Pith. sign in

REVIEW 4 major objections 5 minor 77 references

Topological Machine Learning for Protein-Nucleic Acid Binding Affinity Changes Upon Mutation

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read This paper claims that persistent Laplacian topological features, combined with physicochemical and protein-Transformer descriptors, let a gradient-boosting model outperform existing predictors of mutation-induced protein-nucleic-acid…

desk verdict A plausible first application of persistent Laplacians to protein-nucleic acid mutation prediction, with code and data released, but the outperformance claim is not yet established due to an unresolved split-protocol question and internally inconsistent DNA numbers. read the letter →

arxiv 2505.22786 v1 pith:I56CE5C7 submitted 2025-05-28 q-bio.QM

Xiang Liu , Junjie Wee , Guo-Wei Wei This is my paper
classification q-bio.QM MSC 92-0892C4055N31
keywords protein-DNAinteractionsprotein-RNAbindingaffinitychangeuponmutationpersistentLaplaciantopologicaldataanalysisproteinlanguagemodelgradientboostingfreeenergy
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

The paper proposes TopoML, a machine-learning model that predicts how a single amino-acid mutation changes the binding free energy ($\Delta\Delta G$) of a protein-DNA or protein-RNA complex. Its distinctive input is a set of persistent Laplacian features built from the geometry around mutation and binding sites, which the authors argue capture both topological shape and quantitative spectral information that sequence and physicochemical features miss. On the S596 protein-DNA benchmark and the 710-mutation protein-RNA benchmark, TopoML reports higher Pearson correlation and lower mean absolute error than earlier dedicated predictors, including PCC 0.72 versus 0.67 on the protein-RNA test set. The authors conclude that topological features are the key addition, and that the integrated model is robust enough to work on combined protein-DNA and protein-RNA data.

What carries the argument

The load-bearing object is the persistent Laplacian $L_p^{s,t}$ of a filtered simplicial complex: a family of Laplacian matrices whose zero-eigenvalue (harmonic) spectrum reproduces persistent homology while the non-zero spectrum adds geometric information. The paper constructs Vietoris-Rips and Alpha complexes on mutation-site, neighborhood, protein-binding, and nucleic-acid-binding atom sets, producing a 5140-dimensional topological vector. This is concatenated with a 783-dimensional physicochemical vector and a 2560-dimensional protein-language-model embedding, and the combined vector is fed to a gradient-boosting tree with 10,000 estimators. The persistent Laplacian does the work of converting three-dimensional mutant and wild-type structures into features that survive local perturbation.

What would settle it

Take protein-nucleic-acid complexes that have experimentally determined structures of both wild-type and mutant bound states, compute TopoML's features from those experimental mutant structures, and compare the predictions with the paper's modeled-structure predictions; a large drop in PCC or MAE would show the result depends on the modeled-structure assumption. A simpler diagnostic is to check whether prediction error grows with the root-mean-square deviation between the modeled and experimentally observed mutant structures.

Watch

Extended reading notes

Core claim

The central claim is that persistent Laplacians carry biologically useful signal for mutation effects on nucleic-acid binding. TopoML computes zero-, one-, and two-dimensional persistent Laplacians over Vietoris-Rips and Alpha complex filtrations on 36 element-specific atom sets around mutation and binding sites, extracts harmonic and non-harmonic spectral statistics from them, and concatenates these topological features with 783 physicochemical descriptors and a 650-million-parameter protein-Transformer sequence embedding. A gradient-boosting tree then predicts $\Delta\Delta G$. The paper reports that on the held-out protein-RNA test set the model reaches PCC 0.72 and MAE 0.77 kcal/mol, above the original benchmark method's 0.67 and 0.81, and that protein-DNA cross-validation reaches PCC 0.67 (5-fold) and 0.681 (10-fold); combining all 1306 mutations gives PCC 0.712 and MAE 0.708 kcal/mol. Ablation experiments show that topological features alone give the best single-feature performance on protein-DNA, and that the full combination is more stable than any single feature type.

Load-bearing premise

The load-bearing premise is that the mutant structures generated by the protein-modeling software in the pipeline resemble the true mutant complexes closely enough that the persistent Laplacian features remain physically meaningful; if the modeled structures deviate substantially from reality, the reported gains may not transfer to real complexes.

Editorial extensions

If this is right

  • On the 115-mutation protein-RNA test set, TopoML reaches PCC 0.72 and MAE 0.77 kcal/mol, beating the original benchmark method's 0.67 and 0.81 under the same split.
  • On the independent MPR79 protein-RNA test set, TopoML reaches PCC 0.53 and RMSE 0.70 kcal/mol, above both energy-based baselines.
  • On protein-DNA S596, TopoML reports PCC 0.67 with 5-fold cross-validation and PCC 0.681 with 10-fold, improving on the prior 5-fold result of 0.65.
  • Combining all 596 protein-DNA and 710 protein-RNA mutations into one dataset yields PCC 0.712 and MAE 0.708 kcal/mol, supporting a single model for both interaction types.

