The DOI matches its Crossref record; 54 of the 54 references it cites were found in a registry or an index; 1 of the 27 sources its claims cite were checked for support.
A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation
Open accessAbstract
The ability to predict protein–protein interactions is crucial to our understanding of a wide range of biological activities and functions in the human body, and for guiding drug discovery. Despite considerable efforts to develop suitable computational methods, predicting protein–protein interaction binding affinity changes following mutation (ΔΔG) remains a severe challenge. Algebraic topology, a champion in recent worldwide competitions for protein–ligand binding affinity predictions, is a promising approach to simplifying the complexity of biological structures. Here we introduce element- and site-specific persistent homology (a new branch of algebraic topology) to simplify the structural complexity of protein–protein complexes and embed crucial biological information into topological invariants. We also propose a new deep learning algorithm called NetTree to take advantage of convolutional neural networks and gradient-boosting trees. A topology-based network tree is constructed by integrating the topological representation and NetTree for predicting protein–protein interaction ΔΔG. Tests on major benchmark datasets indicate that the proposed topology-based network tree is an important improvement over the current state of the art in predicting ΔΔG. Persistent homology provides an efficient approach to simplifying the complexity of protein structure. Wang et al. combine this approach with convolutional neural networks and gradient-boosting trees to improve predictions of protein–protein interactions.
Overview
AI snapshot
Written from the paper’s full text.
TL;DR
The paper introduces TopNetTree, combining element- and site-specific persistent homology with CNN-assisted gradient-boosting trees to predict protein–protein binding affinity changes upon mutation (ΔΔG). It reports improved Pearson correlations over prior methods on AB-Bind and SKEMPI benchmarks, including RP 0.65 on AB-Bind S645 and RP 0.82 on SKEMPI 2.0 S4947.
Key findings
- On the AB-Bind S645 set, TopNetTree achieved an RP of 0.65, about 22% better than the previous best RP of 0.53 from mCSM-AB; excluding 27 non-binders increased RP to 0.68, and a blind test on homology structures gave RP 0.55.
- On the SKEMPI S1131 set, the method was reported as about 15% more accurate than the best existing method in tenfold cross-validation.
- On SKEMPI 2.0, TopNetTree achieved RP 0.82 and r.m.s.e. 1.11 kcal mol−1 on S4947; RP 0.79 and r.m.s.e. 1.13 kcal mol−1 on S4169 versus mCSM-PPI2’s RP 0.76 and r.m.s.e. 1.19 kcal mol−1; and RP 0.85 and r.m.s.e. 1.11 kcal mol−1 on S8338 versus mCSM-PPI2’s RP 0.82 and r.m.s.e. 1.18 kcal mol−1.
- Cross-validation analyses showed inter-protein average/median RP of 0.508/0.541 and intra-protein leave-one-out average/median RP of 0.170/0.215; mutation-region RP values were 0.60 (core), 0.66 (rim), 0.66 (support), 0.65 (interior) and 0.48 (surface).
Methods
- The model uses element- and site-specific persistent homology: atoms are classified into mutation-site, neighbourhood, antibody, antigen and element-type subsets, with modified distance matrices excluding same-set interactions; H0 barcodes come from Vietoris–Rips filtration, while H1 and H2 barcodes come from alpha complex filtration.
- H0 barcodes are vectorized by binning the filtration range (for example, [0, 12] Å into 0.5 Å bins) and counting persistence intervals, birth events and death events; H1 and H2 barcodes are summarized by statistics such as sum, maximum, minimum, mean and standard deviation of bar lengths, birth values and death values.
- A CNN extracts higher-level features from H0; these are combined with H1/H2 statistics and fed into a gradient-boosting tree (GBT) for final prediction. The paper evaluates on AB-Bind S645, SKEMPI S2317/S1131, and SKEMPI 2.0 S4947/S4169/S8338/S787 using RP and r.m.s.e., with tenfold, inter-protein, intra-protein and blind-test protocols.
Limitations
- Stated: the characterization of birth and death may not be stable against different discretizations, so only H0 barcodes from Vietoris–Rips filtration are used; datasets are relatively small and complex; non-binders are outliers that strongly affect accuracy; intra-protein cross-validation is limited by small per-complex training sets (average 27 samples); surface mutations have small data and small average energy disturbance; model prediction variance is generally lower than experimental variance.
