Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

Multi-Scale Representation Learning for Protein Fitness Prediction

T0 review · 4 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read The paper claims that integrating sequence, structure, and surface representations yields the best zero-shot protein fitness predictions on ProteinGym.

desk verdict Solid benchmark work showing surface features add a modest gain, but the headline state-of-the-art claim rests on an EVE ensemble, not on the multi-scale architecture alone. read the letter →

arxiv 2412.01108 v1 pith:UAVETA74 submitted 2024-12-02 cs.LG q-bio.BM

Zuobai Zhang , Pascal Notin , Yining Huang , Aurélie Lozano , Vijil Chenthamarakshan , Debora Marks , Payel Das , Jian Tang This is my paper · ORCID
classification cs.LGq-bio.BM
keywords proteinfitnesspredictionzero-shotmutationeffectmulti-scalerepresentationlearningsurfacegeometricvectorperceptronGymdeepmutationalscanningCATH
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper claims that protein fitness prediction improves when a model is trained across three scales at once: amino-acid sequence, backbone structure, and molecular surface. On 217 deep mutational scanning assays from ProteinGym, the full model (S3F-MSA) reaches a Spearman correlation of 0.496, beating the previous best sequence-structure model by 8.5 percent; even without alignment input, the surface-augmented model (S3F) scores 0.470. A sympathetic reader would care because mutation-effect prediction is a bottleneck in protein engineering and clinical variant interpretation, and the paper argues that surface geometry carries signal that sequence and backbone models miss. The method keeps the protein language model frozen and trains only a small structure-surface encoder on 31,000 CATH structures, so the claimed gains come with far fewer trainable parameters than leading baselines.

What carries the argument

The engine is a two-graph geometric message-passing decoder built on top of frozen ESM-2-650M embeddings. A Geometric Vector Perceptron (GVP), a message-passing layer that operates on rotation-equivariant scalar and vector features, updates residue features over a 10-angstrom radius graph on alpha carbons, and a second GVP updates surface-point features over a 16-neighbor point cloud generated by dMaSIF. Each surface point carries Gaussian curvature and Heat Kernel Signature features, and residue and surface representations are pooled back and forth using three nearest residues per surface point and twenty nearest surface points per residue. The whole stack is pre-trained with BERT-style masked residue prediction on CATH, mutation scores are computed as the log-odds ratio between mutant and wild-type residues at masked positions, and a hybrid inference rule uses ESM-2's score whenever AlphaFold2 confidence (pLDDT) is below 70 and the multi-scale score otherwise.

What would settle it

Recompute the 217-assay average with the low-confidence fallback threshold removed or varied across a grid; if no pre-specified threshold reproduces a Spearman of 0.496 for S3F-MSA, or if the ranking against SaProt changes with the threshold, the central claim collapses.

Watch

Extended reading notes

Core claim

The central discovery claimed is that multi-scale representations, namely sequence tokens, wild-type backbone coordinates, and point-cloud surface geometry, combine additively for zero-shot fitness prediction. S2F, the sequence-plus-structure model, roughly matches the best structure-aware baseline, while S3F, which adds the surface encoder, is the best model that does not use MSA input. With alignment-based EVE scores ensembled, S3F-MSA outperforms all 17 baselines on ProteinGym, and the ablation shows that removing the surface encoder drops Spearman from 0.470 to 0.454, while removing the structure encoder drops it to 0.392.

Load-bearing premise

The load-bearing premise is that the rule to fall back to the sequence-only model for low-confidence AlphaFold2 regions was fixed before the authors looked at ProteinGym results, so the reported zero-shot comparison is a fair one.

Editorial extensions

If this is right

  • Surface augmentation yields the largest gains on binding and stability assays, where structure-related mutations matter most.
  • The model's advantage grows with mutation depth, which the authors interpret as better capture of epistatic effects.
  • Because ESM-2 is frozen and only the GVP layers are trained (20M parameters for S3F), the same scaffolding can be re-attached to any future protein language model without retraining the language model.
  • The 8.5% improvement over SaProt in Spearman is statistically significant under ProteinGym's bootstrap difference test, with a standard error of 0.005 for S3F and 0.000 for S3F-MSA.