Reading between the lines

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

  • If the claim generalizes, the same persistent-Laplacian-plus-Transformer recipe could be transferred to protein-protein and protein-ligand mutation benchmarks, where structure-derived spectral features have already shown signal; TopoML would then be an instance of a general structural-spectral prediction pipeline.
  • The ablation in which topology alone beats sequence alone on protein-DNA suggests that for interface-focused tasks, explicit three-dimensional geometry may carry more information than a protein sequence model; a direct test would be to replace the modeled mutant structures with experimentally determined mutant structures and see whether the gain persists, shrinks, or grows.
  • The authors note that predicted values have lower variance than experiment; this regression-to-the-mean tendency implies the model may understate extreme destabilizing mutations, so calibrated uncertainty or a non-mean-regressing objective is a natural next step.
  • An extension that needs no new experiments is to evaluate TopoML separately on alanine-scanning data to test the paper's observation that all mutations to alanine show positive $\Delta\Delta G$; if that pattern is an artifact of training-set composition, a balanced resampling should expose it.
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

4 major / 5 minor

Summary. The paper proposes TopoML, a gradient-boosting model that combines persistent-Laplacian topological features, physicochemical descriptors, and ESM-2 protein sequence embeddings to predict mutation-induced binding affinity changes (ΔΔG) for protein-DNA and protein-RNA complexes. The model is evaluated on a 596-mutation protein-DNA dataset and a 710-mutation protein-RNA dataset, with additional analysis on the MPR79 test set. The authors report that TopoML outperforms existing methods, including PRA-MutPred, mCSM-NA, and SAMPDI-3Dv2, on both interaction types, and they provide an ablation study showing that the topological features contribute to performance.

Significance. If the reported results are reliable, the paper would make a useful contribution by extending persistent Laplacian methods to protein-nucleic acid mutation prediction, a setting where existing models achieve only modest accuracy. The work benefits from a clear feature-generation pipeline, a public code/data repository, repeated evaluation over 100 random splits, and a leakage-aware evaluation on the MPR79 test set. The ablation analysis is a strength because it isolates the contribution of each feature type. However, the central outperformance claim is currently undermined by an unclear and potentially leaky split protocol for the main RNA evaluation, by directly contradictory reported DNA metrics, and by the absence of confidence intervals or significance tests for the small reported margins.

major comments (4)
  1. [Section 2.2.3] The main protein-RNA evaluation uses a mutation-level split without demonstrated complex-level separation. The text states that 710 mutations come from 134 complexes and that 595/115 are used for training/testing, but it does not state that mutations from the same complex are confined to one side of the split. Since the feature vector includes 2560-dimensional ESM-2 embeddings of the wild-type and mutant proteins (Section 4.3), two mutations of the same protein have nearly identical sequence embeddings, allowing a tree model to memorize complex-level bias. This could inflate the reported PCC of 0.72 versus PRA-MutPred's 0.67, which was obtained with other feature types. The authors show awareness of this issue in the MPR79 evaluation ('remove any overlapping mutations from our training set to avoid data leakage'), but no similar separation is reported for the main S710 evaluation. Please report results under a complex-level or leave-one-complex-out split, and state explicitly how mutations from the same complex are allocated.
  2. [Section 2.2.4] The protein-DNA 10-fold cross-validation metrics are internally inconsistent. The text first reports 'an improved PCC of 0.681 and an MAE of 0.612 kcal/mol' and later states 'our model achieves a PCC of 0.681 and an MAE of 0.876 kcal/mol using 10-fold cross-validation.' The MAE values differ by 0.264 kcal/mol while the PCC is identical, so the quantitative basis for the DNA outperformance claim is unstable. The dataset name also changes within the same section, from S596 to S569 ('SAMPDI-3Dv2 also performed the 5-fold cross-validation on the S569 dataset'), which further obscures which dataset was used. Please correct these inconsistencies and ensure all reported numbers in the text, tables, and figures are mutually consistent.
  3. [Sections 2.2.3 and 2.2.4] No confidence intervals or significance tests are provided for the reported performance metrics. The authors state that they repeated the process 100 times and averaged the results, but they do not report the variance across repeats. The central outperformance claims rest on small margins (RNA PCC 0.72 vs. 0.67; DNA PCC 0.67 vs. 0.65 under 5-fold CV), which could easily be within sampling noise. Please report standard deviations or confidence intervals for PCC, MAE, and RMSE over the 100 repeats, and provide a statistical test (e.g., paired comparison of predictions on identical folds) for the difference against each baseline.
  4. [Supplementary Information, Section 3] The model relies on Jackal to generate mutant protein-nucleic acid complex structures, and the persistent Laplacian features are computed from these modeled structures. The accuracy of these modeled mutants is not validated anywhere in the manuscript. If Jackal's mutant structures deviate substantially from the true conformations, the topological descriptors would be noisy and the reported predictive gains might not generalize to experimentally resolved structures. Please provide a validation of the modeled structures (e.g., comparison with available experimental mutant structures, or a robustness check using an alternative structure-generation protocol), or explicitly discuss the limitations this imposes on the biological interpretability of the results.
minor comments (5)
  1. [Abstract and Section 1] The abstract and introduction state that the protein-DNA dataset contains 596 mutations, while Section 2.2.4 refers to the 'S569 dataset' in one place. Please use a single, consistent dataset name and size throughout.
  2. [Figure 2 caption] The caption refers to the 'S710 dataset', but the RNA dataset is introduced as a dataset of 710 mutations without a formal S710 name. Consistent naming would avoid confusion.
  3. [Section 2.2.4] There is a typo, 'datset', in the first sentence of this section.
  4. [Section 1] The word 'biomoleculear' in the introduction should be 'biomolecular'.
  5. [Section 4.3] The model name 'esm2 t33 650M UR50D' should be formatted consistently with the ESM-2 literature, and the reference should include the exact model identifier used.

