optimized by marginal likelihood; the initial values affect the optimization trajectory.
  • Feature addition step and starting sizes = M0=20, N0=20, step=20
    Arbitrary grid for incremental feature subset construction; the selected optimum may depend on this grid.
assumptions (5)
  • domain assumption The determinant and eigenvalue set of a connectivity submatrix uniquely identifies a local atomic environment.
    Invoked in Section 3.1 to group submatrices into feature categories. No proof or failure analysis is given, and the numerical encoding of bond types (1, 1.5, 2, 3) is arbitrary.
  • domain assumption The frequency of occurrence of first through fourth order connectivity submatrices contains sufficient chemical information to predict Tb, Tc, Pc, and Lmv.
    Central representational assumption; only empirical validation from prior work [35] and this paper is offered, with no theoretical guarantee.
  • ad hoc to paper Adjusted R2 with k equal to the number of input features is a valid model complexity penalty for ANN and GPR.
    Equation (8) applies a linear regression degrees of freedom correction to nonlinear models with many more parameters than features; this is not justified in the text.
  • domain assumption The experimental property values in the Alshehri dataset are accurate and comparable across studies.
    Used as ground truth; the paper acknowledges differing sample sizes with Cao et al. [17] but still compares RMSE values directly.
  • ad hoc to paper The cross-validation fold with the lowest test RMSE is a representative measure of model performance.
    Section 5.2 and Tables 8-11 report only the best fold; Table S6 shows Tb test RMSE ranging from 20.24 to 119.50 K across folds, so the best fold is not representative.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Pure Component Property Estimation Framework Using Explainable Machine Learning Methods." pith.science (2026). https://pith.science/paper/2LNYCL4B

@misc{pith2026250509783,
  author       = {Pith},
  title        = {Pith review of: Pure Component Property Estimation Framework Using Explainable Machine Learning Methods},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2LNYCL4B}},
  note         = {Machine review of arXiv:2505.09783}
}
read the original abstract

Accurate prediction of pure component physiochemical properties is crucial for process integration, multiscale modeling, and optimization. In this work, an enhanced framework for pure component property prediction by using explainable machine learning methods is proposed. In this framework, the molecular representation method based on the connectivity matrix effectively considers atomic bonding relationships to automatically generate features. The supervised machine learning model random forest is applied for feature ranking and pooling. The adjusted R2 is introduced to penalize the inclusion of additional features, providing an assessment of the true contribution of features. The prediction results for normal boiling point (Tb), liquid molar volume, critical temperature (Tc) and critical pressure (Pc) obtained using Artificial Neural Network and Gaussian Process Regression models confirm the accuracy of the molecular representation method. Comparison with GC based models shows that the root-mean-square error on the test set can be reduced by up to 83.8%. To enhance the interpretability of the model, a feature analysis method based on Shapley values is employed to determine the contribution of each feature to the property predictions. The results indicate that using the feature pooling method reduces the number of features from 13316 to 100 without compromising model accuracy. The feature analysis results for Tb, Tc, and Pc confirms that different molecular properties are influenced by different structural features, aligning with mechanistic interpretations. In conclusion, the proposed framework is demonstrated to be feasible and provides a solid foundation for mixture component reconstruction and process integration modelling.

Figures

Figures reproduced from arXiv: 2505.09783 by the authors.

