pith. sign in

arxiv: 2606.11243 · v1 · pith:3B62U5OBnew · submitted 2026-06-03 · 💻 cs.LG · cs.CL

ProHiFlo: Hierarchical Flow Matching with Functional Guidance for De Novo Protein Generation

Chuanzhen Wang , Meade Cleti , Pete Jano This is my paper

Pith reviewed 2026-06-28 07:09 UTC · model grok-4.3

classification 💻 cs.LG cs.CL
keywords protein generationflow matchinghierarchical modelingfunctional guidancede novo designenzyme scaffoldingSE(3)-equivariantmotif scaffolding
0
0 comments X

The pith

ProHiFlo generates de novo proteins via hierarchical flow matching and functional guidance from pretrained predictors, achieving higher success rates with four fewer sampling steps.

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

The paper presents ProHiFlo as a framework that generates protein structures by first modeling coarse backbone geometry and then refining to all-atom coordinates. It adds functional guidance that uses existing predictors to direct outputs toward desired properties without any retraining of the main model. An adaptive SE(3)-equivariant architecture supports processing at multiple scales. Tests across unconditional generation, motif scaffolding, and functional design show improved results compared with prior methods while cutting sampling steps by four. Enzyme active site scaffolding reaches 58.9 percent success versus 41.2 percent for RFDiffusion.

Core claim

ProHiFlo shows that a hierarchical flow matching process combined with functional guidance from pretrained predictors produces more accurate and efficient de novo proteins than single-resolution diffusion or flow models, delivering state-of-the-art performance on multiple design tasks with substantially reduced sampling steps.

What carries the argument

The hierarchical flow matching framework with functional guidance, which generates structures coarse-to-fine and steers them using external pretrained predictors.

If this is right

  • Protein design pipelines can run with four fewer sampling steps while maintaining or improving output quality.
  • Functional constraints can be added to existing flow matching models without full retraining.
  • Multi-scale modeling separates backbone geometry from all-atom refinement to lower computational cost.
  • Enzyme active site scaffolding success rises from 41.2 percent to 58.9 percent under the new approach.

Where Pith is reading between the lines

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

  • The same coarse-to-fine structure could be tested on other sequence-to-structure tasks such as antibody design.
  • Predictor-based guidance might reduce the need for task-specific fine-tuning across generative biology models.
  • Integration with downstream experimental assays could create closed-loop design systems that update guidance signals iteratively.

Load-bearing premise

Functional guidance from pretrained predictors can steer generation toward desired properties without retraining the core model.

What would settle it

A controlled test in which functional guidance is removed and success rates on enzyme active site scaffolding fall to or below the 41.2 percent baseline of prior methods.

Figures

Figures reproduced from arXiv: 2606.11243 by Chuanzhen Wang, Meade Cleti, Pete Jano.

Figure 1
Figure 1. Figure 1: Motivation for ProHiFlo. Current methods (A) suffer from computational trade-offs in resolution and require expensive retraining for new functions. ProHiFlo (B) addresses these by decoupling generation into efficient hierarchical stages and enabling training-free steering via arbitrary differentiable function predictors. to steer generation toward desired properties without model retraining. Third, we desi… view at source ↗
Figure 2
Figure 2. Figure 2: In the first stage, we generate the coarse backbone structure represented as residue frames. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Training convergence for both stages. The backbone stage (left) converges faster due to the [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Multi-dimensional performance comparison across six key metrics. ProHiFlo (blue) [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Effect of guidance strength λ on stability and designability. There exists an optimal guidance strength (λ ≈ 0.5) that maximizes stability while maintaining high designability. Stronger guidance can degrade structure quality [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Inference time comparison across different protein lengths. ProHiFlo maintains near-linear [PITH_FULL_IMAGE:figures/full_fig_p011_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Designability across different protein lengths. ProHiFlo maintains higher designability [PITH_FULL_IMAGE:figures/full_fig_p012_7.png] view at source ↗
read the original abstract

De novo protein generation has transformative potential in therapeutic design, enzyme engineering, and synthetic biology. While diffusion-based and flow matching approaches have achieved progress, they typically operate at single resolution and lack mechanisms for incorporating functional constraints. We introduce ProHiFlo, a hierarchical flow matching framework with three innovations: (1) coarse-to-fine generation that models backbone geometry before refining to all-atom coordinates, reducing computational cost while maintaining accuracy; (2) functional guidance leveraging pretrained predictors to steer generation toward desired properties without retraining; (3) adaptive SE(3)-equivariant architecture for efficient multi-scale processing. Experiments on unconditional generation, motif scaffolding, and functional design demonstrate state-ofthe-art performance while requiring 4 fewer sampling steps. On enzyme active site scaffolding, ProHiFlo achieves 58.9% success rate compared to 41.2% for RFDiffusion.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript introduces ProHiFlo, a hierarchical flow matching framework for de novo protein generation featuring three innovations: (1) coarse-to-fine modeling of backbone geometry before all-atom refinement, (2) functional guidance from pretrained predictors to steer toward desired properties without retraining the core model, and (3) an adaptive SE(3)-equivariant architecture. It reports state-of-the-art results on unconditional generation, motif scaffolding, and functional design tasks, including a reduction of 4 sampling steps and a 58.9% success rate on enzyme active site scaffolding versus 41.2% for RFDiffusion.