Circularity Check

0 steps flagged · score 2.0 of 10

No definitional circularity; core derivation is self-contained, with only motivational self-citations and protocol caveats that are not circular.

full rationale

TopoML predicts experimental ΔΔG values from features computed independently of the labels: persistent Laplacian spectra from wild-type and mutant structures, physicochemical descriptors, and ESM-2 sequence embeddings. No fitted parameter is relabeled as a prediction, and the gradient boosting tree is trained on a labeled benchmark with external comparators. The identity that the nullity of the persistent Laplacian equals the persistent Betti number is a standard theorem and is not being used to manufacture the prediction. The paper cites several prior works by the same group to motivate persistent Laplacians, e.g., Refs. [25,29,34,35,38], but these citations are not load-bearing: they do not define the features in terms of the target, and the claimed outperformance is evaluated against external methods (PRA-MutPred, mCSM-NA, PremPRI, FoldX, PEMPNI, PNBACE, SAMPDI-3Dv2). The genuine concerns—mutation-level rather than complex-level train/test splits, inconsistent reported DNA MAE values, and reliance on Jackal-generated mutant structures—are correctness or reproducibility risks, not circularity. Some score is given only because of the pervasive self-citation; no predictive step reduces to its own input by construction.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The model relies on standard mathematical background, the accuracy of structural modeling tools, the representational power of a pretrained language model, and the comparability of benchmark protocols. No new physical entities are introduced.

free parameters (2)
  • Gradient boosting hyperparameters = 10000 estimators, max depth 7, min sample split 3, learning rate 1e-3, max features sqrt, subsample 0.5
    These fixed hyperparameters are chosen by hand (Table 3 of the supplementary) and directly affect the reported performance; no tuning procedure is described.
  • Topological feature construction parameters = filtration 0-8 Å with 0.5 Å step; 16 bins; distance cutoffs r for atom sets (exact values not given)
    The persistent Laplacian features depend on the filtration range, step size, binning, and the neighborhood cutoffs r for the four atom sets; these choices are not derived from theory and likely affect the results.
assumptions (4)
  • standard math Persistent Laplacian nullity equals persistent Betti number
    Invoked in Section 4.1.1 to justify using harmonic spectra as topological features.
  • domain assumption Jackal-generated mutant structures are accurate
    Supplementary Section 3 uses Jackal to build mutant protein-nucleic acid complexes; if these modeled structures are wrong, all structure-based features are affected.
  • domain assumption ESM-2 embeddings capture mutation-relevant sequence information
    Section 4.3 uses the ESM-2 language model embeddings as sequence features without fine-tuning on the binding affinity task.
  • domain assumption The experimental databases and train/test splits are reliable and comparable to those used by baseline papers
    Section 2.2.1 relies on S596 and S710 datasets and on the protocols of SAMPDI-3Dv2 and PRA-MutPred; any mismatch in splits undermines the comparison.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Topological Machine Learning for Protein-Nucleic Acid Binding Affinity Changes Upon Mutation." pith.science (2026). https://pith.science/paper/I56CE5C7

@misc{pith2026250522786,
  author       = {Pith},
  title        = {Pith review of: Topological Machine Learning for Protein-Nucleic Acid Binding Affinity Changes Upon Mutation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/I56CE5C7}},
  note         = {Machine review of arXiv:2505.22786}
}
read the original abstract

Understanding how protein mutations affect protein-nucleic acid binding is critical for unraveling disease mechanisms and advancing therapies. Current experimental approaches are laborious, and computational methods remain limited in accuracy. To address this challenge, we propose a novel topological machine learning model (TopoML) combining persistent Laplacian (from topological data analysis) with multi-perspective features: physicochemical properties, topological structures, and protein Transformer-derived sequence embeddings. This integrative framework captures robust representations of protein-nucleic acid binding interactions. To validate the proposed method, we employ two datasets, a protein-DNA dataset with 596 single-point amino acid mutations, and a protein-RNA dataset with 710 single-point amino acid mutations. We show that the proposed TopoML model outperforms state-of-the-art methods in predicting mutation-induced binding affinity changes for protein-DNA and protein-RNA complexes.

Figures

Figures reproduced from arXiv: 2505.22786 by the authors.