Figure 2
Figure 2. Data distribution and kernel density estimation The molecular sample features in the original dataset were constructed using the GC method, with each molecule represented by a 424-dimensional sparse vector [10]. We refer to this as the GC-dataset. To further characterize the diversity within the GC – dataset, we first count the number of molecules that share the same label but have different feature vectors when usi… view at source ↗
Figure 3
Figure 3. Connectivity matrix of meta-xylene. Taking atom 1 in meta-xylene as an example, its first-order and higher-order features are derived based on its connectivity relationships, as shown in [PITH_FULL_IMAGE:figures/full_fig_p021_3.png] view at source ↗
Figure 4
Figure 4. First-order and higher-order features of atom 1 in meta-xylene [PITH_FULL_IMAGE:figures/full_fig_p022_4.png] view at source ↗
Figures from the paper (5 more)
Figure 5
Figure 5. Figure 5: First-order features for some molecules in the dataset Figs. 6–7 present the feature importance results of the first-order and higher-order feature matrices based on the RF model for Tb prediction. The feature contributions for both matrices are quantified separately w…
Figure 6
Figure 6. Figure 6: Feature importance results for the Top 20 first-order features in Tb prediction [PITH_FULL_IMAGE:figures/full_fig_p023_6.png]
Figure 7
Figure 7. Figure 7: Feature importance results for the higher-order features in Tb prediction [PITH_FULL_IMAGE:figures/full_fig_p023_7.png]
Figure 9
Figure 9. Figure 9 [PITH_FULL_IMAGE:figures/full_fig_p024_9.png]
Figure 10
Figure 10. Figure 10: Parity plot and relative error distribution of the fifth fold for Tb prediction. For Lmv, Tc and Pc, we employ the same modeling approach as for Tb, using the six-fold cross validation. The results are presented in Tables S1 – S3 of the Supplementary Material. We show…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

63 extracted references · 61 canonical work pages

  1. [1]

    Morris, Wetter is better for peat carbon, Nat

    P.J. Morris, Wetter is better for peat carbon, Nat. Clim. Chang. 11 (2021) 561–562

  2. [2]

    Y. Yang, B. Chen, PSE in China: retrospect and prospects, Huagong Jinzhan/Chemical Ind. Eng. Prog. 41 (2022) 3991–4008

  3. [3]

    Wu Y, Molecular management for refining operations, University of Manchester, 2010

  4. [4]

    D. Guan, L. Zhang, Initial guess estimation and fast solving of petroleum complex molecular reconstruction model, AIChE J. 68 (2022) e17782

  5. [5]

    H. Gu, J. Li, P. Mu, Q. Zhu, Improving the Operational Efficiency of Ethylene Cracking Integrated with Refining by Molecular Management, Ind. Eng. Chem. Res. 59 (2020) 13160–13174

  6. [6]

    Vogt, B.M

    E.T.C. Vogt, B.M. Weckhuysen, The refinery of the future, Nature 629 (2024) 295–306

  7. [7]

    V. Mann, R. Gani, V. Venkatasubramanian, Group contribution- based property modeling for chemical product design: A perspective in the AI era, Fluid Phase Equilib. 568 (2023) 113734

  8. [8]

    Y. Wei, L. Shan, T. Qiu, D. Lu, Z. Liu, Machine learning- assisted retrosynthesis planning: Current status and future prospects, Chinese J. Chem. Eng. 77 (2025) 273–292

Show all 63 references
  1. [9]

    Marrero, R

    J. Marrero, R. Gani, Group-contribution based estimation of pure component properties, Fluid Phase Equilib. 183–184 (2001) 183–208

  2. [10]

    Alshehri, A.K

    A.S. Alshehri, A.K. Tula, F. You, R. Gani, Next generation pure component property estimation models: With and without machine learning techniques, AIChE J. 68 (2022)

  3. [11]

    Gani, Group contribution- based property estimation methods: advances and perspectives, Curr

    R. Gani, Group contribution- based property estimation methods: advances and perspectives, Curr. Opin. Chem. Eng. 23 (2019) 184–196

  4. [12]

    Fredenslund, Vapor -liquid equilibria using UNIFAC: a group- contribution method, Elsevier, 2012

    A. Fredenslund, Vapor -liquid equilibria using UNIFAC: a group- contribution method, Elsevier, 2012

  5. [13]

    Abrams, J.M

    D.S. Abrams, J.M. Prausnitz, Statistical thermodynamics of liquid mixtures: A new expression for the excess Gibbs energy of partly or completely miscible systems, AIChE J. 21 (1975) 116–128

  6. [14]

    Aouichaoui, F

    A.R.N. Aouichaoui, F. Fan, J. Abildskov, G. Sin, Application of interpretable group - embedded graph neural networks for pure compound properties, Comput. Chem. Eng. 176 (2023) 108291

  7. [15]