- Evident from reported results: performance varies strongly across protein families, with negative RP for 1FFW (−0.043) and 1YY9 (−0.068); inter-protein cross-validation (average RP 0.508) and blind tests (homology RP 0.55; S787 RP 0.53) are lower than the main tenfold results.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
17 Own result · 15 Not checked yet · 12 Own interpretation · 9 No basis given · 3 Own method · 1 Partly supported · accuracy not yet measured
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“We also applied a blind test on homology structures using the rest of the samples as the training set, achieving an RP of 0.55, as shown in Fig. 4.”
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > TopNetTree model for PPI binding energy change following mutation prediction > Model performance for PPIs.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Tests on major benchmark datasets indicate that the proposed topology-based network tree is an important improvement over the current state of the art in predicting ΔΔG.”
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > Abstract
hedged- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- The authors state this as an assumption: The individual learners in an ensemble are likely to make different mistakes, so summing weak learners reduces overall error.
- Rests on an assumption: PPI ΔΔG prediction data are highly complex due to 3D structures, whereas the datasets are relatively small.
In the paper
“Our method is about 22% more accurate than the best-existing score of RP = 0.53 (given by mCSM-AB), indicating the power of our TopNetTree.”
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > TopNetTree model for PPI binding energy change following mutation prediction > Model performance for PPIs.
What the sources say
- Not checked yetcites Pires et al.21 Pires, D. E. & Ascher, D. B. mCSM-AB: a web server for predicting antibody–antigen affinity changes upon mutation with graph-based signatures. Nucleic Acids Res. 44, W469–W473 (2016). source
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In comparison with non-machine learning methods such as Rosetta and bASA, our method is over 100% more accurate in terms of RP indicating that our topology-based machine learning methods have a better predictive power for PPI systems.”
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > TopNetTree model for PPI binding energy change following mutation prediction > Model performance for PPIs.
hedgedcorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In our case, the RP increased from 0.65 to 0.68 for the same treatment as shown in Fig. 4.”
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > TopNetTree model for PPI binding energy change following mutation prediction > Model performance for PPIs.
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Our model achieved an RP of 0.65 on the AB-Bind S645 dataset, which is significantly better than those of other existing methods as shown in Table 1.”
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > TopNetTree model for PPI binding energy change following mutation prediction > Model performance for PPIs.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“It is found that the proposed TopNetTree is about 15% more accurate than the best-existing method.”
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > TopNetTree model for PPI binding energy change following mutation prediction > Model performance for PPIs.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“however, the performance of the model varies among different protein families”
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > Discussion > Prediction result analysis for different protein complexes.
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“we find that most mutations from polar to hydrophobic residues have a positive free energy change (for example, S to M)”
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > Discussion > Prediction result analysis for different mutation types.
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“The ΔΔG of 27 non-binders (−8 kcal mol−1) did not follow the distribution of the whole dataset.”
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > TopNetTree model for PPI binding energy change following mutation prediction > Model performance for PPIs.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Probably not a substantial claim. 17
- It may state two separate claims as one. 15
- The quote only partly states this claim. 3
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“There are 2,317 single point mutation data entries among the whole database, which are referred to as the SKEMPI S2317 set.”
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > TopNetTree model for PPI binding energy change following mutation prediction > Model performance for PPIs.
What the sources say
- Partly supportedcites [11] Moal, I. H. & Fernández-Recio, J. SKEMPI: a structural kinetic and energetic database of mutant protein interactions and its use in empirical models. Bioinformatics 28, 2600–2607 (2012). sourceExperimental — accuracy not yet measured
The Results section reports that the database contains 2,317 single mutants, supporting the count and mutation type. Our copy does not call these entries the “SKEMPI S2317 set,” so the attributed name is not supported. (LLM, conf=0.88)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim.
- Partly supported
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“We achieved an average RP of 0.53 and r.m.s.e. of 1.45 kcal mol−1 on this blind test (further details are given in the Supplementary Information).”
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > TopNetTree model for PPI binding energy change following mutation prediction > Model performance for PPIs.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one. The quote only partly states this claim.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“We tested set S4947 with the regular tenfold cross-validation, achieving an average RP of 0.82 and an r.m.s.e. of 1.11 kcal mol−1 for the tenfold cross-validation.”
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > TopNetTree model for PPI binding energy change following mutation prediction > Model performance for PPIs.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
In the paper
“For set S4169, we obtained an average RP of 0.79 and r.m.s.e. of 1.13 kcal mol−1, compared with the average RP of 0.76 and r.m.s.e. of 1.19 kcal mol−1 achieved by mCSM-PPI2.”