Figure 1
Figure 1. Illustration of TopoML model architecture. For each protein-nucleic acid complex, TopoML [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Illustration of model performance for predicting protein-RNA binding free energy changes upon [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Illustration of model performance on predicting mutation-induced protein-DNA binding free energy [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Illustration of persistent Laplacians. (a): a filtration process of the Alpha complex from a point cloud data. (b): the persistence barcodes of zero- and one-dimensional persistent homology (left), and the persistent multiplicity of zero eigenvalues from the Laplacian …

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

77 extracted references · 62 canonical work pages

  1. [1]

    Structural basis of protein-nucleic acid interac- tions

    David L Ollis and Stephen W White. Structural basis of protein-nucleic acid interac- tions. Chemical Reviews, 87(5):981–995, 1987

    work page 1987

  2. [2]

    Protein–rna inter- actions: new genomic technologies and perspectives

    Julian K¨ onig, Kathi Zarnack, Nicholas M Luscombe, and Jernej Ule. Protein–rna inter- actions: new genomic technologies and perspectives. Nature Reviews Genetics , 13(2): 77–83, 2012

    work page 2012

  3. [3]

    Comprehensive review and empirical analysis of hallmarks of dna-, rna-and protein-binding residues in protein chains

    Jian Zhang, Zhiqiang Ma, and Lukasz Kurgan. Comprehensive review and empirical analysis of hallmarks of dna-, rna-and protein-binding residues in protein chains. Brief- ings in bioinformatics , 20(4):1250–1268, 2019

    work page 2019

  4. [4]

    A single amino acid can determine the dna binding specificity of homeodomain proteins

    Jessica Trelsman, Pierre G¨ onczy, Malini Vashishtha, Esther Harris, and Claude Desplan. A single amino acid can determine the dna binding specificity of homeodomain proteins. Cell, 59(3):553–562, 1989. 16

    work page 1989

  5. [5]

    Protein–dna interactions: amino acid conservation and the effects of mutations on binding specificity

    Nicholas M Luscombe and Janet M Thornton. Protein–dna interactions: amino acid conservation and the effects of mutations on binding specificity. Journal of molecular biology, 320(5):991–1009, 2002

    work page 2002

  6. [6]

    Dissecting the expression landscape of rna-binding proteins in human cancers

    Bobak Kechavarzi and Sarath Chandra Janga. Dissecting the expression landscape of rna-binding proteins in human cancers. Genome biology, 15:1–16, 2014

    work page 2014

  7. [7]

    Dna- pkcs structure suggests an allosteric mechanism modulating dna double-strand break repair

    Bancinyane L Sibanda, Dimitri Y Chirgadze, David B Ascher, and Tom L Blundell. Dna- pkcs structure suggests an allosteric mechanism modulating dna double-strand break repair. Science, 355(6324):520–524, 2017

    work page 2017

  8. [8]

    Cancer-associated point mutations in the dlc1 tumor sup- pressor and other rho-gaps occur frequently and are associated with decreased function

    Dunrui Wang, Xiaolan Qian, Beatriz Sanchez-Solana, Brajendra K Tripathi, Marian E Durkin, and Douglas R Lowy. Cancer-associated point mutations in the dlc1 tumor sup- pressor and other rho-gaps occur frequently and are associated with decreased function. Cancer research, 80(17):3568–3579, 2020

    work page 2020

Show all 77 references
  1. [9]

    Frontotemporal dementia-linked p112h mutation of tdp-43 induces protein structural change and impairs its rna binding function

    Sashank Agrawal, Monika Jain, Wei-Zen Yang, and Hanna S Yuan. Frontotemporal dementia-linked p112h mutation of tdp-43 induces protein structural change and impairs its rna binding function. Protein Science, 30(2):350–365, 2021

    2021

  2. [10]

    Surface plasmon resonance: a versatile technique for biosensor applications

    Hoang Hiep Nguyen, Jeho Park, Sebyung Kang, and Moonil Kim. Surface plasmon resonance: a versatile technique for biosensor applications. Sensors, 15(5):10481–10510, 2015

    2015

  3. [11]

    Isothermal titration calorimetry

    Adri´ an Vel´ azquez-Campoy, Hiroyasu Ohtaka, Azin Nezami, Salman Muzammil, and Ernesto Freire. Isothermal titration calorimetry. Current protocols in cell biology , 23 (1):17–8, 2004

    2004

  4. [12]

    Recent advances in fret: distance determination in protein–dna complexes

    Alexander Hillisch, Mike Lorenz, and Stephan Diekmann. Recent advances in fret: distance determination in protein–dna complexes. Current opinion in structural biology, 11(2):201–207, 2001

    2001

  5. [13]

    Predicting protein–dna binding free energy change upon missense mutations using modified mm/pbsa approach: Sampdi webserver

    Yunhui Peng, Lexuan Sun, Zhe Jia, Lin Li, and Emil Alexov. Predicting protein–dna binding free energy change upon missense mutations using modified mm/pbsa approach: Sampdi webserver. Bioinformatics, 34(5):779–786, 2018

    2018

  6. [14]

    Sampdi-3d: predicting the effects of protein and dna mutations on protein–dna interactions

    Gen Li, Shailesh Kumar Panday, Yunhui Peng, and Emil Alexov. Sampdi-3d: predicting the effects of protein and dna mutations on protein–dna interactions. Bioinformatics, 37(21):3760–3765, 2021

    2021

  7. [15]

    Further development of sampdi-3d: A machine learning method for predicting binding free energy changes caused by mutations in either protein or dna