    Burger, V

    J. Burger, V. Papaioannou, S. Gopinath, G. Jackson, A. Galindo, C.S. Adjiman, A hierarchical method to integrated solvent and process design of physical CO2 absorption using the SAFT-γ Mie approach, AIChE J. 61 (2015) 3249–3269. 42

  8. [16]

    T. Zhou, Z. Song, X. Zhang, R. Gani, K. Sundmacher, Optimal Solvent Design for Extractive Distillation Processes: A Multiobjective Optimization -Based Hierarchical Framework, Ind. Eng. Chem. Res. 58 (2019) 5777–5786

  9. [17]

    X. Cao, M. Gong, A. Tula, X. Chen, R. Gani, V. Venkatasubramanian, An Improved Machine Learning Model for Pure Component Property Estimation, Engineering 39 (2024) 61–73

  10. [18]

    Aouichaoui, F

    A.R.N. Aouichaoui, F. Fan, S.S. Mansouri, J. Abildskov, G. Sin, Combining Group- Contribution Concept and Graph Neural Networks Toward Interpretable Molecular Property Models, J. Chem. Inf. Model. 63 (2023) 725–744

  11. [19]

    Gani, P.M

    R. Gani, P.M. Harper, M. Hostrup, Automatic Creation of Missing Groups through Connectivity Index for Pure -Component Property Prediction, Ind. Eng. Chem. Res. 44 (2005) 7262–7269

  12. [20]

    Visco, R.S

    D.P. Visco, R.S. Pophale, M.D. Rintoul, J.L. Faulon, Developing a methodology for an inverse quantitative structure -activity relationship using the signature molecular descriptor, J. Mol. Graph. Model. 20 (2002) 429–438

  13. [21]

    Rogers, M

    D. Rogers, M. Hahn, Extended-Connectivity Fingerprints, J. Chem. Inf. Model. 50 (2010) 742–754

  14. [22]

    Dobbelaere, Y

    M.R. Dobbelaere, Y. Ureel, F.H. Vermeire, L. Tomme, C. V. Stevens, K.M. Van Geem, Machine Learning for Physicochemical Property Prediction of Complex Hydrocarbon Mixtures, Ind. Eng. Chem. Res. 61 (2022) 8581–8594

  15. [23]

    W. Mi, H. Chen, D. Zhu, T. Zhang, F. Qian, Melting point prediction of organic molecules by deciphering the chemical structure into a natural language, Chem. Commun. 57 (2021) 2633–2636

  16. [24]

    Z. Zeng, Y. Yao, Z. Liu, M. Sun, A deep- learning system bridging molecule structure and biomedical text with comprehension comparable to human professionals, Nat. Commun. 13 (2022) 862

  17. [25]

    Goh, N.O

    G.B. Goh, N.O. Hodas, C. Siegel, A. Vishnu, SMILES2Vec: An Interpretable General - Purpose Deep Neural Network for Predicting Chemical Properties, (2017)

  18. [26]

    V. Mann, K. Brito, R. Gani, V. Venkatasubramanian, Hybrid, Interpretable Machine Learning for Thermodynamic Property Estimation using Grammar2vec for Molecular Representation, Fluid Phase Equilib. 561 (2022) 113531

  19. [27]

    Wieder, S

    O. Wieder, S. Kohlbacher, M. Kuenemann, A. Garon, P. Ducrot, T. Seidel, T. Langer, A compact review of molecular property prediction with graph neural networks, Drug Discov. Today Technol. 37 (2020) 1–12. 43

  20. [28]

    B. Dou, Z. Zhu, E. Merkurjev, L. Ke, L. Chen, J. Jiang, Y. Zhu, J. Liu, B. Zhang, G.-W. Wei, Machine Learning Methods for Small Data Challenges in Molecular Science, Chem. Rev. 123 (2023) 8736–8780

  21. [29]

    Ishida, T

    S. Ishida, T. Miyazaki, Y. Sugaya, S. Omachi, Graph Neural Networks with Multiple Feature Extraction Paths for Chemical Property Estimation, Molecules 26 (2021)

  22. [30]

