Pith. sign in

REVIEW 3 major objections 7 minor 58 references

Interpretable QSPR Modeling using Recursive Feature Machines and Multi-scale Fingerprints

T0 review · 3 major / 7 minor · reviewed 2026-08-12 · deepseek-v4-flash

Pith's one-line read A kernel machine that learns its own feature weights achieves state-of-the-art molecular solubility prediction and provides chemically interpretable feature importance.

desk verdict The SOTA solubility claims rest on an underspecified correlation-based feature selection that may leak test labels, so the headline numbers are unverified, though the kernel extensions and interpretability analysis are worth a second look. read the letter →

arxiv 2411.14079 v1 pith:SEPAOV2Y submitted 2024-11-21 q-bio.BM

classification q-bio.BM
keywords recursivefeaturemachinesaveragegradientouterproductkernelridgeregressionmolecularsolubilitypredictionmulti-scalehybridfingerprintsimportanceQSPRinterpretablemachinelearning
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 claims that Recursive Feature Machines (RFM) — a fast kernel-ridge model that learns a feature-weighting matrix through the average gradient outer product — can match or beat deep graph neural networks at predicting molecular solubility, while also exposing which substructures drive the prediction. The authors pair RFM with a custom multi-scale hybrid fingerprint (HF) that concatenates global physicochemical descriptors with SMILES-based fragment counts, and report state-of-the-art errors on the ESOL and FreeSolv benchmarks (RMSE 0.24 and 1.12) across nine solubility datasets. They also show that the same learned matrix yields local and global feature-importance scores whose top features overlap substantially with SHAP and permutation importance, and that removing redundant HF features does not degrade accuracy. If these results hold, kernel-based QSPR models become an attractive combination of accuracy, speed, and interpretability for drug-discovery screening.

What carries the argument

The AGOP matrix M, recursively updated as the average gradient outer product of the kernel predictor's output with respect to its inputs, is the load-bearing object. It enters the kernel as $K(x,x') = \exp(-\gamma \sqrt{(x-x')^T M (x-x')})$ for the Laplace case, with analogous modifications for Matern, Gaussian, and rational-quadratic kernels, so that each iteration re-weights features before re-solving kernel ridge regression. The same M produces local importance scores $x_i^T M x_i$ and, averaged over the dataset, global rankings. The companion piece is the multi-scale hybrid fingerprint (HF): 68 global descriptors plus 243 SMILES-fragment counts (41+139 for three smaller Samuel subsets), constructed by SMILES Pair Encoding and functional-group fragmentation, which gives each HF bit a direct chemical meaning and avoids Morgan fingerprint hash collisions.

What would settle it

Re-run RFM-HF on ESOL and FreeSolv with the HF feature screening performed only on the training folds of the same splits and compare the RMSE to the reported 0.24 and 1.12; if the errors increase substantially or the gap to MACCS/Morgan narrows, the SOTA claim is an artifact of leakage.

Watch

Extended reading notes

Core claim

The central discovery is that a kernel machine can acquire deep feature learning through AGOP: in each iteration, RFM solves kernel ridge regression and then updates a matrix M equal to the average outer product of the prediction gradient with respect to the input features; this M is folded into the kernel as a Mahalanobis distance. The paper applies this machinery to molecular solubility prediction for the first time, defining a hybrid fingerprint whose bits are directly interpretable substructures, and obtains results that surpass the reported errors of graph neural networks and graph transformers on ESOL and FreeSolv. The same M matrix doubles as an interpreter: the local score $x_i^T M x_i$ for a molecule and the average over molecules reproduce the global feature rankings that SHAP and permutation importance give, in a fraction of the time. The authors further show that AGOP generalizes beyond the Laplace kernel to Matern, Gaussian, and rational-quadratic kernels, with Matern and Laplace giving the best generalization, and they relate each kernel's overfitting tendency to the eigenvalue spectrum of the learned M matrix.