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > TopNetTree model for PPI binding energy change following mutation prediction > Model performance for PPIs.
What the sources say
- Not checked yetcites David et al.48
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“This dataset, called the AB-Bind S645 set, consists of about 20% stabilizing mutations and 80% destabilizing ones; there are 27 non-binders in the whole dataset, which are variants determined not to bind within the sensitivity of the assay.”
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > TopNetTree model for PPI binding energy change following mutation prediction > Model performance for PPIs.
hedged- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“It is interesting to note that all the mutations to alanine have a positive energy change”
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > Discussion > Prediction result analysis for different mutation types.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“In experimental data, mutations at the core or support region have a higher average energy change of around 1.8 kcal mol−1 (1.72 kcal mol−1 and 1.91 kcal mol−1, respectively), whereas mutations at the rim or interior region have an average energy change of around 0.8 kcal mol−1 (0.82 kcal mol−1 and 0.83 kcal mol−1, respectively)”
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > Discussion > Prediction result analysis for different mutation regions.
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Overall, our predicted patterns are remarkably similar to those of experimental data in terms of both average binding energy changes and variance of binding energy changes, as shown in Fig. 5.”
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > Discussion > Prediction result analysis for different mutation types.
boostedcorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“For this test, our model reached average/median RP values of 0.170/0.215, which are significantly lower than the tenfold cross-validation result over the entire dataset.”
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > Discussion > Prediction result analysis for different protein complexes.
plainn = 27- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“For this test, our model reached an average RP of 0.508 and a median RP of 0.541.”
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > Discussion > Prediction result analysis for different protein complexes.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“We demonstrate that the proposed TopNetTree achieves an RP of 0.65, which is about 22% better than the previous best result for the AB-Bind dataset.”
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > Conclusion
boostedcorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
In the paper
“Sirin et al.10 collected an AB-Bind database of mutation-induced antibody–antigen complex binding free energy changes. This database contains 1,101 mutation data entries, including 645 single-point mutations on 32 different antibody–antigen complexes.”
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > Abstract
What the sources say
- Not checked yetcites Sirin et al.10 Sirin, S., Apgar, J. R., Bennett, E. M. & Keating, A. E. AB-bind: antibody binding mutational database for computational affinity predictions. Protein Sci. 25, 393–409 (2016). source
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.
In the paper
“This dataset contains 7,085 entries, including single- and multi-point mutations.”
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > TopNetTree model for PPI binding energy change following mutation prediction > Model performance for PPIs.
What the sources say
- Not checked yetcites [39] Jankauskaitė, J., Jiménez-García, B., Dapkūnas, J., Fernández-Recio, J. & Moal, I. H. SKEMPI 2.0: an updated benchmark of changes in protein–protein binding energy, kinetics and thermodynamics upon mutation. Bioinformatics 35, 462–469 (2018). source
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim.
In the paper
“Pires et al.21 found that excluding non-binders from the dataset would significantly increase the performance of a prediction model.”
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > TopNetTree model for PPI binding energy change following mutation prediction > Model performance for PPIs.
What the sources say
- Not checked yetcites Pires et al.21 Pires, D. E. & Ascher, D. B. mCSM-AB: a web server for predicting antibody–antigen affinity changes upon mutation with graph-based signatures. Nucleic Acids Res. 44, W469–W473 (2016). source
boosteddirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
In the paper
“SKEMPI is a more general database for protein–protein binding affinity changes following mutation (ΔΔG)11, it contains 3,047 mutation data entries for protein–protein heterodimeric complexes with experimentally determined structures.”
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > Abstract
What the sources say
- Not checked yetcites [11] Moal, I. H. & Fernández-Recio, J. SKEMPI: a structural kinetic and energetic database of mutant protein interactions and its use in empirical models. Bioinformatics 28, 2600–2607 (2012). source
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The limited performance of the current methods therefore highlights a pressing need for a new generation of ΔΔG predictors that are constructed with entirely new design principles and/or innovative machine learning algorithms.”
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > Abstract
plain- Asserted: passes — The authors put it forward as true.
- Checkable: fails — It may be too vague to check against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may be too vague to check against evidence.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“a possible reason is that mutations from a large residue to a small one could lead to a stabilizing effect to the whole system”
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > Discussion > Prediction result analysis for different mutation types.
hedgedconditional- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
None of the sources cited for this claim have been checked yet.