    Prawin Rimal, Shamrat Kumar Paul, Shailesh Kumar Panday, and Emil Alexov. Further development of sampdi-3d: A machine learning method for predicting binding free energy changes caused by mutations in either protein or dna. Genes, 16(1):101, 2025. 17

    2025

  8. [16]

    Prempdi estimates and interprets the effects of missense mutations on protein-dna interactions

    Ning Zhang, Yuting Chen, Feiyang Zhao, Qing Yang, Franco L Simonetti, and Minghui Li. Prempdi estimates and interprets the effects of missense mutations on protein-dna interactions. PLoS computational biology, 14(12):e1006615, 2018

    2018

  9. [17]

    Prempri: Predicting the effects of missense mutations on protein–rna interactions

    Ning Zhang, Haoyu Lu, Yuting Chen, Zefeng Zhu, Qing Yang, Shuqin Wang, and Minghui Li. Prempri: Predicting the effects of missense mutations on protein–rna interactions. International journal of molecular sciences , 21(15):5560, 2020

    2020

  10. [18]

    Pra-mutpred: Predicting the effect of point mutations in protein–rna complexes using structural features.Journal of Chemical Information and Modeling , 2025

    K Harini, M Sekijima, and M Michael Gromiha. Pra-mutpred: Predicting the effect of point mutations in protein–rna complexes using structural features.Journal of Chemical Information and Modeling , 2025

    2025

  11. [19]

    mcsm–na: predicting the effects of mutations on protein–nucleic acids interactions

    Douglas EV Pires and David B Ascher. mcsm–na: predicting the effects of mutations on protein–nucleic acids interactions. Nucleic acids research, 45(W1):W241–W246, 2017

    2017

  12. [20]

    Systematic comparison and prediction of the effects of missense mutations on protein-dna and protein-rna interactions

    Yao Jiang, Hui-Fang Liu, and Rong Liu. Systematic comparison and prediction of the effects of missense mutations on protein-dna and protein-rna interactions. PLoS Computational Biology, 17(4):e1008951, 2021

    2021

  13. [21]

    Pnbace: an ensemble algorithm to predict the effects of mutations on protein-nucleic acid binding affinity

    Si-Rui Xiao, Yao-Kun Zhang, Kai-Yu Liu, Yu-Xiang Huang, and Rong Liu. Pnbace: an ensemble algorithm to predict the effects of mutations on protein-nucleic acid binding affinity. BMC biology, 22(1):203, 2024

    2024

  14. [22]

    Computing persistent homology

    Afra Zomorodian and Gunnar Carlsson. Computing persistent homology. In Proceedings of the twentieth annual symposium on Computational geometry , pages 347–356, 2004

    2004

  15. [23]

    Topological data analy- sis

    Charles Epstein, Gunnar Carlsson, and Herbert Edelsbrunner. Topological data analy- sis. Inverse Problems, 27(12):120201, 2011

    2011

  16. [24]

    A topological approach for protein classification

    Zixuan Cang, Lin Mu, Kedi Wu, Kristopher Opron, Kelin Xia, and Guo-Wei Wei. A topological approach for protein classification. Computational and Mathematical Bio- physics, 3(1), 2015

    2015

  17. [25]

    Topologynet: Topology based deep convolutional and multi-task neural networks for biomolecular property predictions

    Zixuan Cang and Guo-Wei Wei. Topologynet: Topology based deep convolutional and multi-task neural networks for biomolecular property predictions. PLoS computational biology, 13(7):e1005690, 2017

    2017

  18. [26]

    Position: Topological deep learning is the new frontier for relational learning

    Theodore Papamarkou, Tolga Birdal, Michael M Bronstein, Gunnar E Carlsson, Justin Curry, Yue Gao, Mustafa Hajij, Roland Kwitt, Pietro Lio, Paolo Di Lorenzo, et al. Position: Topological deep learning is the new frontier for relational learning. In Forty- first International Co...

    2024

  19. [27]

    Mathematical deep learning for pose and binding affinity prediction and ranking in d3r grand challenges

    Duc Duy Nguyen, Zixuan Cang, Kedi Wu, Menglun Wang, Yin Cao, and Guo-Wei Wei. Mathematical deep learning for pose and binding affinity prediction and ranking in d3r grand challenges. Journal of computer-aided molecular design , 33:71–82, 2019. 18

    2019

  20. [28]

    Mathdl: mathematical deep learning for d3r grand challenge 4

    Duc Duy Nguyen, Kaifu Gao, Menglun Wang, and Guo-Wei Wei. Mathdl: mathematical deep learning for d3r grand challenge 4. Journal of computer-aided molecular design , 34:131–147, 2020

    2020

  21. [29]

    Persistent spectral graph.International journal for numerical methods in biomedical engineering , 36(9):e3376, 2020

    Rui Wang, Duc Duy Nguyen, and Guo-Wei Wei. Persistent spectral graph.International journal for numerical methods in biomedical engineering , 36(9):e3376, 2020

    2020

  22. [30]

    Persistent topological Laplacians–a Survey.Mathematics, 13(2):208, 2025

    Xiaoqi Wei and Guo-Wei Wei. Persistent topological Laplacians–a Survey.Mathematics, 13(2):208, 2025