Load-bearing premise

The hybrid fingerprint is built by a correlation-based screening threshold that may use the target variable on the full dataset; if that selection is not nested inside cross-validation, the reported test RMSE values are inflated by label leakage.

Editorial extensions

If this is right

  • If RFM-HF's reported accuracy is reproducible, kernel ridge regression with AGOP becomes a direct competitor to GNNs and graph transformers for molecular property prediction, at much lower computational cost.
  • The $x_i^T M x_i$ score gives chemists per-molecule, per-substructure attributions without post-hoc explanation methods, so feature importance and prediction come from the same calculation.
  • The redundancy-filtering result implies the HF fingerprint can be aggressively compressed without losing accuracy, which simplifies deployment and interpretation.
  • The kernel comparison suggests AGOP is not Laplace-specific; Matern and Laplace kernels both carry the benefit, and spectral analysis of M can be used to pick among kernels.
  • On the nine solubility datasets, HF representation improves almost every model tried, not just RFM, so hybrid descriptor-fragment fingerprints are a broadly useful representation.

Reading between the lines

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

  • A direct consequence the authors do not state: if the correlation-based HF feature screening is not nested inside the cross-validation loop, the ESOL and FreeSolv RMSE numbers (0.24 and 1.12) are likely optimistic; re-running with screening inside each training fold would settle this without changing the method.
  • The AGOP importance scores could be aggregated across datasets to derive transferable substructure-solubility rules, e.g., fragments that consistently increase solubility, which would give medicinal chemists testable design hypotheses.
  • The same hybrid fingerprint and RFM pipeline could be ported to other endpoints, such as permeability, toxicity, or logP, where bit-level interpretability is equally valuable.
  • The eigenspectrum-based overfitting diagnosis used here for kernels might also serve as a practical diagnostic for choosing between molecular representations, not just kernels.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 7 minor

Summary. The manuscript applies Recursive Feature Machines (RFM) with Average Gradient Outer Product (AGOP) feature learning to Quantitative Structure-Property Relationship (QSPR) modeling for solubility prediction. It introduces a custom multi-scale hybrid fingerprint (HF), compares RFM against traditional ML and deep learning baselines on nine datasets, and claims state-of-the-art results on ESOL and FreeSolv, alongside local and global feature importance analysis. The paper also extends AGOP to Matern, Gaussian, and Rational Quadratic kernels and examines overfitting through kernel spectral properties.

Significance. If the central empirical claims are sound, the paper would be a useful contribution to interpretable QSPR modeling: applying RFM to molecular property prediction is a sensible transfer of recent kernel-feature-learning ideas, the nine-dataset comparison is broad, and the feature importance analysis is a relevant practical extension. The paper also gives explicit credit to the underlying RFM/AGOP theory and presents a concrete interpretability scheme. However, the headline SOTA claim is currently not adequately supported because the HF fingerprint construction is under-specified and a potentially leakage-prone feature screening step is not described relative to the train/test split; the comparison with graph neural networks in Table 15 is also not based on matched evaluation conditions. These issues are fixable but load-bearing.