Reading between the lines

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

  • Editorial inference: an implication the authors leave implicit is that the surface encoder's benefit should transfer to other zero-shot tasks beyond substitution fitness, such as binding-site or interface prediction, since surface geometry is the same modality.
  • Editorial inference: a testable extension is to replace the frozen ESM-2 with a newer language model and measure whether the multi-scale gains persist; the paper's modular design predicts they would.
  • Editorial inference: the low-confidence fallback rule (pLDDT below 70) could be stress-tested by sweeping the threshold; if performance is sensitive to it, the reported zero-shot comparison is less robust than it appears.
  • Editorial inference: another extension is to adapt the method to insertions and deletions by allowing the backbone graph to change at the mutated sites, which the paper explicitly leaves to future work.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper introduces S2F and S3F, multi-scale protein encoders that combine ESM-2 language-model embeddings with GVP message passing over C-alpha backbone graphs and, for S3F, dMaSIF surface point clouds. The models are pre-trained on CATH with a masked-residue prediction objective and evaluated zero-shot on the ProteinGym benchmark's 217 substitution assays. S2F reaches 0.454 average Spearman correlation, S3F reaches 0.470, and ensembling with EVE yields S2F-MSA at 0.487 and S3F-MSA at 0.496, which the paper claims is state of the art and an 8.5% improvement over SaProt's 0.457. The paper also presents breakdown analyses by assay type, MSA depth, taxon, mutation depth, and structure quality, plus an out-of-distribution generalization experiment.

Significance. If verified, the central contribution is meaningful: S3F improves on ESM-2 by 0.056 Spearman with only 20M trainable parameters, and the OOD generalization check supports the transferability of the learned structure and surface representations. Using the established ProteinGym benchmark and bootstrap-based comparisons is a strength, and the multi-scale framework is modular enough to be adapted to future protein language models. However, the headline 8.5% gain is obtained by ensembling with EVE, and no equivalent ensemble baselines are reported for the comparison models; the pLDDT fallback rule in Section 3.5 is not justified as a fixed a priori design choice; and the code, data, and pretrained weights are not available at submission. The evidence for the specific contribution of surface topology is also modest, as the surface-only ablation in Table B.3 underperforms ESM-2.

major comments (4)
  1. [Section 4.1 and 4.2 (Table 1)] The headline comparison "S3F-MSA improves SaProt by 8.5%" is not an apples-to-apples test of the proposed architecture. S3F-MSA is defined as the z-score sum of S3F and EVE, while SaProt is a single model with no equivalent ensemble. No ESM-2+EVE, S2F+EVE, or SaProt+EVE baseline is reported, so the 8.5% gain conflates ensemble gains with gains from the multi-scale architecture. Please add the missing ensemble baselines and reframe the state-of-the-art claim to compare within categories, e.g., S3F versus SaProt in the no-MSA category and S3F-MSA versus TranceptEVE in the MSA category.
  2. [Section 3.5] The pLDDT-based fallback rule ("for mutations on residues with pLDDT score below 70, we use the output scores from ESM-2-650M") is load-bearing for the paper's zero-shot claim, but the threshold is not derived or ablated. If the threshold was chosen after inspecting ProteinGym results, the comparison is not strictly zero-shot. State how the threshold was fixed, report sensitivity to the threshold (e.g., 50/60/70/80) on a held-out validation set, and clarify whether the rule is applied per mutated residue or per assay.
  3. [Table 2 and Section B.2] The reported standard error of 0.000 for SaProt's difference to the best score is implausible and likely a typo. More importantly, Table 2 only reports each baseline's difference to S3F-MSA, not pairwise bootstrap confidence intervals between the models being compared. The sentence "Our performance delta with prior methods are all statistically significant" therefore needs pairwise tests, or at least per-model bootstrap errors, to be supported.
  4. [Checklist items 5 and 7] The paper states that code, data, pretrained weights, and statistical significance results "will be released upon acceptance." Because the central empirical claims depend on trained GVP weights and on the exact AlphaFold2-structure handling, the current submission is not reproducible as written. Please provide the code and weights, or at minimum per-assay prediction tables and a complete specification of hyperparameters, surface point sampling, and pLDDT handling.
minor comments (5)
  1. [Section 4.5 and Figure 4] The text says 23 out-of-distribution assays are selected, but Figure 4 lists 19 proteins; clarify whether multiple assays share the same protein and how grouping was performed.
  2. [Section 3.4] The surface point count is stated as 6K-20K points; specify how this depends on protein size and whether it is fixed during inference.
  3. [Figure 2(f)] The x-axis labels 0 through 4 are described as groups of AlphaFold2-predicted structures ranked by pLDDT, but the ranking procedure is not defined; please state how the five structures are ordered and how ties or missing pLDDT values are handled.
  4. [Section B.3] The ablation shows that removing structure while keeping surface alone drops performance below ESM-2 (0.392 vs 0.414); this negative result is relevant to the claim about surface features and should be discussed in the main text, not only in the appendix.
  5. [Section 4.2] Calling SaProt "the current state-of-the-art method" in the context of the MSA-augmented S3F-MSA comparison is confusing, because SaProt appears in the no-MSA category of Table 1; reword to avoid mixing categories.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: self-supervised CATH pretraining and zero-shot ProteinGym evaluation are structurally separated; self-citations are non-load-bearing.

full rationale

The derivation chain is: (i) freeze ESM-2 and train GVP structure/surface encoders to predict masked residue types on CATH structures (Section 3.5); (ii) at inference score mutants by masked log-odds ratio of the learned conditional distribution (Eq. 1), using AlphaFold2 wild-type structures and the pLDDT-based hybrid rule; (iii) optionally sum z-scores with the pre-existing EVE model (Section 4.1). No ProteinGym fitness label enters pretraining or ensembling, so the zero-shot fitness scores are not fitted to the evaluation target. The pLDDT < 70 fallback to ESM-2 is a stated heuristic, not a parameter fitted to ProteinGym performance, and no text shows it was selected on the benchmark. The S3F-MSA 8.5% improvement is defined as an ensemble with EVE rather than as the multi-scale architecture alone; this is a comparison-fairness concern about attributing gains, not an equation-level circularity. Self-citations (ProteinGym [31], Tranception [14], TranceptEVE [15], Paul et al. [24]) provide benchmark definitions, baselines, or a consistency remark; none is used to justify the model's central premise or to rule out alternatives. The ablation (Table 3) is honest, showing surface-only underperforms ESM2, which further confirms the components are being tested rather than defined into existence.

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

The central claim depends on no fitted constants in the usual sense, but on several modeling choices: a hand-picked pLDDT fallback threshold, an unweighted EVE ensemble for the MSA variant, fixed-backbone and side-chain-free structure representation, and the assumption that CATH masked-residue pretraining transfers to fitness. All are disclosed in the text, but the pLDDT threshold's provenance is not.

free parameters (2)
  • pLDDT fallback threshold = 70
    Section 3.5: above pLDDT 70 use S2F or S3F scores, below use ESM-2. This hand-chosen scalar directly determines which model scores each residue, and no validation or a priori justification is provided.
  • z-score ensemble weights for S3F-MSA = 1.0 (equal sum of S3F and EVE z-scores)
    Section 4.1: the MSA variants are formed by summation of z-scores. Equal weighting is a design choice that drives the headline 8.5% improvement over SaProt.
assumptions (4)
  • domain assumption Protein backbones remain structurally unchanged after substitution (Xmt = Xwt) and side chains can be ignored (Section 3.1).
    Mutation effect scores are computed on wild-type AlphaFold2 structures, so any fitness signal carried by mutation-induced structural or side-chain rearrangements is inaccessible; the paper states this explicitly as a simplification.
  • domain assumption Fitness of multiple mutations is additive in per-position log-likelihood ratios (Eq. 1, Section 3.2).
    Zero-shot scoring sums independent wild-type and mutant log-odds at mutation sites; epistasis is only captured if the model's context features happen to encode it, not by the scoring rule.
  • domain assumption Masked residue prediction on CATH transfers to fitness-relevant signal (Section 3.5).
    The GVP structure and surface encoders are trained only to recover masked residue types; no fitness labels are used, so the benchmark gain assumes this self-supervised objective is a good proxy for mutation effects.
  • domain assumption AlphaFold2-predicted structures and dMaSIF surface point clouds with curvature and Heat Kernel Signature features provide sufficiently accurate surface topology for fitness prediction (Sections 3.4 and 4.4).
    All structures are predicted rather than experimental, and the surface encoder relies on these predicted geometries; the paper shows performance drops with lower-pLDDT structures.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Multi-Scale Representation Learning for Protein Fitness Prediction." pith.science (2026). https://pith.science/paper/UAVETA74

@misc{pith2026241201108,
  author       = {Pith},
  title        = {Pith review of: Multi-Scale Representation Learning for Protein Fitness Prediction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UAVETA74}},
  note         = {Machine review of arXiv:2412.01108}
}
read the original abstract

Designing novel functional proteins crucially depends on accurately modeling their fitness landscape. Given the limited availability of functional annotations from wet-lab experiments, previous methods have primarily relied on self-supervised models trained on vast, unlabeled protein sequence or structure datasets. While initial protein representation learning studies solely focused on either sequence or structural features, recent hybrid architectures have sought to merge these modalities to harness their respective strengths. However, these sequence-structure models have so far achieved only incremental improvements when compared to the leading sequence-only approaches, highlighting unresolved challenges effectively leveraging these modalities together. Moreover, the function of certain proteins is highly dependent on the granular aspects of their surface topology, which have been overlooked by prior models. To address these limitations, we introduce the Sequence-Structure-Surface Fitness (S3F) model - a novel multimodal representation learning framework that integrates protein features across several scales. Our approach combines sequence representations from a protein language model with Geometric Vector Perceptron networks encoding protein backbone and detailed surface topology. The proposed method achieves state-of-the-art fitness prediction on the ProteinGym benchmark encompassing 217 substitution deep mutational scanning assays, and provides insights into the determinants of protein function. Our code is at https://github.com/DeepGraphLearning/S3F.

Figures

Figures reproduced from arXiv: 2412.01108 by the authors.

Figure 1
Figure 1. Multi-scale Pre-training and Inference Frameworks for Protein Fitness Prediction. During pre-training, protein sequences and structures are sampled from a database, with 15% of residue types randomly masked. These sequences are fed into a protein language model, ESM-2- 650M. Then, the output residue representations are used to initialize node features in our structure and surface encoders. Through message passing on… view at source ↗
Figure 2
Figure 2. Results of ESM-2-650M, S2F, S3F, and S3F-MSA for Analyzing Contributions of Sequences, Structures, Surfaces, and Alignments. (a-d) Breakdown performance (Spearman’s rank correlation) on assays grouped by function type (a), MSA depth (b), taxon (c), and mutation depth (d). (e-f) Impact of protein structure quality on performance. (e) Breakdown performance on assays with low, medium, and high-quality structures. (f) R… view at source ↗
Figure 3
Figure 3. Case Study on GB1. (a-c) For each pair of mutation sites, we plot the Spearman’s rank correlation between the experimental values and model-predicted scores for all 361 mutations: ESM (a), S2F (b), and S3F (c). The epistasis between residues 234-252 and residues 266-282 (in the black rectangle) are better captured by S2F and S3F. (d) Visualization of the predicted structure for GB1. Mutation regions 234-252 and 266-… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Spearmanr’s rank correlation for ESM-2-650M, S2F, S3F and S3F-MSA on 19 proteins [PITH_FULL_IMAGE:figures/full_fig_p015_4.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Steering Protein Language Models

    q-bio.BM 2025-07 reject novelty 6.0 of 10

    Activation steering can guide protein language models to generate and optimize sequences with higher predicted thermostability, solubility, or GFP brightness, but only in surrogate-based evaluation.

Reference graph

Works this paper leans on

75 extracted references · 60 canonical work pages · cited by 1 Pith paper

  1. [44]

    Pre-training sequence, structure, and surface features for comprehensive protein representation learning

    Youhan Lee, Hasun Yu, Jaemyung Lee, and Jaehoon Kim. Pre-training sequence, structure, and surface features for comprehensive protein representation learning. In The Twelfth International Conference on Learning Representations, 2024

    work page 2024

  2. [1]

    Romero and Frances H

    Philip A. Romero and Frances H. Arnold. Exploring protein fitness landscapes by directed evolution. Nature Reviews Molecular Cell Biology, 10:866–876, 2009

    work page 2009

  3. [2]

    Rollins, Yarin Gal, Chris Sander, and Debora Marks

    Pascal Notin, Nathan J. Rollins, Yarin Gal, Chris Sander, and Debora Marks. Machine learning for functional protein design. Nature Biotechnology, 42:216–228, 2024

    work page 2024

  4. [3]

    Low-n protein engineering with data-efficient deep learning

    Surojit Biswas, Grigory Khimulya, Ethan C Alley, Kevin M Esvelt, and George M Church. Low-n protein engineering with data-efficient deep learning. Nature methods, 18(4):389–396, 2021

    2021

  5. [4]

    Mutation effects predicted from sequence co-variation

    Thomas A Hopf, John B Ingraham, Frank J Poelwijk, Charlotta PI Schärfe, Michael Springer, Chris Sander, and Debora S Marks. Mutation effects predicted from sequence co-variation. Nature biotechnology, 35(2):128–135, 2017

    work page 2017

  6. [5]

    Language models enable zero-shot prediction of the effects of mutations on protein function

    Joshua Meier, Roshan Rao, Robert Verkuil, Jason Liu, Tom Sercu, and Alex Rives. Language models enable zero-shot prediction of the effects of mutations on protein function. Advances in neural information processing systems, 34:29287–29303, 2021

    work page 2021

  7. [6]

    Deep generative models of genetic variation capture the effects of mutations

    Adam J Riesselman, John B Ingraham, and Debora S Marks. Deep generative models of genetic variation capture the effects of mutations. Nature methods, 15(10):816–822, 2018

    2018

  8. [7]

    Disease variant prediction with deep generative models of evolutionary data

    Jonathan Frazer, Pascal Notin, Mafalda Dias, Aidan Gomez, Joseph K Min, Kelly Brock, Yarin Gal, and Debora S Marks. Disease variant prediction with deep generative models of evolutionary data. Nature, 599(7883):91–95, 2021

    2021

Show all 75 references
  1. [8]

    Gemme: a simple and fast global epistatic model predicting mutational effects

    Elodie Laine, Yasaman Karami, and Alessandra Carbone. Gemme: a simple and fast global epistatic model predicting mutational effects. Molecular biology and evolution, 36(11):2604– 2619, 2019

    2019

  2. [9]

    Protein design and variant prediction using autoregressive generative models

    Jung-Eun Shin, Adam J Riesselman, Aaron W Kollasch, Conor McMahon, Elana Simon, Chris Sander, Aashish Manglik, Andrew C Kruse, and Debora S Marks. Protein design and variant prediction using autoregressive generative models. Nature communications, 12(1):2403, 2021. 10

    2021

  3. [10]

    Unified rational protein engineering with sequence-based deep representation learning