In the paper
“Persistent homology22–25—a new branch of algebraic topology—is able to bridge geometry and topology, leading to a new efficient approach for the simplification of biological structural complexity26–31; however, it neglects critical chemical/biological information when it is directly applied to complex biomolecular structures.”
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > Abstract
What the sources say
- Not checked yetcites [22] Frosini, P. & Landi, C. Size theory as a topological tool for computer vision. Pattern Recognition Image Anal. 9, 596–603 (1999).
- Not checked yetcites [23] Edelsbrunner, H., Letscher, D. & Zomorodian, A. Topological persistence and simplification. Discrete Comput. Geom. 28, 511–533 (2002). source
- Not checked yetcites [24] Zomorodian, A. & Carlsson, G. Computing persistent homology. Discrete Comput. Geom. 33, 249–274 (2005). source
- Not checked yetcites [25] Zomorodian, A. & Carlsson, G. Localized homology. Comput. Geom. 41, 126–148 (2008). source
- Not checked yetcites [26] Xia, K. L. & Wei, G. W. Persistent homology analysis of protein structure, flexibility and folding. Int. J. Numer. Methods Biomed. Eng. 30, 814–844 (2014). source
- Not checked yetcites [27] Gameiro, M. et al. Topological measurement of protein compressibility via persistence diagrams. Japan J. Industr. Appl. Math. 32, 1–17 (2014).
- Not checked yetcites [28] Xia, K. L. & Wei, G. W. Persistent topology for cryo-EM data analysis. Int. J. Numer. Methods Biomed. Eng. 31, e02719 (2015).
- Not checked yetcites [29] Cang, Z. X. et al. A topological approach to protein classification. Mol. Based Math. Biol. 3, 140–162 (2015).
- Not checked yetcites [30] Yao, Y. et al. Topological methods for exploring low-density states in biomolecular folding pathways. J. Chem. Phys. 130, 04B614 (2009).
- Not checked yetcites [31] Kovacev-Nikolic, V., Bubenik, P., Nikolić, D. & Heo, G. Using persistent homology and dynamical distances to analyze protein binding. Stat. Appl. Genet. Mol. Biol. 15, 19–38 (2016). source
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“These non-binders could be regarded as outliers in the database and have a strongly negative impact on the prediction model accuracy.”
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > TopNetTree model for PPI binding energy change following mutation prediction > Model performance for PPIs.
hedgeddirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“A possible reason for these patterns is that different mutation regions vary in their accessibility to water”
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > Discussion > Prediction result analysis for different mutation regions.
hedgedconditional- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
- Rests on an assumption: In general, surface, interior and rim regions have greater access to water than the core and support regions.
In the paper
“Similar patterns regarding mutation sites and energy changes were reported in the literature49.”
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > Discussion > Prediction result analysis for different mutation regions.
What the sources say
- Not checked yetcites [49] Petukh, M., Li, M. & Alexov, E. Predicting binding free energy change caused by point mutations with knowledge-modified MM/PBSA method. PLoS Comput. Biol. 11, e1004276 (2015). source
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
In the paper
“We followed the method of tenfold stratified cross-validation used in mCSM-PPI2 paper for sets S4169 and S833848.”
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > TopNetTree model for PPI binding energy change following mutation prediction > Model performance for PPIs.
What the sources say
- Not checked yetcites [48] Rodrigues, C. H. M., Myung, Y., Pires, D. E. V. & Ascher, D. B. mCSM-PPI2: Predicting the effects of mutations on protein–protein interactions. Nucleic Acids Res. 47, W338– W344 (2019). source
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The quote only partly states this claim.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“Table 2 shows the Pearson correlation coefficients on tenfold cross-validations.”
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > TopNetTree model for PPI binding energy change following mutation prediction > Model performance for PPIs.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The quote only partly states this claim.
In the paper
“They filtered only single-point mutations and selected 4,169 variants in 319 different complexes (denoted set S4169).”
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > TopNetTree model for PPI binding energy change following mutation prediction > Model performance for PPIs.
What the sources say
- Not checked yetcites [48] Rodrigues, C. H. M., Myung, Y., Pires, D. E. V. & Ascher, D. B. mCSM-PPI2: Predicting the effects of mutations on protein–protein interactions. Nucleic Acids Res. 47, W338– W344 (2019). source
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Typically, physics-based methods provide mechanistic interpretations but are not designed for handling large and diverse datasets.”
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > Abstract
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds. It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“We believe that the relative inferior performance for surface mutations is due to its small data size and that the energy disturbance caused by surface mutations is small on average.”