Significance. If the reported performance gains hold under rigorous validation, the work would be significant for computational protein design. The combination of hierarchical flow matching with predictor-based guidance offers a practical route to property-conditioned generation that avoids full retraining, while the reduced sampling steps address a key efficiency bottleneck in diffusion/flow-based methods for biomolecules. The approach extends existing equivariant generative models in a manner that could generalize to other functional constraints.

major comments (2)
  1. The central performance claims (e.g., 58.9% vs. 41.2% success on enzyme scaffolding) are attributed in part to functional guidance, yet the manuscript provides no quantitative assessment of how well the pretrained predictors transfer to generated (as opposed to natural) structures; without held-out accuracy, correlation with downstream success metrics, or ablation on guidance strength, it is unclear whether the observed delta can be credited to this component rather than the hierarchical architecture alone.
  2. The experimental section does not report error bars, number of independent runs, or statistical significance tests for the success-rate comparisons; given that the abstract already highlights specific numerical improvements, the absence of these details makes it impossible to assess whether the gains are robust or within expected variance of the baselines.
minor comments (2)
  1. Abstract: 'state-ofthe-art' contains a typographical error and should read 'state-of-the-art'.
  2. The description of the adaptive SE(3)-equivariant architecture would benefit from an explicit statement of how the multi-scale processing differs from prior equivariant flow-matching models (e.g., a brief comparison paragraph).

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments, which highlight important aspects of validation and statistical rigor. We address each major point below and outline planned revisions.

read point-by-point responses

  1. Referee: The central performance claims (e.g., 58.9% vs. 41.2% success on enzyme scaffolding) are attributed in part to functional guidance, yet the manuscript provides no quantitative assessment of how well the pretrained predictors transfer to generated (as opposed to natural) structures; without held-out accuracy, correlation with downstream success metrics, or ablation on guidance strength, it is unclear whether the observed delta can be credited to this component rather than the hierarchical architecture alone.

    Authors: We agree that isolating the contribution of functional guidance requires additional analysis. The reported gains reflect the full framework, but the manuscript does not include an explicit ablation on guidance strength or direct transfer metrics for generated structures. In revision we will add an ablation varying guidance strength and report any available correlations between predictor outputs and success rates. A full held-out accuracy evaluation on generated structures was not performed in the original experiments. revision: partial

  2. Referee: The experimental section does not report error bars, number of independent runs, or statistical significance tests for the success-rate comparisons; given that the abstract already highlights specific numerical improvements, the absence of these details makes it impossible to assess whether the gains are robust or within expected variance of the baselines.

    Authors: We acknowledge the value of reporting variability. The presented success rates follow the single-run protocols used by prior methods such as RFDiffusion, but we will conduct additional independent runs, include error bars, and add statistical significance tests for the key comparisons in the revised manuscript. revision: yes

Circularity Check

0 steps flagged

No circularity in derivation or claims

full rationale

The paper introduces an empirical ML architecture (hierarchical flow matching plus functional guidance via external pretrained predictors) and supports its performance claims through direct experimental comparisons against independent baselines such as RFDiffusion. No equations, predictions, or uniqueness arguments are shown to reduce to self-definitions, fitted inputs renamed as outputs, or load-bearing self-citations. The functional-guidance step relies on an external assumption about predictor transfer rather than any internal circular reduction. The derivation chain is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract provides no equations, derivations, or modeling details, so no free parameters, axioms, or invented entities can be identified.

pith-pipeline@v0.9.1-grok · 5682 in / 1077 out tokens · 13914 ms · 2026-06-28T07:09:25.865498+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

49 extracted references · 3 linked inside Pith

  1. [1]

    The coming of age of de novo protein design

    Po-Ssu Huang, Scott E Boyken, and David Baker. The coming of age of de novo protein design. Nature, 537(7620):320–327, 2016

    2016

  2. [2]

    Advanced deep learning methods for protein structure prediction and design.BIO Integration, 2025

    Yichao Zhang, Ningyuan Deng, Xinyuan Song, Ziqian Bi, Tianyang Wang, Zheyu Yao, Keyu Chen, Ming Li, Qian Niu, Junyu Liu, et al. Advanced deep learning methods for protein structure prediction and design.BIO Integration, 2025

    2025

  3. [3]

    Highly accurate protein structure prediction with AlphaFold.Nature, 596(7873):583–589, 2021