major comments (3)
  1. [§4.3.3, Tables 11–15] The HF fingerprint is described only as applying "a correlation-based screening threshold" to select 68 descriptors and 243 fragment features, with no statement about whether this threshold is computed inside each cross-validation fold or on the full dataset. If the correlation screening uses the target variable on the full dataset, the reported test RMSE values, including ESOL RMSE 0.24 and FreeSolv RMSE 1.12, are optimistically biased by label leakage. This is the central claim of the paper, so the authors must specify the threshold value, the exact screening protocol, and demonstrate that the screening is nested within the resampling procedure, or re-report the results with leakage-free feature selection.
  2. [Table 15] The SOTA comparison for ESOL and FreeSolv imports MGCN, SchNet, GCN, GIN, DMPNN, ChemBFN, SPMM, ChemRL-GEM, and Uni-Mol results from their original publications, without establishing that the same train/test splits, preprocessing, and evaluation protocol were used. The RFM-HF column reports a single point estimate with no standard deviation or repeated-seed variation, whereas several imported baselines report means with standard deviations. The claimed advantage over "advanced graph neural networks" is therefore not verifiable as a head-to-head benchmark; the authors should either run the baselines under identical splits or clearly state the protocol differences and report variance for their own method.
  3. [§3.2.3, Eq. (10) and §4.3.1] The local feature importance formula is not defined at the resolution needed for the claimed signed per-feature analysis. Equation (10), score(x_i) = x_i^T M x_i, is a scalar for each sample, and because M is an AGOP matrix (a sum of outer products of gradients) it is positive semidefinite, so the scalar quadratic form is nonnegative. The paper's discussion in §4.3.1 of features contributing "positively" or "negatively" to solubility for compounds 29 and 62 (Figure 8) cannot be derived from Eq. (10) unless a different per-feature score is defined. The authors should either specify the per-feature signed score, or soften the local interpretability claims to what Eq. (10) actually supports.
minor comments (7)
  1. [References] Reference [45] duplicates reference [27] (Radhakrishnan et al., Science 2024); the duplicate should be removed and the in-text citation renumbered.
  2. [Table 10] The caption says "Root mean squared error (MSE) results," but the column is RMSE; the acronym should be corrected.
  3. [Tables 4–7] The top-10 overlap between RFM and PI/SHAP on FreeSolv is only 3/10 and 5/10 in Table 6, which does not strongly support the wording "high degree of correlation"; the authors should report a null expectation or statistical significance for these overlaps.
  4. [§4.2 / Tables 9–14] No train/test split ratio, random seed, or repeated-evaluation protocol is stated for the RFM and baseline results; adding these details is essential for reproducibility.
  5. [§4.3.2] The overfitting comparison uses "consistent regularization parameters" across kernels, but the actual values of the regularization parameter and kernel bandwidth are not given, making it hard to judge whether the spectral-decay comparison is confounded by hyperparameter choice.
  6. [§4.3.1, Figure 8] The text refers to compounds "26" and "29" when discussing the figure, while the preceding paragraph identifies the two examples as compounds "29" and "62"; this inconsistency should be fixed.
  7. [Table 3] There is a typo in the MACCS row for Arash ("167bist" instead of "167bits").

Circularity Check

1 steps flagged · score 2.0 of 10

No derivation-level circularity; the central RFM-HF predictive claims rest on external RFM/AGOP theory and benchmark outcomes, with a mild self-referential interpretability validation and an underspecified HF screening step.

  1. other [Section 3.3, 'Demonstrating the Effectiveness of Feature Importance Analysis in RFM'; see also Eq. (10) in Section 3.2.3]
    "To demonstrate the effectiveness of the RFM model in screening feature importance, we applied PI and SHAP to the trained RFM model and compared their analysis results with the directly computed RFM results, finding a high degree of correlation between them."

    The RFM importance score is computed from the same learned matrix M that defines the RFM kernel metric (Eq. 8), and the reference methods PI and SHAP are also applied to this same trained RFM model. Agreement between the three therefore tests whether the RFM's quadratic-form score is consistent with two other post-hoc explainers of the same fitted function; it does not validate against an external ground truth for what features drive solubility. The 'effectiveness' claim is thus supported by a comparison whose reference points are generated from the very model being explained, making the validation self-referential rather than an independent confirmation.