    Ethan C Alley, Grigory Khimulya, Surojit Biswas, Mohammed AlQuraishi, and George M Church. Unified rational protein engineering with sequence-based deep representation learning. Nature methods, 16(12):1315–1322, 2019

    2019

  4. [11]

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

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

    2021

  5. [12]

    Eguchi, Po-Ssu Huang, and Richard Socher

    Ali Madani, Bryan McCann, Nikhil Vijay Naik, Nitish Shirish Keskar, Namrata Anand, Raphael R. Eguchi, Po-Ssu Huang, and Richard Socher. Progen: Language modeling for protein generation. bioRxiv, 2020

    2020

  6. [13]

    Msa transformer

    Roshan M Rao, Jason Liu, Robert Verkuil, Joshua Meier, John Canny, Pieter Abbeel, Tom Sercu, and Alexander Rives. Msa transformer. In Proceedings of the 38th International Conference on Machine Learning , volume 139 of Proceedings of Machine Learning Research , pages 8844–8856...

    2021

  7. [14]

    Tranception: protein fitness prediction with autoregressive transformers and inference-time retrieval

    Pascal Notin, Mafalda Dias, Jonathan Frazer, Javier Marchena Hurtado, Aidan N Gomez, Debora Marks, and Yarin Gal. Tranception: protein fitness prediction with autoregressive transformers and inference-time retrieval. In International Conference on Machine Learning, pages 16990...

    2022

  8. [15]

    Trancepteve: Combining family-specific and family-agnostic models of protein sequences for improved fitness prediction

    Pascal Notin, Lood Van Niekerk, Aaron W Kollasch, Daniel Ritter, Yarin Gal, and Debora S Marks. Trancepteve: Combining family-specific and family-agnostic models of protein sequences for improved fitness prediction. bioRxiv, pages 2022–12, 2022

    2022

  9. [16]

    Marks, Chris Sander, Riccardo Zecchina, José Nelson Onuchic, Terence Hwa, and Martin Weigt

    Faruck Morcos, Andrea Pagnani, Bryan Lunt, Arianna Bertolino, Debora S. Marks, Chris Sander, Riccardo Zecchina, José Nelson Onuchic, Terence Hwa, and Martin Weigt. Direct- coupling analysis of residue coevolution captures native contacts across many protein families. Proceedin...

    2011

  10. [17]

    Garg, Regina Barzilay, and T

    John Ingraham, Vikas K. Garg, Regina Barzilay, and T. Jaakkola. Generative models for graph-based protein design. In DGS@ICLR, 2019

    2019

  11. [18]

    Polarity as a criterion in protein design

    G Baumann, Cornelius Frömmel, and Chris Sander. Polarity as a criterion in protein design. Protein engineering, 2 5:329–34, 1989

    1989

  12. [19]

    Soni, and Ron O

    Bowen Jing, Stephan Eismann, Pratham N. Soni, and Ron O. Dror. Learning from protein structure with geometric vector perceptrons. In International Conference on Learning Representations, 2021

    2021

  13. [20]

    Protein representation learning by geometric structure pretraining

    Zuobai Zhang, Minghao Xu, Arian Jamasb, Vijil Chenthamarakshan, Aurelie Lozano, Payel Das, and Jian Tang. Protein representation learning by geometric structure pretraining. In The Eleventh International Conference on Learning Representations, 2023

    2023

  14. [21]

    Learning inverse folding from millions of predicted structures

    Chloe Hsu, Robert Verkuil, Jason Liu, Zeming Lin, Brian Hie, Tom Sercu, Adam Lerer, and Alexander Rives. Learning inverse folding from millions of predicted structures. InInternational conference on machine learning, pages 8946–8970. PMLR, 2022

    2022

  15. [22]

    Robust deep learning–based protein sequence design using proteinmpnn

    Justas Dauparas, Ivan Anishchenko, Nathaniel Bennett, Hua Bai, Robert J Ragotte, Lukas F Milles, Basile IM Wicky, Alexis Courbet, Rob J de Haas, Neville Bethel, et al. Robust deep learning–based protein sequence design using proteinmpnn. Science, 378(6615):49–56, 2022

    2022

  16. [23]

    Masked inverse folding with sequence transfer for protein representation learning

    Kevin K Yang, Niccolò Zanichelli, and Hugh Yeh. Masked inverse folding with sequence transfer for protein representation learning. Protein Engineering, Design and Selection, 36:gzad015, 2023

    2023

  17. [24]

    Combining structure and sequence for superior fitness prediction

    Steffanie Paul, Aaron Kollasch, Pascal Notin, and Debora Marks. Combining structure and sequence for superior fitness prediction. In NeurIPS 2023 Generative AI and Biology (GenBio) Workshop, 2023

    2023

  18. [25]

    Predicting absolute protein folding stability using generative models

    Matteo Cagiada, Sergey Ovchinnikov, and Kresten Lindorff-Larsen. Predicting absolute protein folding stability using generative models. bioRxiv, 2024. 11

    2024

  19. [26]

    Saprot: Protein language modeling with structure-aware vocabulary

    Jin Su, Chenchen Han, Yuyang Zhou, Junjie Shan, Xibin Zhou, and Fajie Yuan. Saprot: Protein language modeling with structure-aware vocabulary. In The Twelfth International Conference on Learning Representations, 2024

    2024

  20. [27]

    Semantical and topological protein encoding toward enhanced bioactivity and thermostability

    Yang Tan, Bingxin Zhou, Lirong Zheng, Guisheng Fan, and Liang Hong. Semantical and topological protein encoding toward enhanced bioactivity and thermostability. bioRxiv, pages 2023–12, 2023

    2023

  21. [28]

    Accurate proteome- wide missense variant effect prediction with alphamissense

    Jun Cheng, Guido Novati, Joshua Pan, Clare Bycroft, Akvil˙e Žemgulyt˙e, Taylor Applebaum, Alexander Pritzel, Lai Hong Wong, Michal Zielinski, Tobias Sargeant, et al. Accurate proteome- wide missense variant effect prediction with alphamissense. Science, 381(6664):eadg7492, 2023

    2023

  22. [29]

    Deciphering interaction fingerprints from protein molecular surfaces using geometric deep learning

    Pablo Gainza, Freyr Sverrisson, Frederico Monti, Emanuele Rodola, D Boscaini, Michael M Bronstein, and Bruno E Correia. Deciphering interaction fingerprints from protein molecular surfaces using geometric deep learning. Nature Methods, 17(2):184–192, 2020

    2020

  23. [30]

    Cath: an expanded resource to predict protein function through structure and sequence

    Natalie L Dawson, Tony E Lewis, Sayoni Das, Jonathan G Lees, David Lee, Paul Ashford, Christine A Orengo, and Ian Sillitoe. Cath: an expanded resource to predict protein function through structure and sequence. Nucleic acids research, 45(D1):D289–D295, 2017

    2017

  24. [31]

    Proteingym: large- scale benchmarks for protein fitness prediction and design

    Pascal Notin, Aaron Kollasch, Daniel Ritter, Lood Van Niekerk, Steffanie Paul, Han Spinner, Nathan Rollins, Ada Shaw, Rose Orenbuch, Ruben Weitzman, et al. Proteingym: large- scale benchmarks for protein fitness prediction and design. Advances in Neural Information Processing ...

    2023

  25. [32]

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

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

    2023

  26. [33]

    Multi-scale representation learning on proteins

    Vignesh Ram Somnath, Charlotte Bunne, and Andreas Krause. Multi-scale representation learning on proteins. Advances in Neural Information Processing Systems, 34:25244–25255, 2021

    2021

  27. [34]

    Evaluating protein transfer learning with tape

    Roshan Rao, Nicholas Bhattacharya, Neil Thomas, Yan Duan, Xi Chen, John Canny, Pieter Abbeel, and Yun S Song. Evaluating protein transfer learning with tape. In Advances in Neural Information Processing Systems, 2019

    2019

  28. [35]

    Prottrans: Towards cracking the language of lifes code through self-supervised deep learning and high performance computing

    Ahmed Elnaggar, Michael Heinzinger, Christian Dallago, Ghalia Rehawi, Wang Yu, Llion Jones, Tom Gibbs, Tamas Feher, Christoph Angerer, Martin Steinegger, Debsindhu Bhowmik, and Burkhard Rost. Prottrans: Towards cracking the language of lifes code through self-supervised deep l...

    2021

  29. [36]

    Structure-based protein function prediction using graph convolutional networks

    Vladimir Gligorijevi´c, P Douglas Renfrew, Tomasz Kosciolek, Julia Koehler Leman, Daniel Berenberg, Tommi Vatanen, Chris Chandler, Bryn C Taylor, Ian M Fisk, Hera Vlamakis, et al. Structure-based protein function prediction using graph convolutional networks. Nature communicat...

    2021

  30. [37]

    Intrinsic-extrinsic convolution and pooling for learning on 3d protein structures

    Pedro Hermosilla, Marco Schäfer, Matˇej Lang, Gloria Fackelmann, Pere Pau Vázquez, Barbora Kozlíková, Michael Krone, Tobias Ritschel, and Timo Ropinski. Intrinsic-extrinsic convolution and pooling for learning on 3d protein structures. International Conference on Learning Repr...

    2021

  31. [38]

    Structure-aware protein self-supervised learning

    Can (Sam) Chen, Jingbo Zhou, Fan Wang, Xue Liu, and Dejing Dou. Structure-aware protein self-supervised learning. Bioinformatics, 39, 2022

    2022

  32. [39]

    Pre-training protein encoder via siamese sequence-structure diffusion trajectory prediction

    Zuobai Zhang, Minghao Xu, Aurelie Lozano, Vijil Chenthamarakshan, Payel Das, and Jian Tang. Pre-training protein encoder via siamese sequence-structure diffusion trajectory prediction. In Thirty-seventh Conference on Neural Information Processing Systems, 2023

    2023

  33. [40]

    Fast end-to-end learning on protein surfaces

    Freyr Sverrisson, Jean Feydy, Bruno E Correia, and Michael M Bronstein. Fast end-to-end learning on protein surfaces. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 15272–15281, 2021. 12

    2021

  34. [41]

    Atomsurf: Surface representation for learning on protein structures

    Vincent Mallet, Souhaib Attaiki, and Maks Ovsjanikov. Atomsurf: Surface representation for learning on protein structures. arXiv preprint arXiv:2309.16519, 2023

    2023

  35. [42]

    Lm-gvp: an extensible sequence and structure informed deep learning framework for protein property prediction

    Zichen Wang, Steven A Combs, Ryan Brand, Miguel Romero Calvo, Panpan Xu, George Price, Nataliya Golovach, Emmanuel O Salawu, Colby J Wise, Sri Priya Ponnapalli, et al. Lm-gvp: an extensible sequence and structure informed deep learning framework for protein property prediction...

    2022

  36. [43]

    A systematic study of joint representation learning on protein sequences and structures

    Zuobai Zhang, Chuanrui Wang, Minghao Xu, Vijil Chenthamarakshan, Aurelie Lozano, Payel Das, and Jian Tang. A systematic study of joint representation learning on protein sequences and structures. arXiv preprint arXiv:2303.06275, 2023

    2023 arXiv

  37. [45]

    Integration of pre-trained protein language models into geometric deep learning networks

    Fang Wu, Lirong Wu, Dragomir Radev, Jinbo Xu, and Stan Z Li. Integration of pre-trained protein language models into geometric deep learning networks. Communications Biology, 6(1):876, 2023

    2023

  38. [46]

    Machine-learning-guided directed evolution for protein engineering

    Kevin K Yang, Zachary Wu, and Frances H Arnold. Machine-learning-guided directed evolution for protein engineering. Nature methods, 16(8):687–694, 2019

    2019

  39. [47]

    Neural networks to learn protein sequence–function relationships from deep mutational scanning data

    Sam Gelman, Sarah A Fahlberg, Pete Heinzelman, Philip A Romero, and Anthony Gitter. Neural networks to learn protein sequence–function relationships from deep mutational scanning data. Proceedings of the National Academy of Sciences, 118(48):e2104878118, 2021

    2021

  40. [48]

    Flip: Benchmark tasks in fitness landscape inference for proteins

    Christian Dallago, Jody Mou, Kadina E Johnston, Bruce J Wittmann, Nicholas Bhattacharya, Samuel Goldman, Ali Madani, and Kevin K Yang. Flip: Benchmark tasks in fitness landscape inference for proteins. bioRxiv, pages 2021–11, 2021

    2021

  41. [49]

    Proteinnpt: improving protein property prediction and design with non-parametric transformers

    Pascal Notin, Ruben Weitzman, Debora Marks, and Yarin Gal. Proteinnpt: improving protein property prediction and design with non-parametric transformers. Advances in Neural Information Processing Systems, 36, 2023

    2023

  42. [50]

    Progen2: exploring the boundaries of protein language models

    Erik Nijkamp, Jeffrey A Ruffolo, Eli N Weinstein, Nikhil Naik, and Ali Madani. Progen2: exploring the boundaries of protein language models. Cell systems, 14(11):968–978, 2023

    2023

  43. [51]

    Convolutions are competitive with transformers for protein sequence pretraining

    Kevin K Yang, Nicolo Fusi, and Alex X Lu. Convolutions are competitive with transformers for protein sequence pretraining. Cell Systems, 15(3):286–294, 2024

    2024

  44. [52]

    Poet: A generative model of protein families as sequences-of-sequences

    Timothy Truong Jr and Tristan Bepler. Poet: A generative model of protein families as sequences-of-sequences. Advances in Neural Information Processing Systems, 36, 2023

    2023

  45. [53]

    Fast and accurate protein structure search with foldseek

    Michel Van Kempen, Stephanie S Kim, Charlotte Tumescheit, Milot Mirdita, Jeongjae Lee, Cameron LM Gilchrist, Johannes Söding, and Martin Steinegger. Fast and accurate protein structure search with foldseek. Nature Biotechnology, 42(2):243–246, 2024

    2024

  46. [54]

    Deep mutational scanning: a new style of protein science

    Douglas M Fowler and Stanley Fields. Deep mutational scanning: a new style of protein science. Nature methods, 11(8):801–807, 2014

    2014

  47. [55]

    Gaussian processes for machine learning (gpml) toolbox

    Carl Edward Rasmussen and Hannes Nickisch. Gaussian processes for machine learning (gpml) toolbox. The Journal of Machine Learning Research, 11:3011–3015, 2010

    2010

  48. [56]

    Efficient curvature estimation for oriented point clouds

    Yueqi Cao, Didong Li, Huafei Sun, Amir H Assadi, and Shiqiang Zhang. Efficient curvature estimation for oriented point clouds. stat, 1050:26, 2019

    2019

  49. [57]

    A concise and provably informative multi- scale signature based on heat diffusion

    Jian Sun, Maks Ovsjanikov, and Leonidas Guibas. A concise and provably informative multi- scale signature based on heat diffusion. In Computer graphics forum , volume 28, pages 1383–1392. Wiley Online Library, 2009

    2009

  50. [58]

    Bert: Pre-training of deep bidirectional transformers for language understanding

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. Bert: Pre-training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805, 2018. 13

    2018 arXiv

  51. [59]

    Quantifying and understanding the fitness effects of protein mutations: Laboratory versus nature.Protein Science, 25(7):1219–1226, 2016

    Jeffrey I Boucher, Daniel NA Bolon, and Dan S Tawfik. Quantifying and understanding the fitness effects of protein mutations: Laboratory versus nature.Protein Science, 25(7):1219–1226, 2016

    2016

  52. [60]

    Anders Olson, Nicholas C

    C. Anders Olson, Nicholas C. Wu, and Ren Sun. A comprehensive biophysical description of pairwise epistasis throughout an entire protein domain. Current Biology, 24(22):2643–2651, November 2014. 14 Table 2: Average Spearman and Std. Error of Difference to Best Score. The best ...

    2014

  53. [61]

    Claims Question: Do the main claims made in the abstract and introduction accurately reflect the paper’s contributions and scope? Answer: [Yes] Justification: The main claims about the experimental results of our methods are supported in Section 4

  54. [62]

    Limitations Question: Does the paper discuss the limitations of the work performed by the authors? Answer: [Yes] Justification: The limitations of this paper are discussed in Section 5

  55. [63]

    Theory Assumptions and Proofs Question: For each theoretical result, does the paper provide the full set of assumptions and a complete (and correct) proof? Answer: [NA] Justification: This paper does not include theoretical results

  56. [64]

    Experimental Result Reproducibility Question: Does the paper fully disclose all the information needed to reproduce the main experimental results of the paper to the extent that it affects the main claims and/or conclusions of the paper (regardless of whether the code and data...

  57. [65]

    Open access to data and code Question: Does the paper provide open access to the data and code, with sufficient instructions to faithfully reproduce the main experimental results, as described in supplemental material? Answer: [No] Justification: The code, data and pre-trained...

  58. [66]

    Experimental Setting/Details Question: Does the paper specify all the training and test details (e.g., data splits, hyperparameters, how they were chosen, type of optimizer, etc.) necessary to understand the results? Answer: [Yes] Justification: The evaluation setup are provid...

  59. [67]

    Experiment Statistical Significance Question: Does the paper report error bars suitably and correctly defined or other appropriate information about the statistical significance of the experiments? Answer: [No] Justification: The statistical significance will be reported when ...

  60. [68]

    Experiments Compute Resources Question: For each experiment, does the paper provide sufficient information on the computer resources (type of compute workers, memory, time of execution) needed to reproduce the experiments? Answer: [Yes] Justification: The information about com...

  61. [69]

    Code Of Ethics 17 Question: Does the research conducted in the paper conform, in every respect, with the NeurIPS Code of Ethics https://neurips.cc/public/EthicsGuidelines? Answer: [Yes] Justification: The research conform with the NeurIPS Code of Ethics

  62. [70]

    Broader Impacts Question: Does the paper discuss both potential positive societal impacts and negative societal impacts of the work performed? Answer: [Yes] Justification: The broader impacts are discussed in Section A

  63. [71]

    Safeguards Question: Does the paper describe safeguards that have been put in place for responsible release of data or models that have a high risk for misuse (e.g., pretrained language models, image generators, or scraped datasets)? Answer: [NA] Justification: The paper poses...

  64. [72]

    Licenses for existing assets Question: Are the creators or original owners of assets (e.g., code, data, models), used in the paper, properly credited and are the license and terms of use explicitly mentioned and properly respected? Answer: [Yes] Justification: The license of u...

  65. [73]

    New Assets Question: Are new assets introduced in the paper well documented and is the documentation provided alongside the assets? Answer: [NA] Justification: The paper does not release new assets

  66. [74]

    Crowdsourcing and Research with Human Subjects Question: For crowdsourcing experiments and research with human subjects, does the paper include the full text of instructions given to participants and screenshots, if applicable, as well as details about compensation (if any)? A...

  67. [75]

    Institutional Review Board (IRB) Approvals or Equivalent for Research with Human Subjects Question: Does the paper describe potential risks incurred by study participants, whether such risks were disclosed to the subjects, and whether Institutional Review Board (IRB) approvals...

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.