    John Jumper, Richard Evans, Alexander Pritzel, Tim Green, Michael Figurnov, Olaf Ron- neberger, Kathryn Tunyasuvunakool, Russ Bates, Augustin Žídek, Anna Potapenko, et al. Highly accurate protein structure prediction with AlphaFold.Nature, 596(7873):583–589, 2021

    2021

  4. [4]

    R2i-bench: Benchmarking reasoning-driven text-to-image generation

    Kaijie Chen, Zihao Lin, Zhiyang Xu, Ying Shen, Yuguang Yao, Joy Rimchala, Jiaxin Zhang, and Lifu Huang. R2i-bench: Benchmarking reasoning-driven text-to-image generation. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 12606–12641, 2025

    2025

  5. [5]

    De novo design of protein structure and function with RFdiffusion.Nature, 620(7976):1089–1100, 2023

    Joseph L Watson, David Juergens, Nathaniel R Bennett, Brian L Trippe, Jason Yim, Helen E Eisenach, Woody Ahern, Andrew J Borber, Robert J Ragotte, Lukas F Milles, et al. De novo design of protein structure and function with RFdiffusion.Nature, 620(7976):1089–1100, 2023

    2023

  6. [6]

    Illuminating protein space with a programmable generative model.Nature, 623(7989):1070–1078, 2023

    John B Ingraham, Max Barber, Greta Wilber, Luke Strom, Chandra Theesfeld, Julia Listgarten, Gabriele Corso, Tommi Jaakkola, and Regina Barzilay. Illuminating protein space with a programmable generative model.Nature, 623(7989):1070–1078, 2023

    2023

  7. [7]

    Protein generation with evolutionary diffusion: sequence is all you need

    Sarah Alamdari, Nitya Thakkar, Rianne van den Berg, Alex X Lu, Nicolo Fusi, Ava P Amini, and Kevin K Yang. Protein generation with evolutionary diffusion: sequence is all you need. bioRxiv, pages 2023–09, 2023

    2023

  8. [8]

    From noise to nuance: Advances in deep generative image models.arXiv:2412.09656, 2024

    Benji Peng, Chia Xin Liang, Ziqian Bi, Ming Liu, Yichao Zhang, Tianyang Wang, Keyu Chen, Xinyuan Song, and Pohsun Feng. From noise to nuance: Advances in deep generative image models.arXiv:2412.09656, 2024

    arXiv 2024

  9. [9]

    Drdgrl: Dual-relational dynamic graph repre- sentation learning for delay-sensitive stock trend prediction

    Mingjie You, Kaijie Chen, and Dawei Cheng. Drdgrl: Dual-relational dynamic graph repre- sentation learning for delay-sensitive stock trend prediction. InInternational Conference on Database Systems for Advanced Applications, pages 35–50. Springer, 2026

    2026

  10. [10]

    Memmark: State-evolution attribution watermarking for agent long-term memory systems.arXiv preprint arXiv:2605.25002, 2026

    Haobo Zhang, Xutao Mao, Guangyuan Dong, Ziwei Li, Xuanbo Su, Kaijie Chen, Jing Yang, and Zheng Lin. Memmark: State-evolution attribution watermarking for agent long-term memory systems.arXiv preprint arXiv:2605.25002, 2026. 13

    Pith/arXiv arXiv 2026

  11. [11]

    Denoising diffusion probabilistic models.Advances in Neural Information Processing Systems, 33:6840–6851, 2020

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models.Advances in Neural Information Processing Systems, 33:6840–6851, 2020

    2020

  12. [12]

    Flow matching for generative modeling

    Yaron Lipman, Ricky TQ Chen, Heli Ben-Hamu, Maximilian Nickel, and Matthew Le. Flow matching for generative modeling. InInternational Conference on Learning Representations, 2023

    2023

  13. [13]

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

    Zeming Lin, Halil Akin, Roshan Rao, Brian Hie, Zhongkai Zhu, Wenting Lu, 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

  14. [14]

    Attention is all you need

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need. InAdvances in Neural Informa- tion Processing Systems, volume 30, 2017

    2017

  15. [15]

    Prottrans: Toward understanding the language of life through self-supervised learning.IEEE Transactions on Pattern Analysis and Machine Intelligence, 44(10):7112–7127, 2022

    Ahmed Elnaggar, Michael Heinzinger, Christian Dallago, Ghalia Rehawi, Yu Wang, Llion Jones, Tom Gibbs, Tamas Feher, Christoph Angerer, Martin Steinegger, et al. Prottrans: Toward understanding the language of life through self-supervised learning.IEEE Transactions on Pattern Analysis and Machine Intelligence, 44(10):7112–7127, 2022

    2022

  16. [16]

    Structure-based protein function prediction using graph convolutional networks.Nature Communications, 12(1):3168, 2021

    Vladimir Gligorijevi ´c, P Douglas Renfrew, