Pith. sign in

REVIEW 5 major objections 8 minor 59 references

A single multiscale manifold description of binding interfaces predicts affinity for both metal pockets and flat protein surfaces better than prior methods.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-31 00:47 UTC pith:TCUOYTNQ

load-bearing objection Competent dual-benchmark extension of their Hodge/BIG-Laplacian line; real but small gains, and the missing LM-vs-manifold ablation undercuts the geometric story. the 5 major comments →

arxiv 2607.25115 v1 pith:TCUOYTNQ submitted 2026-07-27 q-bio.BM

Persistent Manifold Learning of Protein Properties

Xingjian Xu , Zhe Su , Guo-Wei Wei , Chunmei Wang This is my paper
classification q-bio.BM MSC 92E1055N31
keywords persistent manifold learningde Rham-Hodge theoryHodge LaplacianBoundary-Induced Graph Laplacianbinding affinity predictionmetalloprotein–ligandprotein–protein interactionsBetti numbers
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

How tightly two biomolecules stick together is hard to predict from structure alone, especially when the contact region can be a tight metal-coordinated pocket or a broad, nearly featureless protein surface. This paper argues that much of what sets binding strength is encoded in the shape of that interface, and that one geometric description can serve both classes without hand-crafted, system-specific features. It turns each interface into a nested family of manifolds built from element-specific Gaussian density fields, then reads topology and geometry from the spectra of Boundary-Induced Graph Laplacians (a discrete form of de Rham–Hodge theory). Those manifold embeddings are concatenated with sequence embeddings from protein and molecular language models and fed to gradient boosting trees. On standard metalloprotein–ligand and wild-type protein–protein benchmarks the resulting model beats previously reported state-of-the-art scores, supporting the claim that a shared manifold view of interface shape is enough to drive competitive affinity prediction across dissimilar interaction types.

Core claim

Persistent manifold learning (PML) represents a binding interface as a filtration of manifolds from element-specific atomic density fields and extracts both Betti numbers and nonharmonic spectral geometry via Boundary-Induced Graph Laplacians; when these features are combined with language-model embeddings and gradient boosting, the same pipeline outperforms prior methods on both metalloprotein–ligand and protein–protein binding-affinity benchmarks.

What carries the argument

Boundary-Induced Graph Laplacian (BIG Laplacian) on degree-3 forms under normal boundary conditions: a discrete de Rham–Hodge operator whose zero eigenvalues recover the 0-th Betti number and whose leading nonzero eigenvalues supply multiscale geometric descriptors of each manifold in the filtration.

Load-bearing premise

That the 0-th Betti number plus only the first two nonzero eigenvalues of these Laplacians, taken on nine fixed isovalues of element-pair Gaussian densities, already carry enough binding-relevant shape information for both metal sites and flat protein interfaces.

What would settle it

Retrain and retest on the same held-out metalloprotein–ligand and SKEMPI-WT splits using only the language-model embeddings (no manifold spectra) or only higher/truncated spectra; if the reported PCC gains disappear, the geometric claim is not supported.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • One geometry-first pipeline can replace separate feature engineering for metal-centered pockets and broad protein–protein interfaces.
  • Affinity ranking for virtual screening can be driven by multiscale manifold spectra rather than pocket-specific handcrafted descriptors.
  • The same level-set plus BIG-Laplacian construction is in principle reusable for other structure–sequence quantitative tasks beyond binding free energy.
  • Spectral truncation to leading modes is presented as a resolution-matched regularizer, not a temporary limitation.

Where Pith is reading between the lines

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

  • If the truncated spectra are doing real work, ablating element-specific pairing or the metal-coordination shell cutoff should hurt metalloprotein performance more than PPI performance.
  • The framework’s reliance on static crystal coordinates suggests a natural next test on ensembles or flexible docking poses to see whether leading eigenvalues remain stable.
  • Because language-model embeddings already carry substantial signal, the manifold block is most valuable where sequence alone under-resolves interface packing.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 8 minor

Summary. The manuscript introduces Persistent Manifold Learning (PML) for binding affinity (BA) prediction. A binding interface is converted, via element-specific Gaussian (FRI) density fields, into a filtration of sublevel-set manifolds on a Cartesian grid; the Boundary-Induced Graph (BIG) Laplacian L_{3,n} under normal boundary conditions is assembled on each manifold, and β0 plus the first two nonzero eigenvalues per manifold are used as features. These are concatenated with ESM-2 (protein) and ChemBERTa (ligand) embeddings and regressed with GBDT. On a metalloprotein–ligand benchmark (fixed 1845/618 split) PML reports average PCC 0.753 vs 0.745 for the prior SOTA (CAML); on SKEMPI-WT protein–protein BA it reports average PCC 0.713 (best run 0.731) vs 0.681 for PLNet. Section 2 reviews de Rham–Hodge theory, the Friedrichs theorem, Eulerian DEC discretization, and persistent Hodge Laplacians; the mathematical review is competent and correctly cited.

Significance. If the results hold, the paper demonstrates that a single geometric representation (BIG-Laplacian spectra on density sublevel sets) transfers across two structurally distinct interaction classes — compact metal-coordination sites and flat PPI interfaces — which would be a useful unification. Strengths worth naming: the empirical protocol is better than much of this literature (fixed external test split for MPLI, 10 seeds with mean±std, 10-fold CV, an SVM regressor ablation, tabulated baselines); the mathematical development is a correct, well-cited recall of Hodge/Friedrichs theory and the BIG approximation; and the benchmark claims are concrete and falsifiable. However, the framing claim — that interface geometry drives the SOTA result — is not isolated from the contribution of the language-model embeddings, and on the PPI benchmark PML is worse than PLNet on RMSE, so the current evidence supports a narrower claim than the abstract and Relevance statement make.

major comments (5)
  1. [§4.1.2, §4.2, §5.2; Relevance statement] No ablation separates the manifold features from the LM embeddings they are concatenated with. The only ablation varies the regressor (GBDT vs SVM) with the feature set held fixed. Yet the paper's framing claim is geometric ('much of what determines binding strength is encoded in the shape of the interface itself'; §5 attributes the PPI gain to manifold geometry). On MPLI the margin over CAML is 0.008 PCC (0.753 vs 0.745, Table 2), which a 1664-d ESM-2+ChemBERTa input to a 10,000-tree GBDT could plausibly produce alone; on PPI, PLNet [58] apparently uses no ESM-2, so the comparison confounds representation and feature source, and §5.2 concedes ESM-2 gets 'proportionally greater weight'. LM-only, manifold-only, and leave-one-source-out runs (or GBDT feature-importance per source) are required to support the geometric claim; these are cheap given the existing pipeline.
  2. [Table 3, §4.2, Abstract] On SKEMPI-WT, PML does not outperform PLNet on RMSE: 2.051±0.046 (best 1.996) vs PLNet's 1.533±0.021 — roughly 34% worse — and PML-SVM is worse still (2.127). The statement in §4.2 that 'PML with GBDT consistently achieves the best performance' and the abstract's unqualified 'outperforms state-of-the-art methods' are therefore supported only for PCC. The large RMSE regression despite higher PCC (a calibration/shape issue, possibly from GBDT on 343 samples) needs explicit reporting, qualification of the SOTA claim, and discussion.
  3. [§3.4 vs §5.2; §3.2] The feature accounting is internally inconsistent. §3.4 states 'β0 together with the first 2 non-zero eigenvalues... giving 6 features per manifold' — that is 3 numbers, not 6 — and then computes '3×9×40' (=1080) features per MPLI complex, while §5.2 uses 6×9×9=486 for PPI and quotes 2160 for MPLI (=6×9×40). Which is it, and where does the factor of 2 come from? Similarly §3.2 says PPI pairs are restricted to {C,N,O}×{C,N,O} (=9) but then writes 'pairs between {C,N,O,S} and {C,N,O,S}' (=16). These are basic counts needed to reproduce the input dimensionality.
  4. [§4 (metrics paragraph) vs §3.1/§4.1.1] §4 states that 'PCC and RMSE are computed via 10-fold cross-validation rather than on a single train-test split,' but §3.1 and §4.1.1 describe a fixed 1845/618 train/test partition with 10-fold CV used only for hyperparameter tuning, and Table 2 is a test-set comparison. For SKEMPI-WT the evaluation protocol (CV folds? a held-out split? alignment with PLNet's protocol in [58]) is never stated. Please state precisely which protocol produced each number in Tables 2 and 3, and confirm that all baseline numbers were obtained on identical splits and label conventions (the Table 2 caption says RMSE is on raw pKd labels while the column is labeled kcal/mol).
  5. [§2.3–2.4, Remark 2.13; title/abstract] Sections 2.3–2.4 develop the genuinely persistent Hodge Laplacian ∆_{i,j} (biharmonic extension, commutative diagrams, discrete assembly), but Remark 2.13 states that only the i=j special case is used, stacked over 9 isovalues. The implemented method is therefore single-scale Hodge/BIG spectra evaluated repeatedly along a filtration — closer to the Eulerian method of [47] than to the persistent theory presented. Either the framing/title should be aligned with what is computed, or the authors should justify why stacking i=j spectra constitutes 'persistent' manifold learning in the sense of §2.3 (e.g., what the persistent construction would add and why it is unnecessary here).
minor comments (8)
  1. [§3.2, Eq. (3.1)] The scale parameter τ in ρ(x,τ) is never assigned a value, and the source of the van der Waals radii r_i is not given; both are needed for reproduction. Also, the GBDT implementation is described only as 'the Python library' (§3.6) — name the package and version.
  2. [§5.3] The claim that extending the spectrum to ten eigenvalues 'degrades performance on both benchmarks' is asserted without data. Either show the experiment (a small table would suffice) or soften the claim.
  3. [Table 2 / Figure 3] Table 2 caption says RMSE is computed on raw pKd labels but the column header and figures use kcal/mol; Figure 3(b) caption reads 'PCC 0.756 kcal/mol' — PCC is unitless. Also, comparing the best of 10 seeds (0.756, 0.731) against baselines' reported means is generous; consider a significance test on the average-run margin (0.753±0.002 vs 0.745±0.001).
  4. [§3.3 vs §3.4] §3.3 builds the bounding box per complex from atom coordinate min/max, while §3.4 says 'a fixed Cartesian grid that remains the same across all complexes.' These appear contradictory; clarify whether boxes (and hence grid spacing ℓ) vary per complex, and if so, why spectra remain comparable across complexes given different ℓ.
  5. [§3.3] 'We evaluate the level-set function ρ(x,τ) from Eq. (2.1)' — wrong cross-reference; ρ is defined in Eq. (3.1). Eq. (2.1) is dd=0.
  6. [§4.2 / Table 3] With 343 SKEMPI-WT complexes and ~3000 input features, please state whether model selection used nested CV to avoid leakage, and report the PLNet protocol explicitly enough that the comparison is apples-to-apples (PLNet [58] is the authors' own prior method).
  7. [Declarations] Code is 'available from the authors upon reasonable request.' For a method paper whose claims are empirical, a public repository (GitHub/Zenodo with a DOI) including the feature-extraction pipeline and trained-model scripts would substantially strengthen reproducibility.
  8. [Throughout] Typographical: 'Metallprotein-ligand' (§4.1 heading); 'RMSE are computed' (§4); 'protein-protein' vs 'protein–protein' dash inconsistency; Figure 1 is low-resolution and its panel labels (e.g., 'Hodge Laplaciand') run together.

Circularity Check

1 steps flagged

No load-bearing circularity: BA labels are external experiments; self-citations supply the geometric operators, not the predicted affinities.

specific steps
  1. self citation load bearing [§2.1–2.4; Def. 2.8; citations [6],[41],[47]]
    "Boundary-Induced Graph Laplacian, a discrete realization of de Rham–Hodge theory, then extracts topological invariants together with nonharmonic spectral information... The spectra of the BIG Laplacians have been shown to converge to those of the corresponding Hodge Laplacians up to a scaling factor [41]."

    The geometric descriptor stack (evolutionary/persistent de Rham–Hodge, BIG Laplacian, Eulerian filtration) is justified almost entirely by prior papers with overlapping authors. This is method inheritance, not circular prediction: affinity labels and test metrics remain external. Flagged only as minor non-load-bearing self-citation of the operator family.

full rationale

The paper’s central claim is supervised binding-affinity prediction on held-out experimental labels (metalloprotein–ligand split from [25]/[53]; SKEMPI-WT from [23]), not a first-principles derivation that algebraically equals a fitted constant. Manifold features (FRI Gaussian sublevel sets → BIG Laplacian spectra of L_{3,n}) and LM embeddings are inputs to GBDT; PCC/RMSE are measured against external pK_d / ΔG values. Prior Wei-group citations ([6], [41], [47], [52], etc.) define and discretize the operators used as descriptors; they do not supply the affinity targets or force the reported margins over CAML/PLNet. Comparison to PLNet ([58], overlapping authors) is a same-lab baseline, not a self-definitional loop. Hyperparameters are fixed (Table 1) and performance is averaged over seeds/CV—standard ML, not fitted-input-called-prediction. Minor self-citation of the method stack is present but not load-bearing for the result. Score 1 only for that non-essential self-citation density; no step reduces the SOTA claim to its inputs by construction.

Axiom & Free-Parameter Ledger

7 free parameters · 5 axioms · 2 invented entities

Central performance claims rest on standard differential-geometry theorems plus many modeling and ML knobs (density form, filtration window, spectral truncation, atom-pair chemistry, grid, regressor). No new physical entity is postulated; ‘PML’ and ‘BIG Laplacian’ name computational constructions largely defined in cited prior work. Free parameters dominate what must be trusted for the SOTA tables.

free parameters (7)
  • FRI Gaussian scale τ and van der Waals radii in ρ(x,τ)
    Controls manifold geometry; not fit in a reported sweep here but chosen as part of the density model (§3.2).
  • Isovalue filtration: 9 values in [−0.5, −0.001] = s=9 on [−0.5,−0.001]
    Hand-chosen range and count justified by topology stability vs cost (§3.2); changes feature dimension and multiscale content.
  • Spectral truncation: β0 + first 2 nonzero eigenvalues of L_{3,n} = 2 nonzero eigenvalues
    Explicitly selected; authors state extending to 10 degrades both benchmarks (§3.4, §5.3).
  • Cartesian grid N=100 and 25% bbox padding; 12 Å interface cutoff = N=100, cutoff 12 Å
    Discretization and atom selection hyperparameters that affect spectra (§3.2–3.3).
  • Element-pair schemes (40 MPLI pairs; 9 PPI pairs) and H/S exclusions = 40 / 9 pairs
    Chemistry featurization choices that define which manifolds exist (§3.2).
  • GBDT hyperparameters (Table 1) and 10 random seeds = n_estimators=10000, max_depth=5, subsample=0.5, ...
    Regressor capacity and stochasticity; metrics averaged over seeds but hyperparams fixed without full search report.
  • ESM-2 650M and ChemBERTa-77M pooling (mean final-layer) = 1280-d / 384-d mean pool
    Choice of pretrained models and pooling defines the sequence half of the feature vector (§3.5).
axioms (5)
  • standard math Hodge and Friedrichs theorems: dim ker of normal/tangential Hodge Laplacians equal Betti numbers (relative/absolute cohomology) on compact orientable Riemannian manifolds with boundary.
    Invoked as Theorems 2.3 and 2.6 to justify reading topology from Laplacian kernels.
  • standard math BIG Laplacian spectra converge (up to scaling) to Hodge Laplacian spectra, so identity Hodge stars are valid proxies ([41]).
    Definition 2.8 and surrounding text; underwrites using L^B_{3,n} instead of full Hodge stars.
  • domain assumption FRI negative-sum Gaussians on element-specific atom pairs yield manifolds whose multiscale topology/geometry correlate with binding free energy.
    §3.2–3.4 modeling choice carried from prior FRI/topology papers; not derived from statistical mechanics of binding.
  • domain assumption Wild-type SKEMPI complexes and the [25]/[53] metalloprotein split are adequate, unbiased tests of geometry-based BA models; Profix-repaired structures preserve affinity-relevant geometry.
    §3.1 data construction; repairs and WT-only selection shape the PPI result.
  • ad hoc to paper Persistent (i<j) Hodge features can be omitted; single-scale i=j spectra stacked over isovalues suffice for SOTA BA prediction.
    Remark 2.13 explicitly leaves genuine persistence unused while still branding the method persistent manifold learning.
invented entities (2)
  • Persistent Manifold Learning (PML) pipeline no independent evidence
    purpose: Name the end-to-end BA framework combining multiscale manifold BIG spectra with LM embeddings and GBDT.
    Organizational label for a composition of prior operators plus ML; not a new physical object.
  • Boundary-Induced Graph Laplacian features for binding interfaces independent evidence
    purpose: Practical spectral descriptors (β0 + low eigenvalues of L_{3,n}) on Eulerian interface manifolds.
    Operator attributed to [41]; application to dual BA tasks is the paper’s use. Independent mathematical evidence exists in the cited SIAM Review line, not new here.

pith-pipeline@v1.2.0-grok45-kimik3 · 27081 in / 4152 out tokens · 85899 ms · 2026-07-31T00:47:58.403994+00:00 · methodology

0 comments
read the original abstract

Predicting how tightly two biomolecules bind remains a major challenge, in part because different interaction classes present dissimilar interfaces, from compact metal-coordinated pockets to broad, featureless protein surfaces. We introduce persistent manifold learning (PML), a novel computational framework that describes a binding interface as a family of multiscale manifolds. Boundary-Induced Graph Laplacian, a discrete realization of de Rham-Hodge theory, then extracts topological invariants together with nonharmonic spectral information, capturing the geometry of an interface as well as its topology. These manifold embeddings are combined with protein and molecular language model representations and paired with gradient boosting decision trees. Our PML outperforms state-of-the-art methods on metalloprotein-ligand and protein-protein benchmarks.

Figures

Figures reproduced from arXiv: 2607.25115 by Chunmei Wang, Guo-Wei Wei, Xingjian Xu, Zhe Su.

Figure 1
Figure 1. Figure 1: Overview of the proposed PML framework for metalloprotein-ligand and protein-protein BA prediction. The workflow includes manifold generation from protein structures in the SKEMPI v2 and metalloprotein–ligand datasets, geometry-based algebraic topology feature extraction using multiscale Hodge Laplacian descriptors, and machine learning model training and testing for BA prediction [PITH_FULL_IMAGE:figures… view at source ↗
Figure 2
Figure 2. Figure 2: Homotopic evolution of the Gaussian isosurface manifold generated from protein oxygen atoms and ligand hydrogen atoms in the 4TMN protein–ligand complex. The manifold is constructed from a Gaussian density field over OH atom pairs and evolves as the isovalue decreases 3.3. Grid Construction and Cell Complex. To compute the manifolds described in Sec￾tion 3.2 numerically, we embed each atom pair in a regula… view at source ↗
Figure 3
Figure 3. Figure 3: (a) Performance comparison of PML with previously published models for metalloprotein￾ligand binding affinity prediction, evaluated using the Pearson correlation coefficient. (b) Correlation between experimental and predicted binding affinities from PML’s best-performing run (seed 121314), with PCC 0.756 kcal/mol and RMSE 1.176 kcal/mol. 4.1.2. Cross-Validation Performance. To assess the sensitivity of PML… view at source ↗
Figure 4
Figure 4. Figure 4: Comparison between SVM and GBDT regressors trained on the same PML feature set, for the metalloprotein-ligand binding affinity dataset (seed 181920). (a) Predicted versus experimental binding affinities using an SVM regressor. (b) Predicted versus experimental binding affinities using a GBDT regressor. Both panels show representative results from a single run. used SVM-based baseline for protein-protein BA… view at source ↗
Figure 5
Figure 5. Figure 5: (a) Performance comparison of PML with PLNet and PML-SVM for protein-protein binding affinity prediction, evaluated using the Pearson correlation coefficient. (b) Correlation between experimental and predicted binding affinities from PML’s best-performing run (seed 212223), with PCC = 0.731 and RMSE = 1.996 kcal/mol. 5.1. Element-specific manifolds and the metal coordination shell. The element-specific dec… view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

59 extracted references · 1 linked inside Pith

  1. [1]

    Bennett, Giang T

    Jack L. Bennett, Giang T. H. Nguyen, and William A. Donald. Protein–Small Molecule interactions in native mass spectrometry.Chemical Reviews, 122(8):7327–7385, 2022

    2022

  2. [2]

    Representability of algebraic topology for biomolecules in machine learning based scoring and virtual screening.PLOS Computational Biology, 14(1):e1005929, 2018

    Zixuan Cang, Lin Mu, and Guo-Wei Wei. Representability of algebraic topology for biomolecules in machine learning based scoring and virtual screening.PLOS Computational Biology, 14(1):e1005929, 2018

    2018

  3. [3]

    Topologynet: Topology based deep convolutional and multi-task neural networks for biomolecular property predictions.PLOS Computational Biology, 13(7):e1005690, 2017

    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

  4. [4]

    Zixuan Cang and Guo-Wei Wei. Integration of element specific persistent homology and machine learn- ing for protein–ligand binding affinity prediction.International Journal for Numerical Methods in Biomedical Engineering, 34(2):e2914, 2018

    2018

  5. [5]

    Persistent laplacian projected omicron ba.4 and ba.5 to become new dominating variants.Computers in Biology and Medicine, 151:106262, 2022

    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

  6. [6]

    Evolutionary de Rham–Hodge method

    Jiahui Chen, Rundong Zhao, Yiying Tong, and Guo-Wei Wei. Evolutionary de Rham–Hodge method. Discrete and Continuous Dynamical Systems - Series B, 26(7):3785–3821, 2021

    2021

  7. [7]

    Protein–protein interactions: General trends in the relationship between binding affinity and interfacial buried surface area.Protein Science, 22(4):510– 515, 2013

    Jing Chen, Nicholas Sawyer, and Lynne Regan. Protein–protein interactions: General trends in the relationship between binding affinity and interfacial buried surface area.Protein Science, 22(4):510– 515, 2013

    2013

  8. [8]

    ChemBERTa: Large-scale self- supervised pretraining for molecular property prediction.arXiv preprint arXiv:2010.09885, 2020

    Seyone Chithrananda, Gabriel Grand, and Bharath Ramsundar. ChemBERTa: Large-scale self- supervised pretraining for molecular property prediction.arXiv preprint arXiv:2010.09885, 2020

    Pith/arXiv arXiv 2010

  9. [9]

    DBAC: A simple prediction method for protein binding hot spots based on burial levels and deeply buried atomic contacts.BMC Systems Biology, 5(Suppl 1):S5, 2011

    Lei Deng, Jihong Guan, Xiaoming Wei, Yuan Yi, Qiwen Zhang, and Shuigeng Zhou. DBAC: A simple prediction method for protein binding hot spots based on burial levels and deeply buried atomic contacts.BMC Systems Biology, 5(Suppl 1):S5, 2011

    2011

  10. [10]

    Discrete differential forms for computational modeling

    Mathieu Desbrun, Eva Kanso, and Yiying Tong. Discrete differential forms for computational modeling. InACM SIGGRAPH 2006 Courses, pages 39–54. ACM, 2006

    2006

  11. [11]

    Insights into Protein–Ligand interactions: Mechanisms, models, and methods.International Journal of Molecular Sciences, 17(2):144, 2016

    Xiaoyong Du, Youhua Li, Yao-Liang Xia, Shi-Meng Ai, Jie Liang, Peng Sang, Xing-Lai Ji, and Shihua Liu. Insights into Protein–Ligand interactions: Mechanisms, models, and methods.International Journal of Molecular Sciences, 17(2):144, 2016

    2016

  12. [12]

    First–second shell interactions in metal binding sites in proteins: A PDB survey and DFT/CDM calculations.Journal of the American Chemical Society, 125(10):3168–3180, 2003

    Todor Dudev, Yen-lin Lin, Milena Dudev, and Carmay Lim. First–second shell interactions in metal binding sites in proteins: A PDB survey and DFT/CDM calculations.Journal of the American Chemical Society, 125(10):3168–3180, 2003

    2003

  13. [13]

    Durrant and J

    Jacob D. Durrant and J. Andrew McCammon. NNScore 2.0: A neural-network receptor–ligand scoring function.Journal of Chemical Information and Modeling, 51(11):2897–2903, 2011

    2011

  14. [14]

    Ariel Erijman, Elad Rosenthal, and Julia M. Shifman. How structure defines affinity in Protein–Protein interactions.PLOS ONE, 9(10):e110085, 2014

    2014

  15. [15]

