SKEMPI 2.0: an updated benchmark of changes in protein–protein binding energy, kinetics and thermodynamics upon mutation
Open accessPublished version of a preprint
The DOI matches its Crossref record; 79 of the 80 references it cites were found in a registry or an index.
Abstract
We present SKEMPI 2.0, a major update to our database of binding free energy changes upon mutation for structurally resolved protein-protein interactions. This version now contains manually curated binding data for 7085 mutations, an increase of 133%, including changes in kinetics for 1844 mutations, enthalpy and entropy changes for 443 mutations, and 440 mutations which abolish detectable binding.
Overview
CoChat Summary
## TL;DR
SKEMPI 2.0 is an updated manually curated benchmark of mutation effects on structurally resolved protein–protein interactions, expanding the database to 7085 mutations and adding kinetics, thermodynamics, and non-binding cases. The dataset is broad but strongly biased toward certain interaction classes and well-studied systems, and the paper provides annotations useful for analyzing homology and mutant cycles.
## Key findings
- The database contains 7085 mutations, a 133% increase over the previous version.
- It includes 1844 mutations with kinetics changes, 443 with enthalpy and entropy changes, and 440 mutations that abolish detectable binding.
- The entries span 295 publications and include 610 double mutant cycles, 218 additional double mutant cycles in a third-mutation background, 45 triple mutant cubes, and 14 fourth-order interaction examples.
- Reported ranges include ΔΔG from –12.4 to 12.4 kcal/mol, Δlog10(kon) from –3.6 to 2.4, Δlog10(koff) from –6.0 to 6.8, ΔΔH from –18.3 to 26.5 kcal/mol, and ΔΔS from –61 to 80 cal/mol·K.
## Methods
- Manual literature curation of structurally resolved protein–protein interaction mutations, with stringent checks to match the structure and affinity paper to the same protein, species, cofactors, chains, and post-translational modifications.
- Data were collected from prior SKEMPI entries and new literature searches; the paper states that just over two fifths of the data come from the previous SKEMPI version, and new data were supplemented from sources including ABbind, PROXiMATE, and dbMPIKT.
- Entries were annotated with wild-type and mutant affinities, kinetic and thermodynamic parameters when available, experimental method, residue location, homologous interactions, and cleaned PDB files; the paper also describes calculations from reported values when needed.
## Limitations
- The paper states that the database is biased toward the interests and capabilities of the research community, with over-representation of certain interaction classes and organisms.
- It also notes a structural selection bias: only interactions with structures are included, so there is no representation of “fuzzy” complexes; this is stated in the paper.
- Evident from the text: many error comparisons rely on pairs from different groups or techniques, and the paper notes that some entries near detection limits have large errors, especially very destabilizing mutations.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Figures


Verification
Claims the paper makes
0 Valid · 0 Not valid · 1 Not checked · 74 Own results and interpretations
None of the sources cited for this claim have been checked yet.
In the paper
“The largest contribution comes from the group of the late Michael Laskowski Jr., a systematic study of all possible mutations at selected sites in the turkey ovomucoid third domain and its inhibitory interactions with four proteases (Lu et al., 2001), as well as studies of interactions of the same domain in other bird species and the design of ultra-high affinity broad-spectrum inhibitors.”
§ SKEMPI 2.0: an updated benchmark of changes in protein–protein binding energy, kinetics and thermodynamics upon mutation > 3 Results and discussion > 3.2 Notable studies comprising SKEMPI 2.0 The investigations from which SKEMPI data is derived are diverse, spanning many biological processes and reported in 295 publications including systematic scans, alanine and homolog scanning, design studies including computational design and designs derived from phage display, double mutant cycle studies, antibody engineering, biologic drug design and the evaluation of pathological mutations. The largest contribution comes from the group of the late Michael Laskowski Jr., a systematic study of all possible mutations at selected sites in the turkey ovomucoid third domain and its inhibitory interactions with four proteases (Lu et al., 2001), as well as studies of interactions of the same domain in other bird species and the design of ultra-high affinity broad-spectrum inhibitors. Substantial data also come from investigations into the inhibitory interactions of class A β-lactamases from the groups of Gideon Schreiber (e.g. Reichmann et al., 2007) and Timothy Palzkill (e.g. Brown et al., 2013), as well as cytokine receptor interactions, in particular studies of type I interferons also from the Schreiber group (e.g. Roisman et al., 2001) and that of K. Christopher Garcia (Thomas et al., 2011), but also the study of the GM-CSF/GMRα interaction from the group of Michael W. Parker (Broughton et al., 2016). Other prominent sources of data are studies into hormone receptor interactions, in particular the human growth hormone receptor from the group of Jim Wells (e.g. Cunningham and Wells, 1991) and the prolactin receptor from the group of Michael E. Hodsdon (Kulkarni et al., 2010), as well as studies into antigen recognition including the combined computational and experimental design study to enhance affinity of the AQC2 antibody to integrin α-1 from the group of Herman Van Vlijmen (Clark et al., 2006), the dissection of the interactions of broadly neutralizing antibodies targeting HIV gp120 (Clark et al., 2017) from the group of Richard A. Friesner, and various investigations from the group of Roy A. Mariuzza (e.g. Dall'Acqua et al., 1998). Also notable is the alanine scanning of the urokinase-type plasminogen activator and its receptor from the group of Michael Ploug (Gardsvoll et al., 2006), studies of Ras effector interactions from the group of Christian Herrmann (e.g. Kiel et al., 2004), investigations of pMHC/TCR interactions from the group of Brian Baker (e.g. Piepenbrink et al., 2013), and investigations into the cognate and non-cognate recognition of Escherichia coli colicin DNase bacteriotoxins by their immunity proteins from the group of Colin Kleanthous (e.g. Li et al., 1997).
plain- 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
“For 84% of 1741 such pairs, both entries give a ΔΔG value within 1 kcal.mol−1 of each other.”
§ SKEMPI 2.0: an updated benchmark of changes in protein–protein binding energy, kinetics and thermodynamics upon mutation > 3 Results and discussion > 3.3 Range and error Range: The changes in binding free energy upon mutation range from –12.4 to 12.4 kcal.mol−1, as in SKEMPI 1.1, with Δ log 10kon ranging from –3.6 to 2.4, Δ log 10koff ranging from –6.0 to 6.8, ΔΔH ranging from –18.3 to 26.5 kcal.mol−1, and ΔΔS ranging from –61 to 80 cal.mol−1.K−1. Around 60 mutants are very destabilizing, reducing binding energy by 8 kcal.mol−1 or more. These are all in enzyme/inhibitor complexes such as the inhibition of acetylcholinesterase by the snake venom fasciculin, or the inhibition of enzymes which would be detrimental should they unbind and become active in the wrong location, such as nucleases (barnase/barstar, colicin E9 DNase/Im9, RNase A/angiogenin) and proteases (such as trypsin/BPTI). These interactions tend to be around picomolar affinity and are at the upper limit of what can be detected, due to the time required to reach equilibrium and the low concentrations required by the mass action law to probe informative regions of the binding curve. These very destabilizing mutations reduce affinity into the micromolar range, near the lower limit of what can be quantified using standard methods. As a consequence of both mutant and wild-type affinities being near detection thresholds, errors in these entries are typically large. Further, while some mutations may cause changes in affinity larger than seven orders of magnitude, the absence of affinities for such mutations in the benchmark can be explained by the fact that such mutations would involve affinities beyond the upper or lower limit. Indeed, there are new entries in which single or double substitutions reduce binding from tens of picomolar to having no detectable binding. For many of the highly destabilizing mutations a crystal structure for the mutant has also been solved, and the 30 most stabilizing mutations in the database (ΔΔG ¡ –5 kcal.mol−1) consist of the reverse mutation applied to these structures. These are mostly single or double mutants, but include mutations to up to 27 residues of the non-cognate Colicin E2/Im9 complex, which move it toward the cognate E9/Im9 in sequence space (Li et al., 1997). Errors: Standard errors in KD are typically reported in the order of 50%, around 0.25 kcal.mol−1. These estimates are derived by repeat measurements using the same equipment, environment and protocol, and thus do not include errors arising from systematic bias. Such biases can, however, be estimated from pairs of entries in which the same mutation is evaluated by different groups or using different techniques. For 84% of 1741 such pairs, both entries give a ΔΔG value within 1 kcal.mol−1 of each other. For 704 pairs for which kon is available for both, 80% have Δ log 10(kon) within 0.5 of each other. For 702 pairs for which koff is available, 83% have Δ log 10(koff) within 0.5. For the 62 pair with both ΔΔH and ΔΔS values, 61% have ΔΔH within 3.0 kcal.mol−1 of each other and 58% have ΔΔS within 10 cal.mol−1.K−1 of each other.
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 states — it cites no source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“For many of the highly destabilizing mutations a crystal structure for the mutant has also been solved, and the 30 most stabilizing mutations in the database (ΔΔG ¡ –5 kcal.mol−1) consist of the reverse mutation applied to these structures.”
§ SKEMPI 2.0: an updated benchmark of changes in protein–protein binding energy, kinetics and thermodynamics upon mutation > 3 Results and discussion > 3.3 Range and error Range: The changes in binding free energy upon mutation range from –12.4 to 12.4 kcal.mol−1, as in SKEMPI 1.1, with Δ log 10kon ranging from –3.6 to 2.4, Δ log 10koff ranging from –6.0 to 6.8, ΔΔH ranging from –18.3 to 26.5 kcal.mol−1, and ΔΔS ranging from –61 to 80 cal.mol−1.K−1. Around 60 mutants are very destabilizing, reducing binding energy by 8 kcal.mol−1 or more. These are all in enzyme/inhibitor complexes such as the inhibition of acetylcholinesterase by the snake venom fasciculin, or the inhibition of enzymes which would be detrimental should they unbind and become active in the wrong location, such as nucleases (barnase/barstar, colicin E9 DNase/Im9, RNase A/angiogenin) and proteases (such as trypsin/BPTI). These interactions tend to be around picomolar affinity and are at the upper limit of what can be detected, due to the time required to reach equilibrium and the low concentrations required by the mass action law to probe informative regions of the binding curve. These very destabilizing mutations reduce affinity into the micromolar range, near the lower limit of what can be quantified using standard methods. As a consequence of both mutant and wild-type affinities being near detection thresholds, errors in these entries are typically large. Further, while some mutations may cause changes in affinity larger than seven orders of magnitude, the absence of affinities for such mutations in the benchmark can be explained by the fact that such mutations would involve affinities beyond the upper or lower limit. Indeed, there are new entries in which single or double substitutions reduce binding from tens of picomolar to having no detectable binding. For many of the highly destabilizing mutations a crystal structure for the mutant has also been solved, and the 30 most stabilizing mutations in the database (ΔΔG ¡ –5 kcal.mol−1) consist of the reverse mutation applied to these structures. These are mostly single or double mutants, but include mutations to up to 27 residues of the non-cognate Colicin E2/Im9 complex, which move it toward the cognate E9/Im9 in sequence space (Li et al., 1997). Errors: Standard errors in KD are typically reported in the order of 50%, around 0.25 kcal.mol−1. These estimates are derived by repeat measurements using the same equipment, environment and protocol, and thus do not include errors arising from systematic bias. Such biases can, however, be estimated from pairs of entries in which the same mutation is evaluated by different groups or using different techniques. For 84% of 1741 such pairs, both entries give a ΔΔG value within 1 kcal.mol−1 of each other. For 704 pairs for which kon is available for both, 80% have Δ log 10(kon) within 0.5 of each other. For 702 pairs for which koff is available, 83% have Δ log 10(koff) within 0.5. For the 62 pair with both ΔΔH and ΔΔS values, 61% have ΔΔH within 3.0 kcal.mol−1 of each other and 58% have ΔΔS within 10 cal.mol−1.K−1 of each other.
plain- 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
“For the 62 pair with both ΔΔH and ΔΔS values, 61% have ΔΔH within 3.0 kcal.mol−1 of each other and 58% have ΔΔS within 10 cal.mol−1.K−1 of each other.”
§ SKEMPI 2.0: an updated benchmark of changes in protein–protein binding energy, kinetics and thermodynamics upon mutation > 3 Results and discussion > 3.3 Range and error Range: The changes in binding free energy upon mutation range from –12.4 to 12.4 kcal.mol−1, as in SKEMPI 1.1, with Δ log 10kon ranging from –3.6 to 2.4, Δ log 10koff ranging from –6.0 to 6.8, ΔΔH ranging from –18.3 to 26.5 kcal.mol−1, and ΔΔS ranging from –61 to 80 cal.mol−1.K−1. Around 60 mutants are very destabilizing, reducing binding energy by 8 kcal.mol−1 or more. These are all in enzyme/inhibitor complexes such as the inhibition of acetylcholinesterase by the snake venom fasciculin, or the inhibition of enzymes which would be detrimental should they unbind and become active in the wrong location, such as nucleases (barnase/barstar, colicin E9 DNase/Im9, RNase A/angiogenin) and proteases (such as trypsin/BPTI). These interactions tend to be around picomolar affinity and are at the upper limit of what can be detected, due to the time required to reach equilibrium and the low concentrations required by the mass action law to probe informative regions of the binding curve. These very destabilizing mutations reduce affinity into the micromolar range, near the lower limit of what can be quantified using standard methods. As a consequence of both mutant and wild-type affinities being near detection thresholds, errors in these entries are typically large. Further, while some mutations may cause changes in affinity larger than seven orders of magnitude, the absence of affinities for such mutations in the benchmark can be explained by the fact that such mutations would involve affinities beyond the upper or lower limit. Indeed, there are new entries in which single or double substitutions reduce binding from tens of picomolar to having no detectable binding. For many of the highly destabilizing mutations a crystal structure for the mutant has also been solved, and the 30 most stabilizing mutations in the database (ΔΔG ¡ –5 kcal.mol−1) consist of the reverse mutation applied to these structures. These are mostly single or double mutants, but include mutations to up to 27 residues of the non-cognate Colicin E2/Im9 complex, which move it toward the cognate E9/Im9 in sequence space (Li et al., 1997). Errors: Standard errors in KD are typically reported in the order of 50%, around 0.25 kcal.mol−1. These estimates are derived by repeat measurements using the same equipment, environment and protocol, and thus do not include errors arising from systematic bias. Such biases can, however, be estimated from pairs of entries in which the same mutation is evaluated by different groups or using different techniques. For 84% of 1741 such pairs, both entries give a ΔΔG value within 1 kcal.mol−1 of each other. For 704 pairs for which kon is available for both, 80% have Δ log 10(kon) within 0.5 of each other. For 702 pairs for which koff is available, 83% have Δ log 10(koff) within 0.5. For the 62 pair with both ΔΔH and ΔΔS values, 61% have ΔΔH within 3.0 kcal.mol−1 of each other and 58% have ΔΔS within 10 cal.mol−1.K−1 of each other.
plain- 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
“Standard errors in KD are typically reported in the order of 50%, around 0.25 kcal.mol−1.”
§ SKEMPI 2.0: an updated benchmark of changes in protein–protein binding energy, kinetics and thermodynamics upon mutation > 3 Results and discussion > 3.3 Range and error Range: The changes in binding free energy upon mutation range from –12.4 to 12.4 kcal.mol−1, as in SKEMPI 1.1, with Δ log 10kon ranging from –3.6 to 2.4, Δ log 10koff ranging from –6.0 to 6.8, ΔΔH ranging from –18.3 to 26.5 kcal.mol−1, and ΔΔS ranging from –61 to 80 cal.mol−1.K−1. Around 60 mutants are very destabilizing, reducing binding energy by 8 kcal.mol−1 or more. These are all in enzyme/inhibitor complexes such as the inhibition of acetylcholinesterase by the snake venom fasciculin, or the inhibition of enzymes which would be detrimental should they unbind and become active in the wrong location, such as nucleases (barnase/barstar, colicin E9 DNase/Im9, RNase A/angiogenin) and proteases (such as trypsin/BPTI). These interactions tend to be around picomolar affinity and are at the upper limit of what can be detected, due to the time required to reach equilibrium and the low concentrations required by the mass action law to probe informative regions of the binding curve. These very destabilizing mutations reduce affinity into the micromolar range, near the lower limit of what can be quantified using standard methods. As a consequence of both mutant and wild-type affinities being near detection thresholds, errors in these entries are typically large. Further, while some mutations may cause changes in affinity larger than seven orders of magnitude, the absence of affinities for such mutations in the benchmark can be explained by the fact that such mutations would involve affinities beyond the upper or lower limit. Indeed, there are new entries in which single or double substitutions reduce binding from tens of picomolar to having no detectable binding. For many of the highly destabilizing mutations a crystal structure for the mutant has also been solved, and the 30 most stabilizing mutations in the database (ΔΔG ¡ –5 kcal.mol−1) consist of the reverse mutation applied to these structures. These are mostly single or double mutants, but include mutations to up to 27 residues of the non-cognate Colicin E2/Im9 complex, which move it toward the cognate E9/Im9 in sequence space (Li et al., 1997). Errors: Standard errors in KD are typically reported in the order of 50%, around 0.25 kcal.mol−1. These estimates are derived by repeat measurements using the same equipment, environment and protocol, and thus do not include errors arising from systematic bias. Such biases can, however, be estimated from pairs of entries in which the same mutation is evaluated by different groups or using different techniques. For 84% of 1741 such pairs, both entries give a ΔΔG value within 1 kcal.mol−1 of each other. For 704 pairs for which kon is available for both, 80% have Δ log 10(kon) within 0.5 of each other. For 702 pairs for which koff is available, 83% have Δ log 10(koff) within 0.5. For the 62 pair with both ΔΔH and ΔΔS values, 61% have ΔΔH within 3.0 kcal.mol−1 of each other and 58% have ΔΔS within 10 cal.mol−1.K−1 of each other.
plain- 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.
50 more claims
Create a free account to see all 75 claims and every source verdict.
- Probably not a substantial claim. 12
- It may state two separate claims as one. 11
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?
The cited source says what the claim attributes to it.
In the citing paper
“Additional dedicated datasets summarize curated information about properties and structure of antibodies.These databases include the structural database of kinetics and energetics of mutant protein interactions (SKEMPI) [17], and the antibody binding mutational database (AB-Bind) [18].”
From Computational and artificial intelligence-based methods for antibody development (2023) · cites this paper as [17]
The cited source says what the claim attributes to it.
In the citing paper
“This dataset contains 7,085 entries, including single- and multi-point mutations.”
From A topology-based network tree for the prediction of protein–protein binding affinity changes following mutation (2020) · cites this paper as [39]
In the citing paper
“This model was evaluated by split-by-complex fivefold cross-validation over the Structural Kinetic and Energetic database of Mutant Protein Interactions (SKEMPI) V2.0 dataset (38).”
From Deep learning guided optimization of human antibody against SARS-CoV-2 variants with broad neutralization (2022) · cites this paper as [38]
In the citing paper
“The training and validation datasets were constructed from the SKEMPI V2.0 dataset (38), the largest antibody–antigen binding affinity dataset, which contains 342 complexes and 5,318 mutations in total.”
From Deep learning guided optimization of human antibody against SARS-CoV-2 variants with broad neutralization (2022) · cites this paper as [38]
Citations
Cites
Berman, H. M. (2000) · Nucleic Acids Research · doi:10.1093/nar/28.1.235 · Cited by 40,196 · cited 2×
“Processed PDB files: The PDB files for the interactions, as downloaded from the Protein Data Bank [8], often contain multiple copies of the interacting proteins in the unit cell or other chains irrelevant to the interaction.”
MethodOverview of the CCP 4 suite and current developments
Winn, M. D., Ballard, C. C., Cowtan, K. D., Dodson, E. J., Emsley, P., Evans, P. R., Keegan, R. M., Krissinel, E. B., Leslie, A. G. W., McCoy, A., et al. (2011) · Acta Crystallographica Section D Biological Crystallography · doi:10.1107/s0907444910045749 · Cited by 12,353 · cited 10×
“, 289, 5528–5533. Kastritis,P.L. et al. (2011) A structure-based benchmark for protein-protein”
BackgroundInfluentialmCSM: Predicting the effects of mutations in proteins using graph-based signatures
Pires, D. E. V., Ascher, D. B., & Blundell, T. L. (2014) · Bioinformatics · doi:10.1093/bioinformatics/btt691 · Cited by 1,066
Computational Design of Proteins Targeting the Conserved Stem Region of Influenza Hemagglutinin
Fleishman, S. J., Whitehead, T. A., Ekiert, D. C., Dreyfus, C., Corn, J. E., Strauch, E.-M., Wilson, I. A., & Baker, D. (2011) · Science · doi:10.1126/science.1202617 · Cited by 585
Vreven, T., Moal, I. H., Vangone, A., Pierce, B. G., Kastritis, P. L., Torchala, M., Chaleil, R., Jiménez-García, B., Bates, P. A., Fernandez-Recio, J., et al. (2015) · Journal of Molecular Biology · doi:10.1016/j.jmb.2015.07.016 · Cited by 424
Self-citationThorn, K. S., & Bogan, A. A. (2001) · Bioinformatics · doi:10.1093/bioinformatics/17.3.284 · Cited by 403 · cited 3×
“1 were found by checking the references in the ASEdb [71] and PINT [36] databases,”
MethodInfluentialStructural Linkage between Ligand Discrimination and Receptor Activation by Type I Interferons
Thomas, C., Moraga, I., Levin, D., Krutzik, P. O., Podoplelova, Y., Trejo, A., Lee, C., Yarden, G., Vleck, S. E., Glenn, J. S., et al. (2011) · Cell · doi:10.1016/j.cell.2011.06.048 · Cited by 356
Plattner, N., Doerr, S., De Fabritiis, G., & Noé, F. (2017) · Nature Chemistry · doi:10.1038/nchem.2785 · Cited by 345
BeAtMuSiC: prediction of changes in protein–protein binding affinity on mutations
Dehouck, Y., Kwasigroch, J. M., Rooman, M., & Gilis, D. (2013) · Nucleic Acids Research · doi:10.1093/nar/gkt450 · Cited by 338
Strategy for analysing the co-operativity of intramolecular interactions in peptides and proteins
Horovitz, A., & Fersht, A. R. (1990) · Journal of Molecular Biology · doi:10.1016/0022-2836(90)90275-q · Cited by 335 · cited 2×
“Higher order interaction terms can be garnered from higher cycles, such as triple mutant cubes, constructed from the energies of the wild-type, three single mutants, three corresponding double mutants and the triple mutant (Horovitz and Fersht, 1990).”
Background
Cited by
Items 301–382 of 382
abCAN: a Practical and Novel Attention Network for Predicting Mutant Antibody AffinityPreprint
Gong, C., Shen, Y., Liu, H., & Ming, W. (2024) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.1101/2024.12.02.625958 · Cited by 3
ProAffinity++Preprint
Xu, K., Di, Z., Zhao, J., Zhang, H., & Bu, D. (2025) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.1101/2025.10.31.685718 · Cited by 2
“ProAffinity++ shows strong performance on the antigen-antibody dataset AT-Bind and the missense mutation dataset SKEMPI[18] demonstrates its generalization capability.”
Zhuang, J., Li, Z., Wang, S., Zheng, R., & Zhang, G. (2025) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.1101/2025.11.15.688659 · Cited by 2
A General Framework for Injecting Biophysical Priors into Protein EmbeddingsPreprint
Feldman, J. D., Maechler, A., Wang, D., & Shakhnovich, E. I. (2025) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.64898/2025.12.23.696257 · Cited by 3
“While SKEMPI2 contains roughly 350 curated protein complexes, it suffers from substantial sequential and structural redundancy, with many complexes exhibiting high structural and sequential similarity Bushuiev et al. (2024); Jankauskait˙e et al. (2018).”
InfluentialProtein-peptide Interaction Representation Learning with Pretrained Language ModelsPreprint
Zhan, X., Zhai, S., Liu, T., Lin, S., Bi, T., Zhu, B., & Siu, S. W. I. (2026) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.64898/2026.01.12.699173 · Cited by 1
“0 [30], SabDab [31], UmamiMeta [32] and published literature.”
Wu, Y., Liu, X., Hrovatin, K., Wu, D., Linker, S., Winkel, M., & Tan, F. (2026) · Scientific Data · doi:10.1038/s41597-026-06878-0 · Cited by 4
Improving Hot Region Prediction by Combining Gaussian Naive Bayes and DBSCAN
Hu, J., Zhou, L., Zhang, X., & Chen, N. (2019) · Lecture notes in computer science · doi:10.1007/978-3-030-26969-2_42
Identification of protein hot regions by integrated machine learning algorithm
Hu, J., Gan, H., Zhang, X., & Chen, N. (2019) · IEEE International Conference on Bioinformatics and Biomedicine (BIBM) · doi:10.1109/bibm47256.2019.8983102
Structural Consequences of Disease‐Related Mutations for Protein–Protein Interactions
Rosell, M., & Fernández‐Recio, J. (2020) · Wiley eBooks · doi:10.1002/9780470015902.a0028975
Self-citationSearching for a mechanistic description of pairwise epistasis in protein systemsPreprint
Barnes, J. E., Miller, C. R., & Ytreberg, F. M. (2021) · doi:10.22541/au.162290894.41749480/v2
Prediction of crucial interactions for icosahedral capsid self-assembly by energy landscape atlasingPreprint
Wu, R., Prabhu, R., Bennett, A. D., Ozkan, A., Agbandje‐McKenna, M., & Sitharam, M. (2020) · arXiv (Cornell University)
Koşaca, M., Şamiloğlu, E., & Karaca, E. (2021) · International Journal of Advances in Engineering and Pure Sciences · doi:10.7240/jeps.920075
Akhmanova, A. S. (2019) · Faculty Opinions – Post-Publication Peer Review of the Biomedical Literature · doi:10.3410/f.735160804.793557859
Repertoire-Scale Measures of Antigen BindingPreprint
Arora, R., & Arnaout, R. (2022) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.1101/2022.06.24.497473
Protein–Protein‐Binding Interfaces
Abali, Z., Ovek, D., Senyuz, S., Keskin, Ö., & Gürsoy, A. (2022) · Wiley eBooks · doi:10.1002/9783527830503.ch2
Effects of Mutations in Proteins on Their Interactions
Greß, A., & Kalinina, O. V. (2022) · Wiley eBooks · doi:10.1002/9783527830503.ch15
Liu, Z., Qian, W., Cai, W., Song, W., Wang, W., Maharjan, D. T., Cheng, W., Chen, J., Xu, F., Wang, H., et al. (2022) · Research Square · doi:10.21203/rs.3.rs-2300902/v1
Kumar, A. K., & Rathore, R. S. (2022) · doi:10.22541/au.167101052.24746048/v1
Computational Prediction Approaches for Predicting Mutation Impact on Protein-Protein Interactions
Ping, Y., Hoekstra, L., & Feenstra, K. A. (2023) · Lecture notes on data engineering and communications technologies · doi:10.1007/978-3-031-28893-7_48 · Cited by 1
Automated Protein Affinity Optimization using a 1D-CNN Deep Learning ModelPreprint
McWhirter, J. L., Mukhopadhyay, A., Farber, P. J., Lakatos, G., & Dixit, S. B. (2023) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.1101/2023.04.12.536512
“In this paper, ZymeSwapNet derived affinity metrics were tested on a subset of mutations listed in the open-source protein-protein affinity database, SKEMPI [41].”
MethodInfluentialGeometric and topological AI for molecular sciences
Wee, J. (2023) · DR-NTU (Nanyang Technological University) · doi:10.32657/10356/165903
Knowledge Graphs for Textbooks: Extraction and Completion Techniques
Yao, Y., Ποτίκας, Π., & Potika, K. (2023) · IEEE International Conference on Artificial Intelligence and Knowledge Engineering (AIKE) · doi:10.1109/aike59827.2023.00014
“In contrast, domain-specific KGs like UMLS and SKEMPI, may offer specific information in medical science and protein-protein interaction [10], [11].”
BackgroundChen, Y. C., Sargsyan, K., Wright, J. D., Chen, Y., Huang, Y., & Lim, C. (2024) · Research Square · doi:10.21203/rs.3.rs-3400169/v2
Database construction for mutant protein interactions
Chen, P., Wang, B., Zhang, J., & Liu, Q. (2024) · Elsevier eBooks · doi:10.1016/b978-0-12-824386-2.00004-3
Monajjemi, M., Shahriari, S., Mollaamin, F., & Najaflou, N. (2023) · Journal of the Chilean Chemical Society · doi:10.4067/s0717-97072023000305924
Protein Structural Bioinformatics: An Overview
Gromiha, M. M., Ridha, F., & Selvaraj, S. (2024) · Elsevier eBooks · doi:10.1016/b978-0-323-95502-7.00026-9
Chen, Y. C., Sargsyan, K., Wright, J. D., Chen, Y., Huang, Y., & Lim, C. (2024) · Research Square · doi:10.21203/rs.3.rs-3400169/v3
Software for Predicting Binding Free Energy of Protein–Protein Complexes and Their Mutants
Jarończyk, M. (2024) · Methods in molecular biology · doi:10.1007/978-1-0716-3985-6_9
Jamialahmadi, B., Chamankhah, M., Kohandel, M., & Ghodsi, A. (2024) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.1101/2024.07.16.603820
Chen, Y. C., Sargsyan, K., Wright, J. D., et al. (2024) · eLife · doi:10.7554/elife.96643.2.sa1
Chen, Y. C., Sargsyan, K., Wright, J. D., Chen, Y., Huang, Y., & Lim, C. (2024) · eLife · doi:10.7554/elife.96643.2.sa0
Insight into Protein Engineering: From In silico Modelling to In vitro Synthesis
Balakrishnan, A., Mishra, S. K., & Georrge, J. J. (2024) · Current Pharmaceutical Design · doi:10.2174/0113816128349577240927071706
Prediction of Protein-Protein Interactions
Pathak, A., Tiwari, V. R., Menon, R., & Sowdhamini, R. (2024) · Elsevier eBooks · doi:10.1016/b978-0-323-95502-7.00146-9
An Antigen Design Method Based on Molecular Dynamics and Deep Learning
Li, H., & Zhou, S. (2024) · IEEE International Conference on Medical Artificial Intelligence (MedAI) · doi:10.1109/medai62885.2024.00084
“The dataset utilized in this study was sourced from PDBbind [31], AB-Bind [32], 3DComplex database [33], and SKEMPI 2.0 [34].”
InfluentialAutomatic Explanation of Protein-Protein Binding Mechanism: A Preliminary Study
Tam, J. Z., Liu, Y., Jain, D. S., Armstrong, G. A., & Chen, B. Y. (2025) · Communications in computer and information science · doi:10.1007/978-3-031-85435-4_7
Otun, M. O. (2025) · Journal of applied science and environmental management · doi:10.4314/jasem.v29i3.34
“Independent validation on external datasets is rare but critical, as exemplified by the SKEMPI database, which aggregates mutation-driven PPI energy changes (Jankauskaitė et al ., 2019).”
Zhang, J., & Xiong, Y. (2025) · Protein Science · doi:10.1002/pro.70110
Kuang, Z., Li, Y., Liu, Z., Zhao, J. M., Shi, L., Akira-Kawai, & Wang, H. (2025) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.1101/2025.05.23.655070
“To further verify TAPPI’s Biological Relevance, we collected entries in the SKEMPI v2 dataset [67] with the ΔΔG labels remaining as an external validation set.”
Ridha, F., & Gromiha, M. M. (2025) · Methods in molecular biology · doi:10.1007/978-1-0716-4662-5_20
BindPred: a framework for predicting protein–protein binding affinity from language model embeddings
Piao, H., Boorla, V. S., Santra, S., & Maranas, C. D. (2026) · Bioinformatics · doi:10.1093/bioinformatics/btag309
“0 data-set is dominated by mutations introduced on a fixed wild-type scaffold, with approximately three-quarters of entries involving a single amino acid substitution and most remaining cases consisting of double substitutions (Jankauskaite et al. 2019, Liu et al. 2024).”
Full-Atom Protein-Protein Interaction Prediction via Atomic Equivariant Attention Network
Wang, C., Yang, C. H., Yang, W., Song, L., & Shi, C. (2025) · ACM International Conference on Information and Knowledge Management (CIKM) · doi:10.1145/3746252.3761352 · Cited by 1
Karim, W., Alvi, R., Zaman, S. B., Abian, A. I., Raiaan, M. A. K., Mukta, S. H., Rashid, M. R. U., Islam, R., Sebastian, Y., & Azam, S. (2026) · Archives of Computational Methods in Engineering · doi:10.1007/s11831-026-10743-z · Cited by 7
“SKEMPI [131] and ACE2 [132] datasets focus on protein–protein interaction affinities, which enable the training of models for mutational effect prediction.”
Yu, G., Bi, X., Zhao, Q., & Wang, J. (2026) · Bioinformatics · doi:10.1093/bioinformatics/btag249
Zhang, Y., & Sitharam, M. (2025) · Computer Physics Communications · doi:10.1016/j.cpc.2025.109979
Kosch, R., Yousefi, B., Laubach, N. C., Truger, M. S., Haferlach, T., Gersting, S. W., Teipel, R., Sonnemann, P., Al-Bazaz, M., Artzenroth, J. C., et al. (2026) · Research Square · doi:10.21203/rs.3.rs-8568919/v1
SE3Bind: SE(3)-equivariant model for antibody-antigen binding affinity predictionPreprint
Subedy, A., Bhadra-Lobo, S., & Lamoureux, G. (2026) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.64898/2026.01.17.700115 · Cited by 2
“…(v.2017) [43] 472 mmCSM-AB [44] 301 Random 0.64 Graphinity [24] AB-Bind S645 [32] 645 AB-Bind S645 [32] 645 Random 0.87 Graphinity [24] SKEMPI 2.0 [19] 608 SKEMPI 2.0 [19] 608 100% CDR identity clusters 0.40 Graphinity [24] AB-Bind S645 [32] 618 AB-Bind S645 [32] 618 100% CDR identity clusters…”
InfluentialSong, Y., Wu, F., Wang, R., Zheng, W., He, B., Yan, Q., Huang, X., Li, Y., Chen, S., Yuan, Q., et al. (2026) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.64898/2026.01.20.700456 · Cited by 1
“To explore the predictive capability of the model on mutations, we additionally constructed a new mutation dataset based on the SKEMPI 2.0 [36] database.”
InfluentialA Compilation of Software Tools for Studying Protein–Protein Interactions
Chowdhury, A., & Jois, S. D. S. (2026) · Challenges and advances in computational chemistry and physics · doi:10.1007/978-3-032-07880-3_9
Wei, S., Zhang, J., Chen, Y., Tang, C., & Tan, J. (2026) · Bioinformatics · doi:10.1093/bioinformatics/btag042 · Cited by 3
“0 database (Jankauskait _ e et al. 2019).”
甄卓文, Ma, T., Li, J., Zheng, D., Pang, P. C., Liu, Y., & Zeng, X. (2025) · IEEE International Conference on Bioinformatics and Biomedicine (BIBM) · doi:10.1109/bibm66473.2025.11356250
“We conduct training and evaluation of our model on the SKEMPI 2.0 dataset [22], which is the largest experimentally validated dataset for antigen–antibody binding affinity.”
InfluentialZhang, H., Liu, Y., Yu, L., Wang, Z., Liu, Y., & Guo, M. (2025) · IEEE International Conference on Bioinformatics and Biomedicine (BIBM) · doi:10.1109/bibm66473.2025.11356932
Accurate Protein-Ligand Affinity Prediction by Modeling Molecular Interactions
Guan, Y., Wang, X., Li, X., Liu, Y., Yang, A., & Chen, J. (2025) · IEEE International Conference on Bioinformatics and Biomedicine (BIBM) · doi:10.1109/bibm66473.2025.11356759
A Pretraining-Based Deep Learning Framework for Protein Mutation Effect Prediction
Zhang, P., Deng, J., Gu, M., & Liu, M. (2025) · International Conference on Algorithms, Computing and Artificial Intelligence (ACAI) · doi:10.1109/acai68217.2025.11406652
Deng, J., Gu, M., Zhang, P., Liu, T., Hu, G., Dong, M., Zhang, Y., Song, Y., Zhang, Y., Liu, M., et al. (2026) · Briefings in Bioinformatics · doi:10.1093/bib/bbag105
“0 [39], which compiles 7085 entries of affinity changes in PPI complexes, and the AB-Bind dataset [40], which contains 1101 G measurements across 32 antibody–antigen complexes.”
InfluentialHoffstedt, M., Wowra, J., Wätzig, H., & Baumann, K. (2026) · mAbs · doi:10.1080/19420862.2026.2644655 · Cited by 1
“16,19,27 Expecting robust cross-target generalization under these conditions may therefore be a false premise rather than a solvable modeling problem.”
Wu, S., Xu, N., Xin, X., Zhang, M., Liu, H., Zhu, H., Wei, Z., Zhao, C., Yu, L., & Feng, W. (2026) · Bioinformatics · doi:10.1093/bioinformatics/btag150 · Cited by 3
“The four benchmark datasets S1131, S4169, S8338, and M1707 (numbers denote mutation counts) were derived from the SKEMPI 2.0 database (Jankauskait _ e et al. 2019) (Fig.”
InfluentialSalt Bridge Builder: Using Residue Distances to Predict Salt Bridge Formation
Sanchez, J., Guo, W., Gao, Y., Li, Y., Qiu, W., & Li, L. (2026) · Journal of Chemical Information and Modeling · doi:10.1021/acs.jcim.6c00516 · Cited by 1
Behbahani, Y. M. (2022) · theses.fr (ABES) · doi:10.70675/bfdded7czb638z455azaedaz4ae9f8d0a13d
Lo, Y., Chang, K., Chu, I.-H., Chen, L.-Y., & Wu, K. (2026) · Protein Science · doi:10.1002/pro.70617
PythiaStudio: a one-stop protein engineering platform powered by Pythia model suite
Sun, J., Shi, K., Li, H., 崔颖鹿, Wang, L., & Wu, B. (2026) · Nucleic Acids Research · doi:10.1093/nar/gkag408
“A self-distillation strategy expands the training data from approximately 4000 to 400 000 samples by generating predictions on all protein–protein interfaces in the SKEMPI v2 dataset [ 16 ], and the final model achieves a Pearson correlation of 0.79 (Table 2 ).”
InfluentialProtein-ligand binding kinetics are primarily controlled by the protein, not the ligandPreprint
Srinivasan, B., Corrionero, A., Barone, M., et al. (2026) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.64898/2026.05.20.726507
“…from the SKEMPI-2 database which includes 1,561 entries encompassing mutational variations of 66 distinct pairs of interaction partners ( Figure 5A , Figure S4 ), 48 shows that differences in binding affinity achieved by mutating interacting proteins was most commonly driven by changes in k off .”
Han, R., Zhang, Y., Liu, X., Fu, L., Pan, T., Xu, J., Wang, X. Y., Zhang, P., Chen, X., Lei, J., et al. (2026) · Nature Machine Intelligence · doi:10.1038/s42256-026-01243-7 · Cited by 2
Antibody-Antigen Affinity Prediction with Chain-Aware Protein Language ModelingPreprint
Singh, H., Malhotra, A., Srivastava, S. P., Singh, R. K., & Gorantla, R. (2026) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.64898/2026.06.19.733375 · Cited by 3
“Curated resources such as AB-Bind, SKEMPI 2.0 and more recent structural databases have expanded benchmarking of binding and mutation effects, while also highlighting the limited size, heterogeneity and assay dependence of experimentally measured affinity data 17,18,37 .”
InfluentialZhou, H., Wang, Y., Shi, X., & Ma, Y. (2026) · Journal of Computer-Aided Molecular Design · doi:10.1007/s10822-026-00867-6 · Cited by 1
Li, X., Wen, Q., Zhang, Z., & Cui, F. (2026) · Biotechnology Advances · doi:10.1016/j.biotechadv.2026.108985
The limits of bio-molecular modeling with large language models: a cross-scale evaluation
Xu, Y., Zhou, Y., Zhao, T., Ma, Z., An, F., & Ren, Z. (2026) · Bioinformatics · doi:10.1093/bioinformatics/btag550
EnerBridge-DPO: Energy-Aware Markov Bridge Inverse Folding for Protein Sequence Design
Rong, D., Lu, H., Zhang, X., Zheng, W., Zhang, F., Zheng, S., & Liu, N. (2026) · Journal of Chemical Information and Modeling · doi:10.1021/acs.jcim.6c01697 · Cited by 4
SAASBench: A Synthetic Antibody–antigen Specificity Benchmark
Umerenkov, D., & Poddiakov, I. (2026) · ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD) · doi:10.1145/3770855.3817459
“SKEMPI 2.0 aggregates experimentally measured changes in binding free energy under point mutations [15], while AB-Bind provides curated mutation-level affinity measurements tailored to antibody engineering [27].”
InfluentialLuo, G., Wang, J., Zhang, S., Zhu, Z., Lu, Z., & Li, M. (2026) · Bioinformatics · doi:10.1093/bioinformatics/btag606
“0 (Jankauskait _ e et al .”
Aligning protein-generative models to experimental fitness with ProteinDPO
Widatalla, T., Borah, A. A., King, S. B., Driscoll, C. L., Rafailov, R., & Hie, B. (2026) · Nature Methods · doi:10.1038/s41592-026-03137-3 · Cited by 53
“Due to SKEMPIv2 containing very few variants per complex (as low as one), we report the correlation of all model likelihoods with all affinity measurements, while for AB-Bind we report correlations averaged across each complex.”
InfluentialPark, S. J., Yim, S., Kim, D., Yoo, K., Hwang, D., Lee, K., Jang, J., & Kim, K. (2026) · Bioinformatics · doi:10.1093/bioinformatics/btag489
“SKEMPI 2.0 (Jankauskait _ e et al. 2019) is the most widely used benchmark for binding affinity prediction upon mutation.”
InfluentialGe, W., Yu, Q., Shuai, J., & Zhao, Q. (2026) · Journal of Chemical Information and Modeling · doi:10.1021/acs.jcim.6c01919 · Cited by 1
Niazi, S. K. (2026) · Cells · doi:10.3390/cells15171560
“SKEMPI 2.0 and related benchmarks [20] Predictive accuracy for mutation-induced changes in binding energy, kinetics, and thermodynamics.”
InfluentialAhrens, R., Ahrens, K., Al-Bazaz, M., Leypoldt, L. B., Bokemeyer, C., Alsdorf, W. H., & Weisel, K. C. (2026) · Blood Cancer Journal · doi:10.1038/s41408-026-01616-6
“The full mathematical speci fi cation, tool implementation, parameter perturbation analysis, cross-complex validation, species cross-reactivity analysis and SKEMPI 2.0 benchmark are shown in the Supplementary Information [8].”
InfluentialMultimodal allostery in a single-domain proteinPreprint
Ma, Y., Jin, Q., Lehner, B., & Weng, C. (2026) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.64898/2026.09.14.750605
乔天星, & Feng, Y. (2026) · The Protein Journal · doi:10.1007/s10930-026-10354-4
Bhati, U., Gupta, S., Kesarwani, V., & Shankar, R. (2026) · International Journal of Biological Macromolecules · doi:10.1016/j.ijbiomac.2026.154380
Jiang, H., Bi, X., Ma, W., Huang, X., Nie, J., Wei, Z., & Zhang, S. (2026) · Pattern Recognition · doi:10.1016/j.patcog.2026.114908
Noske, J., Lepoivre, T., Janzen, M., & Höcker, B. (2026) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.64898/2026.09.22.753561
Rogers, J. R., Cardani, L. P., Shah, N. H., & AlQuraishi, M. (2026) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.64898/2026.09.25.754499
Perthold, J. W., Petrov, D., & Oostenbrink, C. (2026) · Journal of Chemical Information and Modeling · doi:10.1021/acs.jcim.6c01771
Mao, S., Meng, S., & Xiang, Y. (2026) · iScience · doi:10.1016/j.isci.2026.117693
Metadata
- Authors
- Jankauskaitė, J., Jiménez-García, B., Dapkūnas, J., Fernández-Recio, J., & Moal, I. H.
- Affiliations
- Vilnius University (LT)
- Barcelona Supercomputing Center (BSC), Barcelona, Spain
- European Bioinformatics Institute (GB)
- Corresponding author
- Moal, Iain H
- Venue
- Bioinformatics
- Publisher
- Oxford University Press (OUP)
- Published
- 2018-07-18
- Volume
- 35
- Issue
- 3
- Pages
- 462-469
- ISSN
- 1367-4803 (ISSN-L)
- 1367-4811
- PMC
- PMC6361233
- OpenAlex
- W2953103783
- Cited by
- 389
- References
- 80
- Citations by year
- 2026: 63 · 2025: 75 · 2024: 61 · 2023: 50 · 2022: 46 · 2021: 37 · 2020: 33 · 2019: 18 · 2018: 5
- FWCI
- 8.26
- Type
- journal article
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- Locations
- Bioinformatics (publishedVersion, cc-by)
- bioRxiv (Cold Spring Harbor Laboratory) (acceptedVersion, cc-by-nc-nd)
- Europe PMC (PubMed Central) (publishedVersion, cc-by)
- PubMed Central (publishedVersion, cc-by)
- PubMed (publishedVersion)
- LA Referencia (Red Federada de Repositorios Institucionales de Publicaciones Científicas) (acceptedVersion, cc-by-nc-nd)
- UPCommons institutional repository (Universitat Politècnica de Catalunya) (acceptedVersion, cc-by-nc-nd)
- DIGITAL.CSIC (Spanish National Research Council (CSIC)) (submittedVersion)
- RIUR (Universidad de La Rioja) (submittedVersion, other-oa)
- RECERCAT (Consorci de Serveis Universitaris de Catalunya) (submittedVersion, cc-by-nc-nd)
- Funders
- European Molecular Biology Laboratory
- Biotechnology and Biological Sciences Research Council
- Future Leader Fellowship (BB/N011600/1)
- Spanish Ministry of Economy and Competitiveness
- MINECO (BIO2016-79930-R)
- Interreg POCTEFA (EFA086/15)
- European Commission (676566)
- Topics
- Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Protein Structure and Dynamics
- Physical Sciences › Chemistry › Physical and Theoretical Chemistry › thermodynamics and calorimetric analyses
- Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Bioinformatics and Genomic Networks
- Keywords
- protein-protein interaction, point mutations, binding kinetics, enthalpy and entropy changes, binding thermodynamics, sequence-structure-function relationships, protein engineering
- MeSH terms
- Kinetics, Mutation, Protein Binding, Thermodynamics, Databases, Protein
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).
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
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.