Pith.

Machine Learning · arXiv:2611.16361 v1 · submitted 12 Nov 2026 · 91 references

MutAtlas: A PDB-Wide Energy-Guided Atlas of Protein Mutation Effects

Ruihan Guo, Chaoran Cheng, Zhanghan Ni, Neil He, Bangji Yang, Ge Liu

The paper says

“Protein mutation effect prediction is fundamental to protein engineering and disease variant interpretation, yet experimentally measured mutation data remain accurate but extremely sparse.”

more of the abstract

To provide scalable supplementary mutation signals, we construct a PDB-wide mutation augmentation dataset that exhaustively enumerates single-site substitutions on experimentally resolved protein structures and aligns mutation signals from physics-based energy models, protein language models, and inverse folding models. Large-scale analysis under a unified mutation preference representation reveals substantial differences in the consistency, concentration, and substitution patterns of mutation distributions across models, indicating that disagreement is pervasive and reflects conflicting inductive biases rather than random noise. Motivated by these observations, we propose an unsupervised multi-source mutation preference distillation framework that learns from relative mutation preferences while explicitly modeling cross-source disagreement. Without using any experimental mutation labels during training, our approach achieves the best overall performance among the evaluated zero-shot baselines and naive multi-source fusion strategies on ProteinGym. We release the dataset and evaluation pipeline to support reproducible studies of protein mutation effects.

from the abstract

Pith says

Machine review · deepseek-v4-flash · 13 Nov 2026 · what was read was not recorded · not checked by a person

It can be wrong. Challenge this review or reply to it.

Holds up, with conditions. A solid resource paper with a real method contribution, but the headline performance claim is weaker than advertised because of missing error bars and a validation procedure that effectively tunes λ on the test set. This line is wrong

4 major objections · 9 minor

How often objections from reviews like this one are right: not yet measured

Are you an author? What you can do.

Not checked: figures, code, what the 91 references say.

The paper says, at Table 1, §6.2

The review did not quote this passage.

Pith says: Objection 1 of 4 · major

The headline claim (best overall zero-shot performance) rests on a single macro-averaged Spearman value (0.498) with no error bars or significance tests. Given that the differences from VespaG (0.487) and the ESM-IF-only ablation (0.480) are modest, it is unclear whether the improvement is statistically meaningful. The authors should report confidence intervals or perform a statistical test (e.g., paired bootstrap across datasets) to support the claim that the disagreement-aware method is significantly better than the next best baseline.

The paper says, at Table 2, §A

The review did not quote this passage.

Pith says: Objection 2 of 4 · major

The regularization weight λ is selected by sweeping over values (0, 0.01, 0.1, 0.2, 0.5, 1.0) and reporting the best overall Spearman on ProteinGym. This constitutes tuning on the evaluation set, which risks overfitting to the benchmark. The authors should clarify whether a held-out validation set was used for λ selection, or if the reported results are the best among a sweep on the test set. If the latter, the central claim is weakened and should be caveated accordingly.

The paper says, at §5.1, Eq. (2)

The review did not quote this passage.

Pith says: Objection 3 of 4 · major

The student model is initialized from ESM2-650M and fine-tuned on the PDB-wide mutation preferences. Since ESM2-650M is also one of the reference sources (ESM2), the student has direct access to ESM2's preferences through its initialization. This raises a potential circularity: the student may simply retain or refine ESM2's own predictions, and the improvement from disagreement-aware training could partly reflect better utilization of the ESM2 prior rather than genuine multi-source integration.

more of objection 3

The authors should discuss this and ideally include an ablation where the student is initialized from a different backbone (e.g., a randomly initialized model or a different PLM) to disentangle the effect of initialization from the distillation objective.

The paper says, at §6.2, Table 1

The review did not quote this passage.

Pith says: Objection 4 of 4 · major

The ESM-IF-only ablation (ρ=0.480) is described as a dataset ablation that excludes FoldX supervision. However, the student is still initialized from ESM2-650M, which already provides ESM2-like signals. The comparison between the full method (0.498) and ESM-IF-only (0.480) is meant to show the benefit of adding FoldX, but the ESM-IF-only ablation still uses ESM2 initialization and ESM-IF as a reference source.

more of objection 4

It is not a clean ablation of FoldX contribution because the student already has ESM2 knowledge. The authors should clarify what exactly is ablated: is FoldX removed from the reference sources, or is the entire FoldX dataset removed? The text says 'ESM-IF aug only' which suggests only ESM-IF augmentation is used, but the student still has ESM2 initialization. This needs clarification.

9 minor objections

The paper says, at Abstract

Pith says: minor objection 1 of 9

The abstract states 'best overall performance among the evaluated zero-shot baselines and naive multi-source fusion strategies on ProteinGym.' The phrase 'naive multi-source fusion strategies' is vague; consider specifying what these strategies are (e.g., averaging, voting) for clarity.

The paper says, at §3.2, Eq. (1)

Pith says: minor objection 2 of 9

The conversion of FoldX energy differences to logits via ℓ = -(E(i,a) - E(i,a_WT)) is a linear transformation. It might be helpful to note that this preserves ranking but changes scale, which is consistent with the subsequent use of pairwise preferences.

show 7 more

The paper says, at §5.2, Eq. (5)

Pith says: minor objection 3 of 9

The normalization constant τ_k is defined as the average absolute pairwise margin over training pairs. It would be useful to specify whether this is computed once over the entire training set or per mini-batch, as this affects reproducibility.

The paper says, at §5.3, Eq. (9)

Pith says: minor objection 4 of 9

The agreement weight A_{i,a,b} is defined as |K^+_{i,a,b}| / |K|. Since |K|=3, this takes values in {0, 1/3, 2/3, 1}. This is a coarse discretization; the authors might discuss whether finer-grained agreement measures (e.g., based on margin magnitudes) could be beneficial.

The paper says, at §6.1

Pith says: minor objection 5 of 9

The evaluation filters out ProteinGym datasets whose corresponding protein sequences contain fewer than 1000 amino acids. The rationale for this filter should be stated (e.g., computational constraints, relevance to the training data).

The paper says, at Table 1

Pith says: minor objection 6 of 9

The row for 'FoldX (reference)' reports 0.424 overall. However, FoldX is also used as a reference source in the proposed method. It would be informative to include a baseline where FoldX scores are used directly (without any student) to contextualize the improvement from distillation.

The paper says, at §4.3, Fig. 5

Pith says: minor objection 7 of 9

The right panel of Fig. 5 shows per-position Spearman correlation distributions. The text states that ESM2 and ESM-IF have higher ranking similarity, while FoldX has lower correlation with both. It would be helpful to report the median or mean correlation values for each pair to quantify this.

The paper says, at Appendix B.6, Fig. 11

Pith says: minor objection 8 of 9

The entropy–disagreement analysis is interesting but the figure is small and the axes are not clearly labeled. Consider enlarging and adding axis labels for readability.

The paper says, at General

Pith says: minor objection 9 of 9

The paper would benefit from a discussion of limitations, particularly regarding the reliance on FoldX for the physics-based signal (FoldX has known limitations in capturing certain mutation effects) and the restriction to single-site substitutions.

What would settle it

Independently recompute the macro-averaged Spearman for the disagreement-aware (λ=0.2) variant on the 201 ProteinGym substitution datasets using the released code and check whether the overall value reproduces within ±0.005 of 0.498. If it does, the central claim holds; if it shifts by more than 0.01, the claim weakens.

from the skeptic's reading

How the review went

4 readings by deepseek-v4-flash · 13 Nov 2026

  1. Reader holds up, with conditions confidence moderate
  2. Skeptic partly agrees 7 s · 16,342 tokens read, 627 written
  3. Editor wrote the verdict above 6 s · 17,219 tokens read, 595 written
  4. Referee major revision: 4 major objections, 9 minor 11 s · 16,579 tokens read, 1,574 written

Not run on this paper: the simulated authors' defence.

Pith did not check this review's objections against the paper.

The review in full

The review in brief, machine-written

The paper builds a PDB-wide atlas of ~300M single-site mutation preference scores from FoldX, ESM2, and ESM-IF, aligned per residue. That resource alone is valuable—two orders of magnitude wider than the largest experimental DMS collections. The descriptive analysis of cross-source disagreement (Section 4) is thorough and genuinely informative: they show that FoldX gives a flatter, less conservative landscape, and that disagreement persists even at low-entropy positions, which supports their claim that the sources encode different inductive biases rather than just noise. The disagreement-aware distillation objective (Section 5) is well-motivated and the ablation ladder (hard 0.468 → soft 0.475 → disagreement-aware 0.498) is internally consistent and directionally clear. The λ=0 row exactly reproduces the soft-consensus row, as Eq. 14 predicts, which is a nice sanity check.

The soft spots are real but not fatal. First, the headline claim—best overall zero-shot performance on ProteinGym—rests on a single macro-averaged Spearman over 201 datasets with no error bars or significance tests. The gap to the next best method (VespaG, 0.487 vs 0.498) is modest, and without uncertainty estimates it is impossible to know whether the difference is meaningful. Second, λ=0.2 is selected by sweeping the same metric on the same benchmark (Table 2) with no stated validation procedure. The authors do not hide this—they report the sweep—but it means the reported 0.498 is an optimistically biased estimate. A proper validation split or a held-out tuning set would have addressed this. Third, the ESM-IF-only ablation is described as a dataset ablation, not a method ablation, so the contribution of FoldX is not cleanly isolated. The paper acknowledges this implicitly but the reader has to work to see it.

These are standard weaknesses for a resource+method paper at a top venue. They do not undermine the atlas or the method design; they just mean the performance claim should be read as "comparable to state-of-the-art zero-shot methods" rather than "clearly best." The paper is for researchers working on mutation effect prediction, protein engineering, or multi-source distillation. It deserves a serious referee—the resource alone justifies it. I would recommend acceptance with a request to add error bars and clarify the λ selection procedure.

The strongest objection

The paper's central claim—that disagreement-aware preference distillation from FoldX, ESM2, and ESM-IF achieves the best overall zero-shot performance on ProteinGym—is supported by a coherent method, a clear ablation ladder, and consistent results. The reader's weakest_assumption (lack of error bars, λ selection on the evaluation set, ambiguous ESM-IF-only ablation) are acknowledged limitations but do not invalidate the claim. The paper itself states that λ=0.2 is the best in a sweep (Table 2) and that the ESM-IF-only ablation is a dataset ablation, not a method ablation; these are disclosed. The performance differences (0.498 vs 0.487 for VespaG, vs 0.480 for ESM-IF-only) are modest but the trend across ablations is monotonic and the method is clearly described. No internal inconsistency or hidden assumption undermines the central argument. The paper's own limitations (b24-b26, b144, b154) are stated and not double-counted. The claim is conditional on the released artifacts verifying the numbers, but the argument as presented is sound.

The reader identified three concerns: (1) no error bars or significance tests, (2) λ selected on the evaluation set, (3) ambiguous ESM-IF-only ablation. These are valid points about presentation and rigor, but they do not constitute a load-bearing attack on the central claim. The paper's method is internally consistent, the ablation ladder is monotonic (hard 0.468 → soft 0.475 → disagreement-aware 0.498), and the λ sweep is disclosed in Table 2. The ESM-IF-only ablation is clearly labeled as a dataset ablation (Table 1: 'ESM-IF aug only'), not a method ablation; the contribution of FoldX is therefore isolated, albeit not perfectly. The performance differences are modest but the trend is clear. The paper's own limitations are stated. The central claim is that the method achieves the best overall performance among evaluated zero-shot baselines and naive fusion strategies; this is supported by the reported numbers. The lack of error bars is a weakness in presentation but not a logical flaw. The reader's verdict of CONDITIONAL is appropriate; no adjustment is needed. I partially agree with the reader because the concerns are real but not fatal.

The referee's report

If the results hold, the work provides a valuable resource (the PDB-wide mutation atlas) and a principled framework for integrating heterogeneous computational mutation signals without experimental labels. The analysis of cross-source disagreement is informative, and the method's performance is competitive. The release of dataset and code would support reproducibility.

Pith's reading of the paper (a machine's summary, not the authors')

Proteins are chains of amino acids, and swapping one amino acid for another can change what a protein does. Laboratories can measure such changes directly, but only for a small number of proteins. This paper tries to fill the gap with computation.

The authors took 86,479 protein chains from the Protein Data Bank and, for each position in each chain, computed what happens when each of the 19 other amino acids is substituted in. They used three kinds of computer models: FoldX, a physics-based energy program; ESM2, a language model trained on many protein sequences; and ESM-IF, a model that designs sequences for a given structure. That is more than 300 million mutation scores. They then put all scores into a common format: a list of 20 numbers per position describing how much each model prefers each amino acid.

The three models often disagree. The authors argue the disagreement is systematic, not random: FoldX prefers small, flexible amino acids and promotes substitutions, while the two neural models stick closer to the original amino acid. Their training method treats this disagreement as useful information. A student model, starting from ESM2, is fine-tuned to match preferences that most sources support, while being discouraged from making strong confident calls where the sources conflict. No experimental mutation data is used anywhere in training.

On ProteinGym, a standard collection of measured mutation effects, the resulting model scores 0.498 macro-averaged Spearman, slightly above the best sequence-only baseline they compared (VespaG at 0.487). The gaps are small and reported without error bars, and the main guarantee depends on the promised release of the dataset and code.

Core claim. The abstract states: 'Without using any experimental mutation labels during training, our approach achieves the best overall performance among the evaluated zero-shot baselines and naive multi-source fusion strategies on ProteinGym' (b4), operationalized as 0.498 macro-averaged Spearman overall in Table 1. If correct, disagreement-aware distillation of FoldX, ESM2, and ESM-IF preferences ranks single-site mutations better than any of the compared sources or consensus fusions on ProteinGym's substitution benchmarks.

The premise it rests on. The headline ranking depends on the premise that the reported differences are statistically real: each method is a single macro-averaged Spearman over 201 datasets with no error bars or significance tests, and the best λ (=0.2) is the value that maximizes the same metric in the Table 2 sweep, with no stated validation procedure. If 0.498 vs 0.487 (VespaG) or 0.498 vs 0.480 (the ESM-IF-only ablation) is within noise, or if λ was effectively fit to ProteinGym, the central claim reduces to 'comparable to existing zero-shot methods.' A separate premise, that FoldX adds complementary signal, rests on the same ambiguously described ESM-IF-only ablation.

Check this review

Pith stored no record of this review's passes and took no fingerprint of it. Reviews made by code from 13 Nov 2026 on carry both.

How often reviews like this one are right

Reviews like this one: written by deepseek-v4-flash, where what was read was not recorded.

How often objections from reviews like this one are right: not yet measured

Every value above, in the review's public record (JSON)

Reviews by people

This review was written by a machine. No person has reviewed this paper on Pith yet. A scientist listed on Pith can review it for a fee, and their review is published here whatever it concludes.

Request a human review

How Pith checks papers · The full audit of this paper

The paper

Abstract

Protein mutation effect prediction is fundamental to protein engineering and disease variant interpretation, yet experimentally measured mutation data remain accurate but extremely sparse. To provide scalable supplementary mutation signals, we construct a PDB-wide mutation augmentation dataset that exhaustively enumerates single-site substitutions on experimentally resolved protein structures and aligns mutation signals from physics-based energy models, protein language models, and inverse folding models. Large-scale analysis under a unified mutation preference representation reveals substantial differences in the consistency, concentration, and substitution patterns of mutation distributions across models, indicating that disagreement is pervasive and reflects conflicting inductive biases rather than random noise. Motivated by these observations, we propose an unsupervised multi-source mutation preference distillation framework that learns from relative mutation preferences while explicitly modeling cross-source disagreement. Without using any experimental mutation labels during training, our approach achieves the best overall performance among the evaluated zero-shot baselines and naive multi-source fusion strategies on ProteinGym. We release the dataset and evaluation pipeline to support reproducible studies of protein mutation effects.

The paper's own sentences, arranged, kept beside the review and separate from it. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. No sentence in this record was written by a model.

abstract compiled · 2026-11-13

The authors' main claim, as the classifier picked it out 0.99

“Without using any experimental mutation labels during training, our approach achieves the best overall performance among the evaluated zero-shot baselines and naive multi-source fusion strategies on ProteinGym.”

How the main claim is stated: asserted 1.00

What the abstract itself states, as probabilities: claims first 0.07 · states limitation 0.29 · human participants 0.03 · controlled comparison 0.84 · randomized assignment 0.01 · states quantitative result 0.07 · compares with prior results 0.95 · mentions code or data availability 0.96

The abstract, sentence by sentence

This table labels the abstract only (arXiv listing abstract v1). The number is the probability the classifier gave the label.

SentenceRoleThe authors' own workStates the main conclusionStrength
Protein mutation effect prediction is fundamental to protein engineering and disease variant interpretation, yet experimentally measured mutation data remain accurate but extremely sparse. background 0.78 0.07 0.03 asserted 1.00
To provide scalable supplementary mutation signals, we construct a PDB-wide mutation augmentation dataset that exhaustively enumerates single-site substitutions on experimentally resolved protein structures and aligns mutation signals from physics-based energy models, protein language models, and inverse folding models. method 0.61 0.96 0.05 asserted 1.00
Large-scale analysis under a unified mutation preference representation reveals substantial differences in the consistency, concentration, and substitution patterns of mutation distributions across models, indicating that disagreement is pervasive and reflects conflicting inductive biases rather than random noise. result 0.67 0.91 0.19 asserted 0.54
Motivated by these observations, we propose an unsupervised multi-source mutation preference distillation framework that learns from relative mutation preferences while explicitly modeling cross-source disagreement. method 0.59 0.81 0.12 asserted 0.95
Without using any experimental mutation labels during training, our approach achieves the best overall performance among the evaluated zero-shot baselines and naive multi-source fusion strategies on ProteinGym. result 1.00 0.96 0.51 asserted 1.00
We release the dataset and evaluation pipeline to support reproducible studies of protein mutation effects. contribution 0.93 0.89 0.04 asserted 0.96

What kind of paper the classifier reports this is: method or tool 0.81 · evidence from simulation or computation 0.78

What was examined, and what was not

Sources read

  • arXiv listing abstract v1

Not examined

  • full text
  • figures and tables
  • references and citing sentences
  • relations to other papers

Levels run

  • P abstract

Not run

  • S sections
  • T body sentences
  • C citations
  • R cross-paper relations

How the labels were made

Compiled by classifier-1.13.0 via TypeSafe API, question bank abstract-p1-20260923, 2026-11-13: 1 request, 5,019 input tokens.

Every quoted sentence on this record is the authors'. Every label is from Pith's closed vocabulary, with the probability shown beside it. This record as JSON

Figures (11)
Figure 1 of the paper
Figure 1. Overview of mutation signal construction and representation. For a residue selected from an experimentally resolved protein structure, mutation signals are obtained from three complementary sources: physics-based energy evaluation using FoldX, sequence-based prediction using a masked protein language model (ESM2), and structure-conditioned inverse folding using ESM-IF. All signals are aligned to the same residue position and expressed as 20-dimensional mutation logits over the standard amino acid alphabet, forming a unified per-residue mutation representation. ness, and (iii)…
Figure 2 of the paper
Figure 2. WT self-preference at the decision level. Fraction of positions where the WT amino acid is selected as the Top-1 mutation, grouped by WT residue type. ESM2 and ESM-IF exhibit strong WT preservation, whereas FoldX more frequently promotes substitutions
Figure 3 of the paper
Figure 3. WT-to-Mutant substitution behavior at the decision level. FoldX shows weaker WT preservation and broader substitution preferences compared to ESM2 and ESM-IF. Full score-level and decision-level landscapes are reported in Appendix B. 4.2. Native Distributional Sharpness We next analyze how sharply each source expresses its mu￾tation preferences. Here, native distributional sharpness refers to the concentration of each source’s own preference distribution before any cross-source calibration or fusion. As shown in
Figure 4 of the paper
Figure 4. Per-position uncertainty in mutation space. FoldX exhibits substantially higher entropy than ESM2 and ESM-IF, indi￾cating broader and less confident mutation preference distributions. hibit higher ranking similarity, whereas FoldX shows sub￾stantially lower correlation with both neural sources (
Figure 5 of the paper
Figure 5. Cross-method agreement and ranking structure. Left: Top-1 agreement rates. Right: per-position Spearman rank correlation. FoldX exhibits low agreement with the other two models, providing a complementary mutation ranking signal. acids (a, b), we define the reference pairwise margin as ∆ (k) i,a,b = R (k) i (a) − R (k) i (b). (4) A positive margin indicates that source k prefers amino acid a over b under the same sequence context. Although the sign of ∆ (k) i,a,b is invariant to source-specific score scaling, its magnitude is not directly comparable across sources. We therefore…
Figure 6 of the paper
Figure 6. Additional WT self-preference statistics. Top: mean predicted probability assigned to the WT amino acid. Bottom: fraction of positions where WT appears among the Top-3 mutations. Results are grouped by WT residue type. B.3. Detailed WT→Mutant Substitution Matrices We provide full WT→Mutant substitution matrices to complement the qualitative discussion in the main text (Section 4). As shown in
Figure 7 of the paper
Figure 7. Global mutant preference across methods. Frequency with which each amino acid is selected as the Top-1 mutation, aggregated across all residue positions. B.6. Entropy–Disagreement Relationship We further analyze the relationship between per-position uncertainty and cross-method disagreement. While entropy correlates with Top-1 disagreement to some extent, we observe a substantial number of residue positions with low entropy but high disagreement. As illustrated in
Figure 8 of the paper
Figure 8. WT-to-Mutant substitution landscapes (full matrices). Top: average WT-to-Mutant logit matrix (score-level). Bottom: WT-to-Mutant Top-1 optimality matrix (decision-level). Variants of the proposed method. To assess the effect of different supervision formulations, we report results for multiple variants of our approach, including hard-consensus distillation, soft-consensus distillation, and the full disagreement-aware formulation. All variants share the same backbone architecture and training data, and differ only in the design of the training objective. 15
Figure 9 of the paper
Figure 9. Top-1−Top-2 logit gap distributions
Figure 10 of the paper
Figure 10. Top-k overlap between mutation predictions. 16
Figure 11 of the paper
Figure 11. Entropy vs. cross-method disagreement. Relationship between mean per-position entropy and disagreement under k = 3 (top) and k = 5 (bottom). 17

Figures are the authors', reproduced from the paper.

References (91)
  1. [1] Yoon, Jisang , title=. 2021 , url= on Pith
  2. [2] Pultz, Dennis and Friis, Esben and Salomon, Jesper and Hallin, Peter Fischer and Jørgensen, Sarah Baagøe and Reade, Walter and Demkin, Maggie , title=. 2022 , url= on Pith
  3. [3] Announcing the worldwide protein data bank , author=. Nature structural & molecular biology , volume=. 2003 , publisher= on Pith
  4. [4] Stability oracle: a structure-based graph-transformer for identifying stabilizing mutations , author=. BioRxiv , pages=. 2023 , publisher= on Pith
  5. [5] Mega-scale experimental analysis of protein folding stability in biology and design , author=. Nature , volume=. 2023 , publisher= on Pith
  6. [6] Rotamer Density Estimator is an Unsupervised Learner of the Effect of Mutations on Protein-Protein Interaction , author=. bioRxiv , pages=. 2023 , publisher= on Pith
  7. [7] Predicting protein stability changes upon single-point mutation: a thorough comparison of the available tools on a new dataset , author=. Briefings in Bioinformatics , volume=. 2022 , publisher= on Pith
  8. [8] SKEMPI 2.0: an updated benchmark of changes in protein--protein binding energy, kinetics and thermodynamics upon mutation , author=. Bioinformatics , volume=. 2019 , publisher= on Pith
  9. [9] CATH--a hierarchic classification of protein domain structures , author=. Structure , volume=. 1997 , publisher= on Pith
  10. [10] mTM-align: an algorithm for fast and accurate multiple protein structure alignment , author=. Bioinformatics , volume=. 2018 , publisher= on Pith
  11. [11] CLUSTAL W: improving the sensitivity of progressive multiple sequence alignment through sequence weighting, position-specific gap penalties and weight matrix choice. , author=. Nucleic acids research , year= on Pith
  12. [12] The CLUSTAL\_X windows interface: flexible strategies for multiple sequence alignment aided by quality analysis tools. , author=. Nucleic acids research , year= on Pith
  13. [13] CCMpred—fast and precise prediction of protein residue--residue contacts from correlated mutations , author=. Bioinformatics , volume=. 2014 , publisher= on Pith
  14. [14] Discovery of novel gain-of-function mutations guided by structure-based deep learning , author=. ACS synthetic biology , volume=. 2020 , publisher= on Pith
  15. [15] Superluminescent variants of marine luciferases for bioassays , author=. Analytical chemistry , volume=. 2011 , publisher= on Pith
  16. [16] Machine learning-aided engineering of hydrolases for PET depolymerization , author=. Nature , volume=. 2022 , publisher= on Pith
  17. [17] Enzymatic and whole cell catalysis: finding new strategies for old processes , author=. Biotechnology advances , volume=. 2011 , publisher= on Pith
  18. [18] Pharmaceutical and industrial protein engineering: where we are? , author=. Pakistan Journal of Pharmaceutical Sciences , volume= on Pith
  19. [19] Recent advances in microbial production of fuels and chemicals using tools and strategies of systems metabolic engineering , author=. Biotechnology advances , volume=. 2015 , publisher= on Pith
  20. [20] Protein engineering and its applications in food industry , author=. Critical reviews in food science and nutrition , volume=. 2017 , publisher= on Pith
show 71 more
  1. [21] Enzymatic biofuel cells based on protein engineering: Recent advances and future prospects , author=. Biomaterials science , volume=. 2020 , publisher= on Pith
  2. [22] Protein engineering of subtilisins to improve stability in detergent formulations , author=. Journal of biotechnology , volume=. 1993 , publisher= on Pith
  3. [23] P. Langley , title =. Proceedings of the 17th International Conference on Machine Learning (ICML 2000) , address =. 2000 , pages = on Pith
  4. [24] T. M. Mitchell. The Need for Biases in Learning Generalizations. 1980 on Pith
  5. [25] M. J. Kearns , title = on Pith
  6. [26] Machine Learning: An Artificial Intelligence Approach, Vol. I. 1983 on Pith
  7. [27] R. O. Duda and P. E. Hart and D. G. Stork. Pattern Classification. 2000 on Pith
  8. [28] Suppressed for Anonymity , author= on Pith
  9. [29] A. Newell and P. S. Rosenbloom. Mechanisms of Skill Acquisition and the Law of Practice. Cognitive Skills and Their Acquisition. 1981 on Pith
  10. [30] A. L. Samuel. Some Studies in Machine Learning Using the Game of Checkers. IBM Journal of Research and Development. 1959 on Pith
  11. [31] Schymkowitz, Joost and Borg, Jesper and Stricher, Francois and Nys, Robby and Rousseau, Frederic and Serrano, Luis , title = ". Nucleic Acids Research , volume =. 2005 , month =. doi:10.1093/nar/gki387 , url = on Pithdoi:10.1093/nar/gki387
  12. [32] Park, Hahnbeom and Bradley, Philip and Greisen, Per Jr. and Liu, Yuan and Mulligan, Vikram Khipple and Kim, David E. and Baker, David and DiMaio, Frank , title =. Journal of Chemical Theory and Computation , volume =. 2016 , doi = on Pith
  13. [33] Alford, Rebecca F. and Leaver-Fay, Andrew and Jeliazkov, Jeliazko R. and O’Meara, Matthew J. and DiMaio, Frank P. and Park, Hahnbeom and Shapovalov, Maxim V. and Renfrew, P. Douglas and Mulligan, Vikram K. and Kappel, Kalli and Labonte, Jason W. and Pacella, Michael S. and Bon... on Pith
  14. [34] Thomas Steinbrecher and Robert Abel and Anthony Clark and Richard Friesner , keywords =. Free Energy Perturbation Calculations of the Thermodynamics of Protein Side-Chain Mutations , journal =. 2017 , issn =. doi:https://doi.org/10.1016/j.jmb.2017.03.002 , url = on Pithdoi:10.1016/j.jmb.2017.03.002
  15. [35] Geng, Cunliang and Vangone, Anna and Folkers, Gert E. and Xue, Li C. and Bonvin, Alexandre M. J. J. , title =. Proteins: Structure, Function, and Bioinformatics , volume =. doi:https://doi.org/10.1002/prot.25630 , url =. https://onlinelibrary.wiley.com/doi/pdf/10.1002/prot.256... on Pithdoi:10.1002/prot.25630
  16. [36] Zhang, Ning and Chen, Yuting and Lu, Haoyu and Zhao, Feiyang and Alvarez, Roberto Vera and Goncearenco, Alexander and Panchenko, Anna R and Li, Minghui. MutaBind2 : Predicting the impacts of single and multiple mutations on protein-protein interactions. iScience on Pith
  17. [37] Sisi Shan and Shitong Luo and Ziqing Yang and Junxian Hong and Yufeng Su and Fan Ding and Lili Fu and Chenyu Li and Peng Chen and Jianzhu Ma and Xuanling Shi and Qi Zhang and Bonnie Berger and Linqi Zhang and Jian Peng , title =. Proceedings of the National Academy of Sciences... on Pithdoi:10.1073/pnas.2122954119
  18. [38] Deep geometric representations for modeling effects of mutations on protein-protein binding affinity , year =. PLOS Computational Biology , publisher =. doi:10.1371/journal.pcbi.1009284 , author = on Pithdoi:10.1371/journal.pcbi.1009284
  19. [39] Masked inverse folding with sequence transfer for protein representation learning , author=. 2023 , url= on Pith
  20. [40] MSA Transformer , author =. Proceedings of the 38th International Conference on Machine Learning , pages =. 2021 , editor = on Pith
  21. [41] Meier, Joshua and Rao, Roshan and Verkuil, Robert and Liu, Jason and Sercu, Tom and Rives, Alex , booktitle =. Language models enable zero-shot prediction of the effects of mutations on protein function , url = on Pith
  22. [42] Notin, Pascal and Dias, Mafalda and Frazer, Jonathan and Hurtado, Javier Marchena and Gomez, Aidan N. and Marks, Debora and Gal, Yarin , month = jun, year =. Tranception:. Proceedings of the 39th on Pith
  23. [43] Hopf, Thomas A. and Ingraham, John B. and Poelwijk, Frank J. and Sch. Mutation effects predicted from sequence co-variation , journal=. 2017 , month=. doi:10.1038/nbt.3769 , url= on Pithdoi:10.1038/nbt.3769
  24. [44] Riesselman, Adam J. and Ingraham, John B. and Marks, Debora S. , title=. Nature Methods , year=. doi:10.1038/s41592-018-0138-4 , url= on Pithdoi:10.1038/s41592-018-0138-4
  25. [45] Luo, Yunan and Jiang, Guangde and Yu, Tianhao and Liu, Yang and Vo, Lam and Ding, Hantian and Su, Yufeng and Qian, Wesley Wei and Zhao, Huimin and Peng, Jian , title=. Nature Communications , year=. doi:10.1038/s41467-021-25976-8 , url= on Pithdoi:10.1038/s41467-021-25976-8
  26. [46] Frazer, Jonathan and Notin, Pascal and Dias, Mafalda and Gomez, Aidan and Min, Joseph K. and Brock, Kelly and Gal, Yarin and Marks, Debora S. , title=. Nature , year=. doi:10.1038/s41586-021-04043-8 , url= on Pithdoi:10.1038/s41586-021-04043-8
  27. [47] Protein Representation Learning by Geometric Structure Pretraining , author=. The Eleventh International Conference on Learning Representations , year= on Pith
  28. [48] Learning inverse folding from millions of predicted structures , author =. Proceedings of the 39th International Conference on Machine Learning , pages =. 2022 , editor = on Pith
  29. [49] Retrieval-based Controllable Molecule Generation , author=. The Eleventh International Conference on Learning Representations , year= on Pith
  30. [50] Highly accurate protein structure prediction with AlphaFold , author=. Nature , volume=. 2021 , publisher= on Pith
  31. [51] Accurate prediction of protein structures and interactions using a 3-track neural network , author=. Science (New York, N.Y.) , year= on Pith
  32. [52] Ruidong Wu and Fan Ding and Rui Wang and Rui Shen and Xiwen Zhang and Shitong Luo and Chenpeng Su and Zuofan Wu and Qi Xie and Bonnie Berger and Jianzhu Ma and Jian Peng , title =. 2022 , doi =. https://www.biorxiv.org/content/early/2022/07/22/2022.07.21.500999.full.pdf , journal = on Pith
  33. [53] Zeming Lin and Halil Akin and Roshan Rao and Brian Hie and Zhongkai Zhu and Wenting Lu and Nikita Smetanin and Robert Verkuil and Ori Kabeli and Yaniv Shmueli and Allan dos Santos Costa and Maryam Fazel-Zarandi and Tom Sercu and Salvatore Candido and Alexander Rives , title =.... on Pithdoi:10.1126/science.ade2574
  34. [54] Guu, Kelvin and Lee, Kenton and Tung, Zora and Pasupat, Panupong and Chang, Ming-Wei , title =. Proceedings of the 37th International Conference on Machine Learning , articleno =. 2020 , publisher = on Pith
  35. [55] Memorizing Transformers , author=. International Conference on Learning Representations , year= on Pith
  36. [56] Chang Ma and Haiteng Zhao and Lin Zheng and Jiayi Xin and Qintong Li and Lijun Wu and Zhihong Deng and Yang Lu and Qi Liu and Lingpeng Kong , title =. 2023 , doi =. https://www.biorxiv.org/content/early/2023/05/23/2023.02.22.529597.full.pdf , journal = on Pith
  37. [57] Malkov, Yu A. and Yashunin, D. A. , title =. IEEE Trans. Pattern Anal. Mach. Intell. , month =. 2020 , issue_date =. doi:10.1109/TPAMI.2018.2889473 , abstract = on Pithdoi:10.1109/tpami.2018.2889473
  38. [58] Robust deep learning--based protein sequence design using ProteinMPNN , author=. Science , volume=. 2022 , publisher= on Pith
  39. [59] FoldX 5.0: working with RNA, small molecules and a new graphical interface , author=. Bioinformatics , volume=. 2019 , publisher= on Pith
  40. [60] Predicting mutational effects on protein-protein binding via a side-chain diffusion probabilistic model , author=. Advances in Neural Information Processing Systems , volume= on Pith
  41. [61] The PDB\_REDO server for macromolecular structure model optimization , author=. IUCrJ , volume=. 2014 , publisher= on Pith
  42. [62] Predicting changes in the stability of proteins and protein complexes: A study of more than 1000 mutations , author=. Journal of Molecular Biology , volume= on Pith
  43. [63] Computational redesign of protein--protein interaction specificity , author=. Nature Structural & Molecular Biology , volume=. 2004 , doi= on Pith
  44. [64] Energy landscape roughness of protein folding , author=. Proteins , volume= on Pith
  45. [65] Deep generative models of genetic variation capture the effects of mutations , author=. Nature Methods , volume=. 2018 , doi= on Pith
  46. [66] Biological structure and function emerge from scaling unsupervised learning to 250 million protein sequences , author=. Proceedings of the National Academy of Sciences , volume=. 2021 , doi= on Pith
  47. [67] Tranception: Protein Fitness Prediction with Autoregressive Transformers and Inference-time Retrieval , author=. Proceedings of the 39th International Conference on Machine Learning , series=. 2022 , publisher= on Pith
  48. [68] Deep mutational scanning: a new style of protein science , author=. Nature Methods , volume=. 2014 , doi= on Pith
  49. [69] Deep mutational scanning of SARS-CoV-2 receptor binding domain reveals constraints on folding and ACE2 binding , author=. Cell , volume=. 2020 , doi= on Pith
  50. [70] MaveDB: an open-source platform to distribute and interpret data from multiplexed assays of variant effect , author=. Genome Biology , volume=. 2019 , doi= on Pith
  51. [71] ProteinGym: Large-Scale Benchmarks for Protein Fitness Prediction and Design , author=. Proceedings of the 37th Conference on Neural Information Processing Systems , year= on Pith
  52. [72] Low-N protein engineering with data-efficient deep learning , author=. Nature Methods , volume=. 2021 , doi= on Pith
  53. [73] Direct Preference Optimization: Your Language Model is Secretly a Reward Model , author=. Advances in Neural Information Processing Systems , year= on Pith
  54. [74] Role of conformational sampling in computing mutation-induced changes in protein structure and stability , author=. Proteins: Structure, Function, and Bioinformatics , volume=. 2011 , publisher= on Pith
  55. [75] FoldX force field revisited, an improved version , author=. Bioinformatics , volume=. 2025 , publisher= on Pith
  56. [76] GEMME: a simple and fast global epistatic model predicting mutational effects , author=. Molecular biology and evolution , volume=. 2019 , publisher= on Pith
  57. [77] Assessing computational methods for predicting protein stability upon mutation: good on average but not in the details , author=. Protein engineering, design & selection , volume=. 2009 , publisher= on Pith
  58. [78] Fast and accurate predictions of protein stability changes upon mutations using statistical potentials and neural networks: PoPMuSiC-2.0 , author=. Bioinformatics , volume=. 2009 , publisher= on Pith
  59. [79] Stability effects of mutations and protein evolvability , author=. Current opinion in structural biology , volume=. 2009 , publisher= on Pith
  60. [80] Protein language models learn evolutionary statistics of interacting sequence motifs , author=. Proceedings of the National Academy of Sciences , volume=. 2024 , publisher= on Pith
  61. [81] Distilling the knowledge in a neural network , author=. arXiv preprint arXiv:1503.02531 , year= on PithPith's reviewarXiv
  62. [82] Learning from multiple teacher networks , author=. Proceedings of the 23rd ACM SIGKDD international conference on knowledge discovery and data mining , pages= on Pith
  63. [83] Snorkel: Rapid training data creation with weak supervision , author=. Proceedings of the VLDB endowment. International conference on very large data bases , volume= on Pith
  64. [84] Maximum likelihood estimation of observer error-rates using the EM algorithm , author=. Journal of the Royal Statistical Society: Series C (Applied Statistics) , volume=. 1979 , publisher= on Pith
  65. [85] Co-teaching: Robust training of deep neural networks with extremely noisy labels , author=. Advances in neural information processing systems , volume= on Pith
  66. [86] Deep mutual learning , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages= on Pith
  67. [87] Deep learning from crowds , author=. Proceedings of the AAAI conference on artificial intelligence , volume= on Pith
  68. [88] Marquet, Céline and Schlensok, Julius and Abakarova, Marina and Rost, Burkhard and Laine, Elodie , date =. Expert-guided protein language models enable accurate and blazingly fast fitness prediction , volume =. doi:10.1093/bioinformatics/btae621 , abstract = on Pithdoi:10.1093/bioinformatics/btae621
  69. [89] Guo, Ruihan and Wang, Rui and Wu, Ruidong and Ren, Zhizhou and Li, Jiahan and Luo, Shitong and Wu, Zuofan and Liu, Qiang and Peng, Jian and Ma, Jianzhu , booktitle =. Enhancing Protein Mutation Effect Prediction through a Retrieval-Augmented Framework , url =. doi:10.52202/079... on Pithdoi:10.52202/079017-1556
  70. [90] ProteinZero: Self-Improving Protein Generation via Online Reinforcement Learning , author=. 2026 , eprint= on Pith
  71. [91] Full-Atom Peptide Design based on Multi-modal Flow Matching , author =. Proceedings of the 41st International Conference on Machine Learning , pages =. 2024 , editor = on Pith

Challenges and discussion

Disagree with the review, or know something it missed? Leave a comment, or challenge this review formally: a challenge becomes a permanent entry in this paper's record, and its decision is published here.

No comments yet.

Sign in to comment