    Feinberg, Debnil Sur, Zhenqin Wu, Brooke E

    Evan N. Feinberg, Debnil Sur, Zhenqin Wu, Brooke E. Husic, Huanghao Mai, Yang Li, Saisai Sun, Jianyi Yang, Bharath Ramsundar, and Vijay S. Pande. Potentialnet for molecular property prediction.ACS Central Science, 4(11):1520–1530, 2018

    2018

  16. [16]

    CAML: Commutative algebra machine learning—a case study on protein–ligand binding affinity prediction.Journal of Chemical Information and Modeling, 2025

    Hongsong Feng, Faisal Suwayyid, Mushal Zia, JunJie Wee, Yuta Hozumi, Chun-Long Chen, and Guo- Wei Wei. CAML: Commutative algebra machine learning—a case study on protein–ligand binding affinity prediction.Journal of Chemical Information and Modeling, 2025

    2025

  17. [17]

    Algebraic connectivity of graphs.Czechoslovak Mathematical Journal, 23(2):298–305, 1973

    Miroslav Fiedler. Algebraic connectivity of graphs.Czechoslovak Mathematical Journal, 23(2):298–305, 1973

    1973

  18. [18]

    Fleishman and David Baker

    Sarel J. Fleishman and David Baker. Hotspot-centric de novo design of protein binders.Journal of Molecular Biology, 413(5):1047–1062, 2011

    2011

  19. [19]

    K. O. Friedrichs. Differential forms on Riemannian manifolds.Communications on Pure and Applied 24 X. XU, Z. SU, G. WEI, C. WANG Mathematics, 8(4):551–590, 1955

    1955

  20. [20]

    Gilson, James A

    Michael K. Gilson, James A. Given, Bruce L. Bush, and J. Andrew McCammon. The statistical- thermodynamic basis for computation of binding affinities: A critical review.Biophysical Journal, 72(3):1047–1069, 1997

    1997

  21. [21]

    Hussein Hassan-Harrirou, Ce Zhang, and Thomas Lemmin. RosENet: Improving binding affinity pre- diction by leveraging molecular mechanics energies with an ensemble of 3D convolutional neural networks.Journal of Chemical Information and Modeling, 60(6):2791–2802, 2020

    2020

  22. [22]

    William V. D. Hodge.The Theory and Applications of Harmonic Integrals. Cambridge University Press, 1941

    1941

  23. [23]

    Justina Jankauskait˙ e, Brian Jim´ enez-Garc ´ ıa, Justas Dapkunas, Juan Fern´ andez-Recio, and Iain H. Moal. SKEMPI 2.0: An updated benchmark of changes in protein–protein binding energy, kinetics and thermodynamics upon mutation.Bioinformatics, 35(3):462–469, 2019

    2019

  24. [24]

    Interactiongraphnet: A novel and efficient deep graph representation learning framework for accurate protein–ligand interaction predictions

    Dejun Jiang, Chang-Yu Hsieh, Zhenxing Wu, Yu Kang, Jike Wang, Ercheng Wang, Ben Liao, Chao Shen, Lei Xu, Jian Wu, Dejun Cao, and Tingjun Hou. Interactiongraphnet: A novel and efficient deep graph representation learning framework for accurate protein–ligand interaction predictions. Journal of Medicinal Chemistry, 2021

    2021

  25. [25]

    Metalprognet: A structure-based deep graph model for metalloprotein– ligand interaction predictions.Chemical Science, 14(8):2054–2069, 2023

    Dejun Jiang, Zixuan Ye, Chang-Yu Hsieh, Ziyi Yang, Xiaohui Zhang, Yu Kang, Haitao Du, Zhenxing Wu, Jike Wang, and Yi Zeng. Metalprognet: A structure-based deep graph model for metalloprotein– ligand interaction predictions.Chemical Science, 14(8):2054–2069, 2023

    2054

  26. [26]

    Kitchen, H´ el` ene Decornez, John R

    Douglas B. Kitchen, H´ el` ene Decornez, John R. Furr, and J¨ urgen Bajorath. Docking and scoring in virtual screening for drug discovery: Methods and applications.Nature Reviews Drug Discovery, 3(11):935–949, 2004

    2004

  27. [27]

    A simple physical model for binding energy hot spots in protein– protein complexes.Proceedings of the National Academy of Sciences, 99(22):14116–14121, 2002

    Tanja Kortemme and David Baker. A simple physical model for binding energy hot spots in protein– protein complexes.Proceedings of the National Academy of Sciences, 99(22):14116–14121, 2002

    2002

  28. [28]

    Monn: A multi- objective neural network for predicting compound–protein interactions and affinities.Cell Systems, 10(4):308–322.e11, 2020

    Shuya Li, Fangping Wan, Hantao Shu, Tao Jiang, Dan Zhao, and Jianyang Zeng. Monn: A multi- objective neural network for predicting compound–protein interactions and affinities.Cell Systems, 10(4):308–322.e11, 2020

    2020

  29. [29]

    Evolutionary-scale prediction of atomic- level protein structure with a language model.Science, 379(6637):1123–1130, 2023

    Zeming Lin, Halil Akin, Roshan Rao, Brian Hie, Zhongkai Zhu, Wenting Lu, Allan dos Santos Costa, Maryam Fazel-Zarandi, Tom Sercu, Salvatore Candido, et al. Evolutionary-scale prediction of atomic- level protein structure with a language model.Science, 379(6637):1123–1130, 2023

    2023

  30. [30]

    The algebraic stability for persistent laplacians.Homology, Homotopy and Applications, 26(2):297–323, 2024

    Jian Liu, Jingyan Li, and Jie Wu. The algebraic stability for persistent laplacians.Homology, Homotopy and Applications, 26(2):297–323, 2024

    2024

  31. [31]

    Manifold topological deep learning for biomedical data.Nature Communications, 17:4710, 2026

    Xiang Liu, Zhe Su, Yongyi Shi, Yiying Tong, Ge Wang, and Guo-Wei Wei. Manifold topological deep learning for biomedical data.Nature Communications, 17:4710, 2026

    2026

  32. [32]

    Long et al

    Y. Long et al. Interpretable binding affinity prediction with persistent homology.PLOS Computational Biology, 21(6):e1013216, 2025

    2025

  33. [33]

    Recent advances in the development of Protein–Protein Interaction modulators: Mechanisms and clinical trials.Signal Transduction and Targeted Therapy, 5:213, 2020

    Haiying Lu, Qiaodan Zhou, Jun He, Zhongliang Jiang, Cheng Peng, Rongsheng Tong, and Jianyou Shi. Recent advances in the development of Protein–Protein Interaction modulators: Mechanisms and clinical trials.Signal Transduction and Targeted Therapy, 5:213, 2020

    2020

  34. [34]

    Persistent laplacians: Properties, algorithms and implications.SIAM Journal on Mathematics of Data Science, 4(2):858–884, 2022

    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

  35. [35]

    Persistent spectral-based machine learning (PerSpect ML) for protein– ligand binding affinity prediction.Science Advances, 7(19):eabc5329, 2021

    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

  36. [36]

    Moesser, Dominik K

    Marc A. Moesser, Dominik K. Klein, Fergus Boyles, Charlotte M. Deane, Alice Baxter, and Garrett M. Morris. Protein–ligand interaction graphs: Learning from ligand-shaped 3d interaction graphs to improve binding affinity prediction.bioRxiv, 2022

    2022

  37. [37]

    Meanwell, and Kyeong Lee

    Hossam Nada, Yongseok Choi, Sungdo Kim, Kwon Su Jeong, Nicholas A. Meanwell, and Kyeong Lee. New insights into Protein–Protein Interaction modulators in drug discovery and therapeutic advance. Signal Transduction and Targeted Therapy, 9:341, 2024

    2024

  38. [38]

    DG-GL: Differential geometry-based geometric learning of molecular datasets.International Journal for Numerical Methods in Biomedical Engineering, 35(3):e3179, 2019

    Duc Duy Nguyen and Guo-Wei Wei. DG-GL: Differential geometry-based geometric learning of molecular datasets.International Journal for Numerical Methods in Biomedical Engineering, 35(3):e3179, 2019

    2019

  39. [39]

    Quinn, Tri Nguyen, Truyen Le, and Svetha Venkatesh

    Thin Nguyen, Hang Le, Thomas P. Quinn, Tri Nguyen, Truyen Le, and Svetha Venkatesh. Graphdta: Predicting drug–target binding affinity with graph neural networks.Bioinformatics, 37(8):1140–1147, 2021

    2021

  40. [40]

    Permyakov

    Eugene A. Permyakov. Metal binding proteins.Encyclopedia, 1(1):261–292, 2021

    2021

  41. [41]

    Combinatorial and Hodge PERSISTENT MANIFOLD LEARNING 25 Laplacians: Similarities and differences.SIAM Review, 2024

    Emily Ribando-Gros, Rui Wang, Jiahui Chen, Yiying Tong, and Guo-Wei Wei. Combinatorial and Hodge PERSISTENT MANIFOLD LEARNING 25 Laplacians: Similarities and differences.SIAM Review, 2024

    2024

  42. [42]

    Lawrence Zitnick, Jerry Ma, and Rob Fergus

    Alexander Rives, Joshua Meier, Tom Sercu, Siddharth Goyal, Zeming Lin, Jason Liu, Demi Guo, Myle Ott, C. Lawrence Zitnick, Jerry Ma, and Rob Fergus. Biological structure and function emerge from scaling unsupervised learning to 250 million protein sequences.Proceedings of the National Academy of Sciences, 118(15):e2016239118, 2021

    2021

  43. [43]

    Sorzano, Jos´ e Mar ´ ıa Carazo, and Joan Segura

    Ruben S´ anchez-Garc ´ ıa, Carlos Oscar S. Sorzano, Jos´ e Mar ´ ıa Carazo, and Joan Segura. PPI-Affinity: A web tool for the prediction and optimization of protein–peptide and protein–protein binding affinity. Journal of Proteome Research, 21(8):1829–1841, 2022

    2022

  44. [44]

    Springer, Berlin, 1995

    G¨ unter Schwarz.Hodge Decomposition: A Method for Solving Boundary Value Problems, volume 1607 ofLecture Notes in Mathematics. Springer, Berlin, 1995

    1995

  45. [45]

    Scott, Andrew R

    Duncan E. Scott, Andrew R. Bayly, Chris Abell, and John Skidmore. Small molecules, big targets: Drug discovery faces the Protein–Protein interaction challenge.Nature Reviews Drug Discovery, 15(8):533–550, 2016

    2016

  46. [46]

    Topological data analysis and topological deep learning beyond persistent homology: a review

    Zhe Su, Xiang Liu, Layal Bou Hamdan, Vasileios Maroulas, Jie Wu, Gunnar Carlsson, and Guo-Wei Wei. Topological data analysis and topological deep learning beyond persistent homology: a review. Artificial Intelligence Review, 59:58, 2025

    2025

  47. [47]

    Persistent de Rham–Hodge laplacians in Eulerian representation for manifold topological learning.AIMS Mathematics, 9(10):27438–27470, 2024

    Zhe Su, Yiying Tong, and Guo-Wei Wei. Persistent de Rham–Hodge laplacians in Eulerian representation for manifold topological learning.AIMS Mathematics, 9(10):27438–27470, 2024

    2024

  48. [48]

    Topology-preserving hodge decomposition in the eulerian representation.Beijing Journal of Pure and Applied Mathematics, 2(2):619–657, 2025

    Zhe Su, Yiying Tong, and Guo-Wei Wei. Topology-preserving hodge decomposition in the eulerian representation.Beijing Journal of Pure and Applied Mathematics, 2(2):619–657, 2025

    2025

  49. [49]

    Representation of molecular structures with persistent homology for machine learning applications in chemistry.Nature communications, 11(1):3230, 2020

    Jacob Townsend, Cassie Putman Micucci, John H Hymel, Vasileios Maroulas, and Konstantinos D Vogiatzis. Representation of molecular structures with persistent homology for machine learning applications in chemistry.Nature communications, 11(1):3230, 2020

    2020

  50. [50]

    Anna Vangone and Alexandre M. J. J. Bonvin. Contacts-based prediction of binding affinity in Protein– Protein complexes.eLife, 4:e07454, 2015

    2015

  51. [51]

    Persistent path laplacian.Foundations of Data Science, 5(1):26–55, 2023

    Rui Wang et al. Persistent path laplacian.Foundations of Data Science, 5(1):26–55, 2023

    2023

  52. [52]

    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

  53. [53]

    Join persistent homology (jph)- based machine learning for metalloprotein–ligand binding affinity prediction.Journal of Chemical Information and Modeling, 65(6):2785–2793, 2025

    Yaxing Wang, Xiang Liu, Yipeng Zhang, Xiangjun Wang, and Kelin Xia. Join persistent homology (jph)- based machine learning for metalloprotein–ligand binding affinity prediction.Journal of Chemical Information and Modeling, 65(6):2785–2793, 2025

    2025

  54. [54]

    Join persistent homology (JPH)-based machine learning for metalloprotein–ligand binding affinity prediction.Journal of Chemical Information and Modeling, 2025

    Yaxing Wang, Xiang Liu, Yipeng Zhang, Xiangjun Wang, and Kelin Xia. Join persistent homology (JPH)-based machine learning for metalloprotein–ligand binding affinity prediction.Journal of Chemical Information and Modeling, 2025

    2025

  55. [55]

    Persistent sheaf laplacians.Foundations of data science (Springfield, Mo.), 7(2):446, 2025

    Xiaoqi Wei and Guo-Wei Wei. Persistent sheaf laplacians.Foundations of data science (Springfield, Mo.), 7(2):446, 2025

    2025

  56. [56]

    Persistent homology analysis of protein structure, flexibility, and folding

    Kelin Xia and Guo-Wei Wei. Persistent homology analysis of protein structure, flexibility, and folding. International Journal for Numerical Methods in Biomedical Engineering, 30(8):814–844, 2014

    2014

  57. [57]

    Extending the accuracy limits of prediction for side-chain conformations

    Zhexin Xiang and Barry Honig. Extending the accuracy limits of prediction for side-chain conformations. Journal of Molecular Biology, 311(2):421–430, 2001

    2001

  58. [58]

    PLNet: Persistent laplacian neural network for protein–protein binding free energy prediction.Protein Science, 34(12):e70377, 2025

    Xingjian Xu, Chunmei Wang, Guo-Wei Wei, and Jiahui Chen. PLNet: Persistent laplacian neural network for protein–protein binding free energy prediction.Protein Science, 34(12):e70377, 2025

    2025

  59. [59]

    Ciallella, Lauren M

    Lingling Zhao, Hao L. Ciallella, Lauren M. Aleksunes, and Hao Zhu. A brief review of Protein–Ligand interaction prediction.Computational and Structural Biotechnology Journal, 20:2831–2838, 2022

    2022