    2025

  23. [31]

    Persistent laplacians: Properties, algorithms and implications

    Facundo M´ emoli, Zhengchao Wan, and Yusu Wang. Persistent laplacians: Properties, algorithms and implications. SIAM Journal on Mathematics of Data Science , 4(2): 858–884, 2022

    2022

  24. [32]

    The algebraic stability for persistent laplacians

    Jian Liu, Jingyan Li, and Jie Wu. The algebraic stability for persistent laplacians. arXiv preprint arXiv:2302.03902, 2023

    2023 arXiv

  25. [33]

    Persistent spectral–based machine learning (perspect ml) for protein-ligand binding affinity prediction

    Zhenyu Meng and Kelin Xia. Persistent spectral–based machine learning (perspect ml) for protein-ligand binding affinity prediction. Science advances, 7(19):eabc5329, 2021

    2021

  26. [34]

    Persistent spectral theory-guided protein engineering

    Yuchi Qiu and Guo-Wei Wei. Persistent spectral theory-guided protein engineering. Nature computational science, 3(2):149–163, 2023

    2023

  27. [35]

    Persistent laplacian projected omicron ba

    Jiahui Chen, Yuchi Qiu, Rui Wang, and Guo-Wei Wei. Persistent laplacian projected omicron ba. 4 and ba. 5 to become new dominating variants. Computers in Biology and Medicine, 151:106262, 2022

    2022

  28. [36]

    A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation

    Menglun Wang, Zixuan Cang, and Guo-Wei Wei. A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation. Nature Machine Intelligence , 2(2):116–123, 2020

    2020

  29. [37]

    Persistent spectral based ensemble learning (perspect-el) for protein–protein binding affinity prediction

    JunJie Wee and Kelin Xia. Persistent spectral based ensemble learning (perspect-el) for protein–protein binding affinity prediction. Briefings in Bioinformatics , 23(2):bbac024, 2022

    2022

  30. [38]

    Hom-complex-based machine learning (hcml) for the prediction of protein–protein binding affinity changes upon mutation

    Xiang Liu, Huitao Feng, Jie Wu, and Kelin Xia. Hom-complex-based machine learning (hcml) for the prediction of protein–protein binding affinity changes upon mutation. Journal of chemical information and modeling , 62(17):3961–3969, 2022

    2022

  31. [39]

    Persistent tor-algebra for protein– protein interaction analysis

    Xiang Liu, Huitao Feng, Zhi L¨ u, and Kelin Xia. Persistent tor-algebra for protein– protein interaction analysis. Briefings in Bioinformatics , 24(2):bbad046, 2023

    2023

  32. [40]

    Greedy function approximation: a gradient boosting machine

    Jerome H Friedman. Greedy function approximation: a gradient boosting machine. Annals of statistics , pages 1189–1232, 2001. 19

    2001

  33. [41]

    Pronab: database for binding affinities of protein–nucleic acid complexes and their mutants

    Kannan Harini, Ambuj Srivastava, Arulsamy Kulandaisamy, and M Michael Gromiha. Pronab: database for binding affinities of protein–nucleic acid complexes and their mutants. Nucleic acids research, 50(D1):D1528–D1534, 2022

    2022

  34. [42]

    Protherm and pronit: thermodynamic databases for proteins and protein–nucleic acid interactions

    MD Shaji Kumar, K Abdulla Bava, M Michael Gromiha, Ponraj Prabakaran, Koji Kitajima, Hatsuho Uedaira, and Akinori Sarai. Protherm and pronit: thermodynamic databases for proteins and protein–nucleic acid interactions. Nucleic acids research, 34 (suppl 1):D204–D206, 2006

    2006

  35. [43]

    dbamepni: a database of alanine mutagenic effects for protein–nucleic acid interactions

    Ling Liu, Yi Xiong, Hongyun Gao, Dong-Qing Wei, Julie C Mitchell, and Xiaolei Zhu. dbamepni: a database of alanine mutagenic effects for protein–nucleic acid interactions. Database, 2018:bay034, 2018

    2018

  36. [44]

    The foldx web server: an online force field

    Joost Schymkowitz, Jesper Borg, Francois Stricher, Robby Nys, Frederic Rousseau, and Luis Serrano. The foldx web server: an online force field. Nucleic acids research , 33 (suppl 2):W382–W388, 2005

    2005

  37. [45]

    A simple definition of structural regions in proteins and its use in analyzing interface evolution

    Emmanuel D Levy. A simple definition of structural regions in proteins and its use in analyzing interface evolution. Journal of molecular biology , 403(4):660–670, 2010

    2010

  38. [46]

    Persistent sheaf Laplacians.Foundations of Data Science, 7(2):446–463, 2025

    Xiaoqi Wei and Guo-Wei Wei. Persistent sheaf Laplacians.Foundations of Data Science, 7(2):446–463, 2025

    2025

  39. [47]

    Quantum persistent ho- mology

    Bernardo Ameneyro, Vasileios Maroulas, and George Siopsis. Quantum persistent ho- mology. Journal of Applied and Computational Topology , 8(7):1961–1980, 2024

    1961

  40. [48]

    Persistent dirac of paths on digraphs and hyper- graphs

    Faisal Suwayyid and Guo-Wei Wei. Persistent dirac of paths on digraphs and hyper- graphs. Foundations of data science (Springfield, Mo.) , 6(2):124, 2024

    2024

  41. [49]

    Random forests

    Leo Breiman. Random forests. Machine learning, 45:5–32, 2001

    2001

  42. [50]

    Xgboost: A scalable tree boosting system

    Tianqi Chen and Carlos Guestrin. Xgboost: A scalable tree boosting system. In Pro- ceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining , pages 785–794, 2016

    2016

  43. [51]

    Stacked generalization

    David H Wolpert. Stacked generalization. Neural networks, 5(2):241–259, 1992

    1992

  44. [52]

    Integration of element specific persistent homology and machine learning for protein-ligand binding affinity prediction

    Zixuan Cang and Guo-Wei Wei. Integration of element specific persistent homology and machine learning for protein-ligand binding affinity prediction. International journal for numerical methods in biomedical engineering , 34(2):e2914, 2018

    2018

  45. [53]

    ¨Uber den h¨ oheren zusammenhang kompakter r¨ aume und eine klasse von zusammenhangstreuen abbildungen

    Leopold Vietoris. ¨Uber den h¨ oheren zusammenhang kompakter r¨ aume und eine klasse von zusammenhangstreuen abbildungen. Mathematische Annalen, 97(1):454–472, 1927

    1927

  46. [54]

    Alpha shapes-a survey

    Herbert Edelsbrunner. Alpha shapes-a survey. In Tessellations in the sciences: Virtues, techniques and applications of geometric tilings . 2011. 20

    2011

  47. [55]

    MIBPB: a software package for electrostatic analysis

    Duan Chen, Zhan Chen, Changjun Chen, Weihua Geng, and Guo-Wei Wei. MIBPB: a software package for electrostatic analysis. Journal of computational chemistry , 32(4): 756–770, 2011

    2011

  48. [56]

    The Journal of Physical Chemistry B , 105 (28):6507–6514, 2001

    Extending the applicability of the nonlinear poisson- boltzmann equation: multiple dielectric constants and multivalent ions. The Journal of Physical Chemistry B , 105 (28):6507–6514, 2001

    2001

  49. [57]

    Improvements to the apbs biomolecular solvation software suite

    Elizabeth Jurrus, Dave Engel, Keith Star, Kyle Monson, Juan Brandi, Lisa E Felberg, David H Brookes, Leighton Wilson, Jiahui Chen, Karina Liles, et al. Improvements to the apbs biomolecular solvation software suite. Protein science, 27(1):112–128, 2018

    2018

  50. [58]

    Biological structure and function emerge from scaling unsupervised learning to 250 million protein sequences

    Alexander Rives, Joshua Meier, Tom Sercu, Siddharth Goyal, Zeming Lin, Jason Liu, Demi Guo, Myle Ott, C Lawrence Zitnick, Jerry Ma, et al. Biological structure and function emerge from scaling unsupervised learning to 250 million protein sequences. Proceedings of the National ...

    2021

  51. [59]

    Prottrans: Toward understanding the language of life through self-supervised learn- ing

    Ahmed Elnaggar, Michael Heinzinger, Christian Dallago, Ghalia Rehawi, Yu Wang, Llion Jones, Tom Gibbs, Tamas Feher, Christoph Angerer, Martin Steinegger, et al. Prottrans: Toward understanding the language of life through self-supervised learn- ing. IEEE transactions on patter...

    2021

  52. [60]

    Evolutionary-scale pre- diction of atomic-level protein structure with a language model

    Zeming Lin, Halil Akin, Roshan Rao, Brian Hie, Zhongkai Zhu, Wenting Lu, Nikita Smetanin, Robert Verkuil, Ori Kabeli, Yaniv Shmueli, et al. Evolutionary-scale pre- diction of atomic-level protein structure with a language model. Science, 379(6637): 1123–1130, 2023

    2023

  53. [61]

    ESES: Software for Eulerian solvent excluded surface, 2017

    Beibei Liu, Bao Wang, Rundong Zhao, Yiying Tong, and Guo-Wei Wei. ESES: Software for Eulerian solvent excluded surface, 2017

    2017

  54. [62]

    Pdb2pqr: an automated pipeline for the setup of poisson–boltzmann electrostatics calculations

    Todd J Dolinsky, Jens E Nielsen, J Andrew McCammon, and Nathan A Baker. Pdb2pqr: an automated pipeline for the setup of poisson–boltzmann electrostatics calculations. Nucleic acids research, 32(suppl 2):W665–W667, 2004

    2004

  55. [63]

    Very fast prediction and rational- ization of pka values for protein–ligand complexes

    Delphine C Bas, David M Rogers, and Jan H Jensen. Very fast prediction and rational- ization of pka values for protein–ligand complexes. Proteins: Structure, Function, and Bioinformatics, 73(3):765–783, 2008

    2008

  56. [64]

    Gapped blast and psi-blast: a new generation of protein database search programs

    Stephen F Altschul, Thomas L Madden, Alejandro A Sch¨ affer, Jinghui Zhang, Zheng Zhang, Webb Miller, and David J Lipman. Gapped blast and psi-blast: a new generation of protein database search programs. Nucleic acids research, 25(17):3389–3402, 1997

    1997

  57. [65]

    Spider2: a package to predict 21 secondary structure, accessible surface area, and main-chain torsional angles by deep neural networks

    Yuedong Yang, Rhys Heffernan, Kuldip Paliwal, James Lyons, Abdollah Dehzangi, Alok Sharma, Jihua Wang, Abdul Sattar, and Yaoqi Zhou. Spider2: a package to predict 21 secondary structure, accessible surface area, and main-chain torsional angles by deep neural networks. Predicti...

    2017

  58. [66]

    Jackal: A protein structure modeling package

    Jason Z Xiang and B Honig. Jackal: A protein structure modeling package. Columbia University and Howard Hughes Medical Institute, New York , 2002

    2002

  59. [67]

    The gudhi library: Simplicial complexes and persistent homology

    Cl´ ement Maria, Jean-Daniel Boissonnat, Marc Glisse, and Mariette Yvinec. The gudhi library: Simplicial complexes and persistent homology. In Mathematical Software–ICMS 2014: 4th International Congress, Seoul, South Korea, August 5-9, 2014. Proceedings 4, pages 167–174. Sprin...

    2014

  60. [68]

    scipy/scipy: Scipy 1.15

    Ralf Gommers, Pauli Virtanen, Matt Haberland, Evgeni Burovski, Tyler Reddy, Warren Weckesser, Travis E Oliphant, David Cournapeau, Andrew Nelson, Pamphile Roy, et al. scipy/scipy: Scipy 1.15. 0. Zenodo, 2024. 22 Supplementary Information 1 Structural Region Types Table 2: Stru...

    2024

  61. [69]

    Atom surface areas within each group are summed up to generate one feature, leading to (5 × 7 + 1) × 3 = 108 features

    Surface Area: Atom-level solvent-excluded surface areas are calculated using ESES [61]. Atom surface areas within each group are summed up to generate one feature, leading to (5 × 7 + 1) × 3 = 108 features

  62. [70]

    For each atom group, both the sum of partial charges and the sum of 23 absolute partial charges are considered

    Partial Charge: Partial charges are computed using PDB2PQR [62] with the AMBER force field. For each atom group, both the sum of partial charges and the sum of 23 absolute partial charges are considered. This leads to 5 × 7 × 3 × 2 + 2 × 3 = 216 features

  63. [71]

    The constant of 1 is used in our computation

    Coulomb Interaction: The Coulomb energy of the i-th atom is computed as the sum of pairwise Coulomb interactions with every other atom according to the following formula: Ci = X j̸=i ke qiqj rij where ke is the Coulomb constant. The constant of 1 is used in our computation. qi...

  64. [72]

    The van der Waals energy of the i-th atom is computed as the sum of pairwise Lennard-Jones potentials with all other atoms

    Van der Waals interaction. The van der Waals energy of the i-th atom is computed as the sum of pairwise Lennard-Jones potentials with all other atoms. Vi = X j̸=i ϵ[( ri + rj rij )12 − 2( ri + rj rij )6] where ri is the atom radius of i-th atom and ϵ is the depth of potential ...

  65. [73]

    The electrostatic solvation free energy for each atom is computed using the Poisson-Boltzmann model through the software MIBPM [55]

    Electrostatic Solvation Free Energy. The electrostatic solvation free energy for each atom is computed using the Poisson-Boltzmann model through the software MIBPM [55]. By summing up all solvation free energies within each group, 5 × 7 × 3 + 3 = 108 features are generated. 2....

  66. [74]

    Neighboring residues within 12 ˚A of the mutation site are considered

    Mutation Neighborhood Amino Acid Composition. Neighboring residues within 12 ˚A of the mutation site are considered. These residues are classified into six categories: hydrophobic, polar, polar uncharged, positively charged, negatively charged, and spe- cial cases. Both the co...

  67. [75]

    The pKa values of residues are computed using PROPKA software [63]

    pKa shifts. The pKa values of residues are computed using PROPKA software [63]. Specifically, the seven ionizable amino acids are considered: ASP, GLU, ARG, LYS, HIS, CYS, and TYR. The pKa values of the mutation site, N-terminal, and C-terminal residues are considered for both...

  68. [76]

    Features are computed from the conservation scores in the PSSM of the mutation site

    Position Specific Scoring Matrix. Features are computed from the conservation scores in the PSSM of the mutation site. The conservation scores are computed by PSI- BLAST [64]. The feature size is 8 × 2 = 16 for wild-type and mutant proteins

  69. [77]

    The SPIDER software [65] is used to compute the probability of residue torsion angle and a residue being in a coil, alpha helix, and beta strand

    Secondary Structure. The SPIDER software [65] is used to compute the probability of residue torsion angle and a residue being in a coil, alpha helix, and beta strand. The features are computed for wild-typ, mutant type, and their difference, resulting in 17 × 3 = 51 features. ...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.