full rationale

The core predictive chain is not circular: RFM is an externally defined algorithm (cited to Radhakrishnan et al.), the kernel is defined in Eq. 8 from input features and the learned M, and the ESOL/FreeSolv RMSE values in Tables 14-15 are benchmark outcomes, not re-statements of the inputs. M is learned from training labels and then used both for prediction and for post-hoc feature scoring; using the same M for interpretation is not a derivation of the predictive results. The only genuinely self-referential element is the validation of the feature importance scores: PI and SHAP are run on the same trained RFM model and compared with the RFM score, so high overlap demonstrates internal consistency among explainers of one model, not external confirmation that those features drive solubility; this does not affect the benchmark predictions. The construction of HF is described only as 'applying a correlation-based screening threshold' with no nested cross-validation statement; if that threshold uses full-data labels, the reported HF RMSE values would be leakage-inflated, but that is a validation-protocol concern, not a circular derivation. The orphan self-citation [43] is not load-bearing.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The central claim rests on several unstated modeling choices, most importantly the correlation-based feature screening for HF and the use of x^T M x as an importance score. These are methodological assumptions rather than fitted physical parameters. No new physical entities are introduced.

free parameters (4)
  • kernel bandwidth gamma inverse length scale = not reported
    Equation 8 defines gamma = 1/L, with L not specified as tuned per dataset.
  • ridge regularization lambda = not reported
    Equation 9 requires lambda; no value or tuning procedure is given.
  • RFM iteration count = not reported
    Recursive updates of M require a stopping rule; none is given.
  • HF correlation screening threshold = not reported (yields 68+243 or 41+139 bits per dataset)
    HF dimensions vary by dataset in Table 3, and the text says a correlation-based screening threshold is applied, implying a label-dependent cutoff.
assumptions (4)
  • domain assumption Neural Feature Ansatz: the neural feature matrix of a trained DNN correlates with AGOP (Equation 2)
    Used to justify RFM feature learning; this is an empirical and theoretical result from refs 26-27, not re-derived in the paper.
  • standard math Kernel ridge regression solution (Equation 7) is the correct minimizer of the regularized loss (Equation 5)
    Standard RKHS theory; accepted without proof.
  • ad hoc to paper x^T M x is a valid feature-importance score for each sample
    Equation 10 is introduced without derivation or formal link to prediction attribution.
  • ad hoc to paper Correlation-based screening of HF features does not leak label information
    The paper applies a correlation-based threshold but does not state that selection is nested inside cross-validation; the validity of the benchmark depends on this.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Interpretable QSPR Modeling using Recursive Feature Machines and Multi-scale Fingerprints." pith.science (2026). https://pith.science/paper/SEPAOV2Y

@misc{pith2026241114079,
  author       = {Pith},
  title        = {Pith review of: Interpretable QSPR Modeling using Recursive Feature Machines and Multi-scale Fingerprints},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SEPAOV2Y}},
  note         = {Machine review of arXiv:2411.14079}
}
read the original abstract

This study pioneers the application of Recursive Feature Machines (RFM) in QSPR modeling, introducing a tailored feature importance analysis approach to enhance interpretability. By leveraging deep feature learning through AGOP, RFM achieves state-of-the-art (SOTA) results in predicting molecular properties, as demonstrated through solubility prediction across nine benchmark datasets. To capture a wide array of structural information, we employ diverse molecular representations, including MACCS keys, Morgan fingerprints, and a custom multi-scale hybrid fingerprint (HF) derived from global descriptors and SMILES local fragmentation techniques. Notably, the HF offers significant advantages over MACCS and Morgan fingerprints in revealing structural determinants of molecular properties. The feature importance analysis in RFM provides robust local and global explanations, effectively identifying structural features that drive molecular behavior and offering valuable insights for drug development. Additionally, RFM demonstrates strong redundancy-filtering abilities, as model performance remains stable even after removing redundant features within custom fingerprints. Importantly, RFM introduces the deep feature learning capabilities of the average gradient outer product (AGOP) matrix into ultra-fast kernel machine learning, to imbue kernel machines with interpretable deep feature learning capabilities. We extend this approach beyond the Laplace Kernel to the Matern, Rational Quadratic, and Gaussian kernels, to find that the Matern and Laplace kernels deliver the best performance, thus reinforcing the flexibility and effectiveness of AGOP in RFM. Experimental results show that RFM-HF surpasses both traditional machine learning models and advanced graph neural networks.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