    X. Zang, X. Zhao, B. Tang, Hierarchical Molecular Graph Self-Supervised Learning for property prediction, Commun. Chem. 6 (2023) 1–10

  23. [31]

    Jain, B.C

    S. Jain, B.C. Wallace, Attention is not explanation, ArXiv Prepr. ArXiv1902.10186 (2019)

  24. [32]

    Wiegreffe, Y

    S. Wiegreffe, Y. Pinter, Attention is not not explanation, ArXiv Prepr. ArXiv1908.04626 (2019)

  25. [33]

    Rozemberczki, L

    B. Rozemberczki, L. Watson, P. Bayer, H.-T. Yang, O. Kiss, S. Nilsson, R. Sarkar, The shapley value in machine learning, in: 31st Int. Jt. Conf. Artif. Intell. 25th Eur. Conf. Artif. Intell., International Joint Conferences on Artificial Intelligence Organization, 2022: pp. 5572–5579

  26. [34]

    A. Yang, S. Sun, Y. Su, Z.Y. Kong, J. Ren, W. Shen, Insight to the prediction of CO2 solubility in ionic liquids based on the interpretable machine learning model, Chem. Eng. Sci. 297 (2024) 120266

  27. [35]

    Q. Pan, X. Fan, J. Li, Automatic creation of molecular substructures for accurate estimation of pure component properties using connectivity matrices, Chem. Eng. Sci. 265 (2023) 118214

  28. [36]

    K. Bi, T. Qiu, An intelligent SVM modeling process for crude oil properties prediction based on a hybrid GA-PSO method, Chinese J. Chem. Eng. 27 (2019) 1888–1894

  29. [37]

    Ranjan, A.K

    G.S.K. Ranjan, A.K. Verma, S. Radhika, K -Nearest Neighbors and Grid Search CV Based Real Time Fault Monitoring System for Industries, in: 2019 IEEE 5th Int. Conf. Converg. Technol., 2019: pp. 1–5

  30. [38]

    Y. Wang, P. Wang, K. Tansey, J. Liu, B. Delaney, W. Quan, An interpretable approach combining Shapley additive explanations and LightGBM based on data augmentation for improving wheat yield estimates, Comput. Electron. Agric. 229 (2025) 109758

  31. [39]

    Y. Shi, W. Zhong, X. Peng, M. Yang, W. Du, Interpretable reconstruction of naphtha components using property- based extreme gradient boosting and compositional - weighted Shapley additive explanation values, Chem. Eng. Sci. 284 (2024)

  32. [40]

    Landrum, Rdkit documentation, Release 1 (2013) 4

    G. Landrum, Rdkit documentation, Release 1 (2013) 4. 44

  33. [41]

    Greenacre, P.J.F

    M. Greenacre, P.J.F. Groenen, T. Hastie, A.I. d’Enza, A. Markos, E. Tuzhilina, Principal component analysis, Nat. Rev. Methods Prim. 2 (2022) 100

  34. [42]

    Y. Mu, X. Liu, L. Wang, A Pearson’s correlation coefficient based decision tree and its parallel implementation, Inf. Sci. (Ny). 435 (2018) 40–58

  35. [43]

    Rigatti, Random Forest, J

    S.J. Rigatti, Random Forest, J. Insur. Med. 47 (2017) 31–39

  36. [44]

    Schonlau, R.Y

    M. Schonlau, R.Y. Zou, The random forest algorithm for statistical learning, Stata J. 20 (2020) 3–29

  37. [45]

    F. Jin, Y. Guo, Y. Zhang, X. Ma, B. Liu, J. Jiao, X. Yin, H. Xu, J. Gong, K. Wen, A new gas pipeline network simulation method based on BHC -PINN, Nat. Gas Ind. 45 (2025) 164–174

  38. [46]

    Z. Wang, D. He, H. Nie, Operational optimization of copper flotation process based on the weighted Gaussian process regression and index- oriented adaptive differential evolution algorithm, Chinese J. Chem. Eng. 66 (2024) 167–179

  39. [47]