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > Discussion > Prediction result analysis for different mutation regions.
hedgeddirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Models trained on some protein families could extrapolate to other families; for example, the two protein families with the best results, 1KTZ and 2JEL, can reach RP of 0.866 and 0.818, respectively, whereas the two families with the poorest results, 1FFW and 1YY9, have RP values of −0.043 and −0.068, respectively.”
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > Discussion > Prediction result analysis for different protein complexes.
hedgedcorrelationaln = 2- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
None of the sources cited for this claim have been checked yet.
In the paper
“Topology has recently been shown to be surprisingly effective in simplifying biomolecular structural complexity26,27,29.”
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > Conclusion
What the sources say
- Not checked yetcites [26] Xia, K. L. & Wei, G. W. Persistent homology analysis of protein structure, flexibility and folding. Int. J. Numer. Methods Biomed. Eng. 30, 814–844 (2014). source
- Not checked yetcites [27] Gameiro, M. et al. Topological measurement of protein compressibility via persistence diagrams. Japan J. Industr. Appl. Math. 32, 1–17 (2014).
- Not checked yetcites [29] Cang, Z. X. et al. A topological approach to protein classification. Mol. Based Math. Biol. 3, 140–162 (2015).
boosteddirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim.
In the paper
“It has been devised to win worldwide competitions in computer-aided drug design36.”
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > Conclusion
What the sources say
- Not checked yetcites [36] Nguyen, D. D. et al. Mathematical deep learning for pose and binding affinity prediction and ranking in D3R grand challenges. J. Compurt. Aided Mol. Design https://doi.org/10.1007/s10822-018-0146-6 (2018). source
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
The authors state this as an assumption.
The individual learners in an ensemble are likely to make different mistakes, so summing weak learners reduces overall error.
“It improves the performance on the weak learners with the assumption that the individual learners are likely to make different mistakes and thus summing the weak learners will reduce the overall error.”
The paper relies on this without support.
PPI ΔΔG prediction data are highly complex due to 3D structures, whereas the datasets are relatively small.
“A major challenge in the prediction of binding affinity changes following mutation for PPIs is that the data is highly complex due to 3D structures, whereas the datasets are relatively small.”
The paper relies on this without support.
In general, surface, interior and rim regions have greater access to water than the core and support regions.
“in general, surface, interior and rim regions have greater access to water than the core and support regions.”
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
In the citing paper
“TopGBT uses topology-based features to represent the complex, which is not initially designed for representing the interactions between atoms, limiting its predictive power for binding affinity changes upon mutations.”
From Deep geometric representations for modeling effects of mutations on protein-protein binding affinity (2021) · cites this paper as [31]
In the citing paper
“As some complexes in the above benchmark datasets are highly related (S2 and S3 Figs), machine learning methods tend to be overtrained on these datasets, as evidenced by the drops of the prediction performance when machine learning methods meet unseen structures [31].”
From Deep geometric representations for modeling effects of mutations on protein-protein binding affinity (2021) · cites this paper as [31]
In the citing paper
“These datasets have been widely used to test the predictive power of the PPI prediction methods [12, 13, 27–29, 29–35].”
From Deep geometric representations for modeling effects of mutations on protein-protein binding affinity (2021) · cites this paper as [31]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > Abstract
“This database contains 1,101 mutation data entries, including 645 single-point mutations on 32 different antibody–antigen complexes.”
§ A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation > TopNetTree model for PPI binding energy change following mutation prediction > Model performance for PPIs.
“We follow Pires et al.21 by considering only 645 single mutations across 29 antibody–antigen complexes.”
Citations
Cites
Scikit-learn: Machine Learning in Python
Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., Blondel, M., Müller, A., Nothman, J., Louppe, G., et al. (2012) · arXiv · doi:10.48550/arxiv.1201.0490 · Cited by 71,016 · cited 1×
“The implementation provided by the scikit-learn package (v.0.18.1) 38 is used.”
MethodInfluentialBerman, H. M. (2000) · Nucleic Acids Research · doi:10.1093/nar/28.1.235 · Cited by 40,196
The CCP4 suite: programs for protein crystallography
Number Collaborative Computational Project (1994) · Acta Crystallographica Section D Biological Crystallography · doi:10.1107/s0907444994003112 · Cited by 17,317
Comparative Protein Structure Modeling Using MODELLER
Webb, B., & Šali, A. (2016) · Current Protocols in Bioinformatics · doi:10.1002/cpbi.3 · Cited by 4,404 · cited 1×
“Computational methods for generating protein structures from sequences (for example, MODELLER 16 ) and predicting docking poses for protein–protein complexes (for example, BioLuminate 17 ) are also available.”