58 extracted references · 53 canonical work pages

  1. [1]

    Chemical representation learning for toxicity prediction[J]

    Born J, Markert G, Janakarajan N, et al. Chemical representation learning for toxicity prediction[J]. Digital Discovery, 2023, 2(3): 674-691

  2. [3]

    Deep exploration of random forest model boosts the interpretability of machine learning studies of complicated immune responses and lung burden of nanoparticles[J]

    Yu F, Wei C, Deng P, et al. Deep exploration of random forest model boosts the interpretability of machine learning studies of complicated immune responses and lung burden of nanoparticles[J]. Science advances, 2021, 7(22): eabf4130

  3. [4]

    Machine learning toxicity prediction: latest advances by toxicity end point[J]

    Cavasotto C N, Scardino V . Machine learning toxicity prediction: latest advances by toxicity end point[J]. ACS omega, 2022, 7(51): 47536-47546

  4. [5]

    DeepTox: toxicity prediction using deep learning[J]

    Mayr A, Klambauer G, Unterthiner T, et al. DeepTox: toxicity prediction using deep learning[J]. Frontiers in Environmental Science, 2016, 3: 80

  5. [6]

    Accurate clinical toxicity prediction using multi-task deep neural nets and contrastive molecular explanations[J]

    Sharma B, Chenthamarakshan V , Dhurandhar A, et al. Accurate clinical toxicity prediction using multi-task deep neural nets and contrastive molecular explanations[J]. Scientific Reports, 2023, 13(1): 4908

  6. [7]

    Machine learning -quantitative structure property relationship (ML-QSPR) method for fuel physicochemical properties prediction of multiple fuel types[J]

    Li R, Herreros J M, Tsolakis A, et al. Machine learning -quantitative structure property relationship (ML-QSPR) method for fuel physicochemical properties prediction of multiple fuel types[J]. Fuel, 2021, 304: 121437

  7. [8]

    QSPR for predicting the hydrophile-lipophile balance (HLB) of non- ionic surfactants[J]

    Wu J, Yan F, Jia Q, et al. QSPR for predicting the hydrophile-lipophile balance (HLB) of non- ionic surfactants[J]. Colloids and Surfaces A: Physicochemical and Engineering Aspects, 2021, 611: 125812

  8. [9]

    Topological indices and QSPR modeling of some novel drugs used in the cancer treatment

    Havare ÖÇ. Topological indices and QSPR modeling of some novel drugs used in the cancer treatment. International Journal of Quantum Chemistry. 2021 Dec 15;121(24):e26813

Show all 58 references
  1. [10]

    Molecular descriptors[J]

    Consonni V , Todeschini R. Molecular descriptors[J]. Recent advances in QSAR studies: methods and applications, 2010: 29-102

  2. [11]

    One molecular fingerprint to rule them all: drugs, biomolecules, and the metabolome[J]

    Capecchi A, Probst D, Reymond J L. One molecular fingerprint to rule them all: drugs, biomolecules, and the metabolome[J]. Journal of cheminformatics, 2020, 12: 1-15

  3. [12]

    Nevae: A deep generative model for molecular graphs[J]

    Samanta B, De A, Jana G, et al. Nevae: A deep generative model for molecular graphs[J]. Journal of machine learning research, 2020, 21(114): 1-33

  4. [13]

    Evolution of support vector machine and regression modeling in chemoinformatics and drug discovery[J]

    Rodríguez-Pérez R, Bajorath J. Evolution of support vector machine and regression modeling in chemoinformatics and drug discovery[J]. Journal of Computer -Aided Molecular Design, 2022, 36(5): 355-362

  5. [14]

    Support vector machine[M]//Machine learning

    Pisner D A, Schnyer D M. Support vector machine[M]//Machine learning. Academic Press, 2020: 101-121

  6. [15]

    Random forests[M]

    Genuer R, Poggi J M, Genuer R, et al. Random forests[M]. Springer International Publishing, 2020

  7. [16]

    Kernel ridge regression[M]//Empirical inference: Festschrift in honor of vladimir n

    V ovk V . Kernel ridge regression[M]//Empirical inference: Festschrift in honor of vladimir n. vapnik. Berlin, Heidelberg: Springer Berlin Heidelberg, 2013: 105-116

  8. [17]

    Machine learning methods for small data challenges in molecular science[J]

    Dou B, Zhu Z, Merkurjev E, et al. Machine learning methods for small data challenges in molecular science[J]. Chemical Reviews, 2023, 123(13): 8736-8780

  9. [18]

    MLP-based regression prediction model for compound bioactivity[J]

    Qin Y , Li C, Shi X, et al. MLP-based regression prediction model for compound bioactivity[J]. Frontiers in Bioengineering and Biotechnology, 2022, 10: 946329

  10. [19]

    Zhong S, Hu J, Y u X, et al. Molecular image-convolutional neural network (CNN) assisted QSAR models for predicting contaminant reactivity toward OH radicals: Transfer learning, data augmentation and model interpretation[J]. Chemical Engineering Journal, 20 21, 408: 127998

  11. [20]

    Molecular generation with recurrent neural networks (RNNs)[J]

    Bjerrum E J, Threlfall R. Molecular generation with recurrent neural networks (RNNs)[J]. arXiv preprint arXiv:1705.04612, 2017

  12. [21]

    Message passing neural networks[J]

    Gilmer J, Schoenholz S S, Riley P F, et al. Message passing neural networks[J]. Machine learning meets quantum physics, 2020: 199-214

  13. [22]

    Motif -based graph self -supervised learning for molecular property prediction[J]

    Zhang Z, Liu Q, Wang H, et al. Motif -based graph self -supervised learning for molecular property prediction[J]. Advances in Neural Information Processing Systems, 2021, 34: 15870- 15882

  14. [23]

    Explaining the explainer: A first theoretical analysis of LIME[C]//International conference on artificial intelligence and statistics

    Garreau D, Luxburg U. Explaining the explainer: A first theoretical analysis of LIME[C]//International conference on artificial intelligence and statistics. PMLR, 2020: 1287- 1296

  15. [24]

    From explanations to feature selection: assessing SHAP values as feature selection mechanism[C]//2020 33rd SIBGRAPI conference on Graphics, Patterns and Images (SIBGRAPI)

    Marcílio W E, Eler D M. From explanations to feature selection: assessing SHAP values as feature selection mechanism[C]//2020 33rd SIBGRAPI conference on Graphics, Patterns and Images (SIBGRAPI). Ieee, 2020: 340-347

  16. [25]

    A unified approach to interpreting model predictions[J]

    Lundberg S M, Lee S I. A unified approach to interpreting model predictions[J]. Advances in neural information processing systems, 2017, 30

  17. [26]

    Mechanism of feature learning in deep fully connected networks and kernel machines that recursively learn features[J]

    Radhakrishnan A, Beaglehole D, Pandit P, et al. Mechanism of feature learning in deep fully connected networks and kernel machines that recursively learn features[J]. arXiv preprint arXiv:2212.13881, 2022

  18. [28]

    Profile scaling increases the similarity search performance of molecular fingerprints containing numerical descriptors and structural keys

    Xue, Ling, et al. "Profile scaling increases the similarity search performance of molecular fingerprints containing numerical descriptors and structural keys." Journal of chemical information and computer sciences 43.4 (2003): 1218-1225

  19. [29]

    Descriptor generation from Morgan fingerprint using persistent homology

    Ehiro, T. "Descriptor generation from Morgan fingerprint using persistent homology." SAR and QSAR in Environmental Research 35.1 (2024): 31-51

  20. [30]

    P value interpretations and considerations[J]

    Thiese M S, Ronna B, Ott U. P value interpretations and considerations[J]. Journal of thoracic disease, 2016, 8(9): E928

  21. [31]

    False Discovery Rate[J]

    Storey J D. False Discovery Rate[J]. International encyclopedia of statistical science, 2011, 1: 504-508

  22. [32]

    The role of family-wise error rate in determining statistical significance[J]

    Nicholson K J, Sherman M, Divi S N, et al. The role of family-wise error rate in determining statistical significance[J]. Clinical spine surgery, 2022, 35(5): 222-223

  23. [33]

    Permutation importance: a corrected feature importance measure[J]

    Altmann A, Toloşi L, Sander O, et al. Permutation importance: a corrected feature importance measure[J]. Bioinformatics, 2010, 26(10): 1340-1347

  24. [34]

    Extended-connectivity fingerprints

    Rogers, David, and Mathew Hahn. "Extended-connectivity fingerprints." Journal of chemical information and modeling 50.5 (2010): 742-754

  25. [35]

    AqSolDB, a curated reference set of aqueous solubility and 2D descriptors for a diverse set of compounds

    Sorkun, Murat Cihan, Abhishek Khetan, and Süleyman Er. "AqSolDB, a curated reference set of aqueous solubility and 2D descriptors for a diverse set of compounds." Scientific data 6.1 (2019): 143

  26. [36]

    Prediction of organic compound aqueous solubility using machine learning: a comparison study of descriptor -based and fingerprints-based models

    Tayyebi, Arash, et al. "Prediction of organic compound aqueous solubility using machine learning: a comparison study of descriptor -based and fingerprints-based models." Journal of Cheminformatics 15.1 (2023): 99

  27. [37]

    ESOL: estimating aqueous solubility directly from molecular structure

    Delaney, John S. "ESOL: estimating aqueous solubility directly from molecular structure." Journal of chemical information and computer sciences 44.3 (2004): 1000-1005

  28. [38]

    FreeSolv: a database of experimental and calculated hydration free energies, with input files

    Mobley, David L., and J. Peter Guthrie. "FreeSolv: a database of experimental and calculated hydration free energies, with input files." Journal of computer -aided molecular design 28 (2014): 711-720

  29. [39]

    Machine learning with physicochemical relationships: solubility prediction in organic solvents and water

    Boobier, Samuel, et al. "Machine learning with physicochemical relationships: solubility prediction in organic solvents and water." Nature communications 11.1 (2020): 5753

  30. [40]

    Gradient descent induces alignment between weights and the empirical NTK for deep non -linear networks

    Beaglehole, Daniel, Ioannis Mitliagkas, and Atish Agarwala. "Gradient descent induces alignment between weights and the empirical NTK for deep non -linear networks." arXiv preprint arXiv:2402.05271 (2024)

  31. [41]

    Neural tangent kernel: Convergence and generalization in neural networks

    Jacot, Arthur, Franck Gabriel, and Clément Hongler. "Neural tangent kernel: Convergence and generalization in neural networks." Advances in neural information processing systems 31 (2018)

  32. [42]

    Generalization in kernel regression under realistic assumptions

    Barzilai, Daniel, and Ohad Shamir. "Generalization in kernel regression under realistic assumptions." arXiv preprint arXiv:2312.15995 (2023)

  33. [43]

    Neural Network Layer Matrix Decomposition reveals Latent Manifold Encoding an d Memory Capacity

    Shyh-Chang, Ng, Luo, A -Li, and Qiu, Bo. “Neural Network Layer Matrix Decomposition reveals Latent Manifold Encoding an d Memory Capacity.” arXiv preprint arXiv:2309.05968 (2023)

  34. [44]

    Characterizing overfitting in kernel ridgeless regression through the eigenspectrum

    Cheng, Tin Sum, et al. "Characterizing overfitting in kernel ridgeless regression through the eigenspectrum." arXiv preprint arXiv:2402.01297 (2024)

  35. [45]

    Mechanism for feature learning in neural networks and backpropagation -free machine learning models[J]

    Radhakrishnan A, Beaglehole D, Pandit P, et al. Mechanism for feature learning in neural networks and backpropagation -free machine learning models[J]. Science, 2024, 383(6690): 1461-1467

  36. [46]

    Lightgbm: A highly efficient gradient boosting decision tree

    Ke, Guolin, et al. "Lightgbm: A highly efficient gradient boosting decision tree." Advances in neural information processing systems 30 (2017)

  37. [47]

    Xgboost: A scalable tree boosting system

    Chen, Tianqi, and Carlos Guestrin. "Xgboost: A scalable tree boosting system." Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining. 2016

  38. [48]

    Random forests

    Breiman, Leo. "Random forests." Machine learning 45 (2001): 5-32

  39. [49]

    Deep residual learning for image recognition

    He, Kaiming, et al. "Deep residual learning for image recognition." Proceedings of the IEEE conference on computer vision and pattern recognition. 2016

  40. [50]

    Revisiting deep learning models for tabular data

    Gorishniy, Yury, et al. "Revisiting deep learning models for tabular data." Advances in Neural Information Processing Systems 34 (2021): 18932-18943

  41. [51]

    Degeneration of kernel regression with Matern kernels into low-order polynomial regression in high dimension

    Manzhos, Sergei, and Manabu Ihara. "Degeneration of kernel regression with Matern kernels into low-order polynomial regression in high dimension." The Journal of Chemical Physics 160.2 (2024)

  42. [52]

    Comparing support vector machines with Gaussian kernels to radial basis function classifiers

    Scholkopf, Bernhard, et al. "Comparing support vector machines with Gaussian kernels to radial basis function classifiers." IEEE transactions on Signal Processing 45.11 (1997): 2758- 2765

  43. [53]

    Improved msvr-based range-free localization using a rational quadratic kernel function

    Mohammadzadeh, Paria, et al. "Improved msvr-based range-free localization using a rational quadratic kernel function." Electrical Engineering (ICEE), Iranian Conference on. IEEE, 2018

  44. [54]

    Molecular set representation learning

    Boulougouri, Maria, Pierre Vandergheynst, and Daniel Probst. "Molecular set representation learning." Nature Machine Intelligence 6.7 (2024): 754-763

  45. [55]

    A Bayesian Flow Network Framework for Chemistry Tasks

    Tao, Nianze, and Minori Abe. "A Bayesian Flow Network Framework for Chemistry Tasks." arXiv preprint arXiv:2407.20294 (2024)

  46. [56]

    Bidirectional generation of structure and properties through a single molecular foundation model

    Chang, Jinho, and Jong Chul Ye. "Bidirectional generation of structure and properties through a single molecular foundation model." Nature Communications 15.1 (2024): 2323

  47. [57]

    Geometry -enhanced molecular representation learning for property prediction

    Fang, Xiaomin, et al. "Geometry -enhanced molecular representation learning for property prediction." Nature Machine Intelligence 4.2 (2022): 127-134

  48. [58]

    Uni-mol: A universal 3d molecular representation learning framework

    Zhou, Gengmo, et al. "Uni-mol: A universal 3d molecular representation learning framework." (2023)

  49. [59]

    Direct fit to nature: an evolutionary perspective on biological and artificial neural networks

    Hasson, Uri, Samuel A. Nastase, and Ariel Goldstein. "Direct fit to nature: an evolutionary perspective on biological and artificial neural networks." Neuron 105.3 (2020): 416-434

  50. [60]

    Bidirectional generation of structure and properties through a single molecular foundation model

    Chang, Jinho, and Jong Chul Ye. "Bidirectional generation of structure and properties through a single molecular foundation model." Nature Communications 15.1 (2024): 2323. Appendix RMSE - narrow Methods/ Representations MACCS(167bits) Morgan2(2048bits) HF(68+243bits) RFM 0.81...

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.