    Gulli, S

    A. Gulli, S. Pal, Deep learning with Keras, Packt Publishing Ltd, 2017

  40. [48]

    B. Pang, E. Nijkamp, Y.N. Wu, Deep learning with tensorflow: A review, J. Educ. Behav. Stat. 45 (2020) 227–248

  41. [49]

    X. Yin, K. Wen, Y. Wu, X. Han, Y. Mukhtar, J. Gong, A machine learning- based surrogate model for the rapid control of piping flow: Application to a natural gas flowmeter calibration system, J. Nat. Gas Sci. Eng. 98 (2022) 104384

  42. [50]

    X. Yin, K. Wen, W. Huang, Y. Luo, Y. Ding, J. Gong, J. Gao, B. Hong, A high-accuracy online transient simulation framework of natural gas pipeline network by integrating physics-based and data-driven methods, Appl. Energy 333 (2023) 120615

  43. [51]

    K. Wen, J. Jiao, K. Zhao, X. Yin, Y. Liu, J. Gong, C. Li, B. Hong, Rapid transient operation control method of natural gas pipeline networks based on user demand prediction, Energy 264 (2023) 126093

  44. [52]

    Kramer, O

    O. Kramer, O. Kramer, Scikit-learn, Mach. Learn. Evol. Strateg. (2016) 45–53

  45. [53]

    Sundararajan, A

    M. Sundararajan, A. Najmi, The many Shapley values for model explanation, in: Int. Conf. Mach. Learn., PMLR, 2020: pp. 9269–9278

  46. [54]

    H. Cai, Y. Yang, Y. Tang, Z. Sun, W. Zhang, Shapley value -based class activation mapping for improved explainability in neural networks, Vis. Comput. (2025) 1–19

  47. [55]

    Modla, Energy saving methods for the separation of a minimum boiling point azeotrope using an intermediate entrainer, Energy 50 (2013) 103–109

    G. Modla, Energy saving methods for the separation of a minimum boiling point azeotrope using an intermediate entrainer, Energy 50 (2013) 103–109. 45

  48. [56]

    Robinson, D.Y

    D.B. Robinson, D.Y. Peng, S.Y.K. Chung, The development of the Peng - Robinson equation and its application to phase equilibrium in a system containing methanol, Fluid Phase Equilib. 24 (1985) 25–41

  49. [57]

    Constantinou, R

    L. Constantinou, R. Gani, J.P. O’Connell, Estimation of the acentric factor and the liquid molar volume at 298 K using a new group contribution method, Fluid Phase Equilib. 103 (1995) 11–22

  50. [58]

    Chen, A tutorial on kernel density estimation and recent advances, Biostat

    Y.-C. Chen, A tutorial on kernel density estimation and recent advances, Biostat. Epidemiol. 1 (2017) 161–187

  51. [59]

    Shannon, A mathematical theory of communication, Bell Syst

    C.E. Shannon, A mathematical theory of communication, Bell Syst. Tech. J. 27 (1948) 379–423

  52. [60]

    Prechelt, Automatic early stopping using cross validation: quantifying the criteria, Neural Networks 11 (1998) 761–767

    L. Prechelt, Automatic early stopping using cross validation: quantifying the criteria, Neural Networks 11 (1998) 761–767

  53. [61]

    Next generation pure component property estimation models: With and without machine learning techniques,

    A.S. Alshehri, A.K. Tula, F. You, R. Gani, Corrections to “Next generation pure component property estimation models: With and without machine learning techniques,” AIChE J. 69 (2023) e18086

  54. [62]

    Zhou, Machine learning, Springer nature, 2021

    Z.-H. Zhou, Machine learning, Springer nature, 2021

  55. [63]

    Edmister, Thermodynamic Properties of Hydrocarbons, Ind

    W.C. Edmister, Thermodynamic Properties of Hydrocarbons, Ind. Eng. Chem. 30 (1938) 352–358. 1 Supplemental Materials for Pure Component Property Estimation Framework Using Explainable Machine Learning Methods Jianfeng Jiao1, Xi Gao2,3,§ and Jie Li1,* 1Centre for Process Integr...

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.