MethodComparative Protein Structure Modeling Using MODELLER
Eswar, N., Webb, B., Martı́-Renom, M. A., Madhusudhan, M. S., Eramian, D., Shen, M., Pieper, U., & Šali, A. (2007) · Current Protocols in Protein Science · doi:10.1002/0471140864.ps0209s50 · Cited by 2,874
The FoldX web server: an online force field
Schymkowitz, J., Borg, J., Stricher, F., Nys, R., Rousseau, F., & Serrano, L. (2005) · Nucleic Acids Research · doi:10.1093/nar/gki387 · Cited by 2,759
Topological Persistence and Simplification
Edelsbrunner, Letscher, & Zomorodian (2002) · Discrete & Computational Geometry · doi:10.1007/s00454-002-2885-2 · Cited by 1,779
Zomorodian, A., & Carlsson, G. (2005) · Discrete & Computational Geometry · doi:10.1007/s00454-004-1146-y · Cited by 1,591
Conformations of immunoglobulin hypervariable regions
Chothia, C., Lesk, A. M., Tramontano, A. L., Levitt, M., Smith‐Gill, S. J., Air, G. M., Sheriff, S., PADLAN, E. A., Davies, D. R., Tulip, W. R., et al. (1989) · Nature · doi:10.1038/342877a0 · Cited by 1,191
Comparative Protein Structure Modeling Using MODELLER
Webb, B., & Šali, A. (2014) · Current Protocols in Bioinformatics · doi:10.1002/0471250953.bi0506s47 · Cited by 1,156
Cited by
Sorted by citations
Items 201–202 of 202
Mao, S., Meng, S., & Xiang, Y. (2026) · iScience · doi:10.1016/j.isci.2026.117693
“47 With the datasets S1131 and S4169, the two most commonly used benchmarks for PPI affinity prediction task, 19,24,26,29,31,48 we made parallel comparisons of our model with other models that presented a SOTA performance once, including UniBind, 32 MpbPPI, 29 Hom-ML-V2, 48 PerSpect-EL, 24 GeoPPI, 26 TopNetTree, 25 mC…”
InstructPLM-mu: protein fitness prediction from structure instructions
Xu, J., Shi, Y., Lang, L., Cui, T., Zhang, Z., Chen, G., Qiu, J., & Heng, P.-A. (2026) · Briefings in Bioinformatics · doi:10.1093/bib/bbag400 · Cited by 2
Metadata
- Authors
- Wang, M., Cang, Z., & Wei, G.-W.
- Affiliations
- Michigan State University (US)
- Corresponding author
- Wei, Guo-Wei
- Venue
- Nature Machine Intelligence
- Published
- 2020-02-14
- Volume
- 2
- Issue
- 2
- Pages
- 116-123
- ISSN
- 2522-5839 (ISSN-L)
- PMC
- PMC7223817
- OpenAlex
- W3005822506
- Semantic Scholar
- 544ee14a801abef4019e1962a38e9a1f936ef01a
- Cited by
- 202
- References
- 54
- Citations by year
- 2026: 19 · 2025: 41 · 2024: 39 · 2023: 41 · 2022: 32 · 2021: 22 · 2020: 7
- FWCI
- 8.27
- Type
- journal article
- Language
- en
- Open access
- green
- www.ncbi.nlm.nih.gov
- Licence
- Locations
- Nature Machine Intelligence (publishedVersion)
- PubMed Central (publishedVersion)
- PubMed (publishedVersion)
- Funders
- National Science Foundation (DMS-1721024, DMS 1761320, DMS-1761320, IIS1900473)
- U.S. Department of Health & Human Services | NIH | National Institute of General Medical Sciences (GM126189)
- Bristol-Myers Squibb
- Topics
- Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Machine Learning in Bioinformatics
- Physical Sciences › Computer Science › Computational Theory and Mathematics › Topological and Geometric Data Analysis
- Life Sciences › Pharmacology, Toxicology and Pharmaceutics › Drug Discovery › Computational Drug Discovery Methods
- Keywords
- protein-protein interaction, ΔΔG, mutation effects, persistent homology, algebraic topology, protein-protein complex, convolutional neural networks, gradient boosting decision tree, topological invariants
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
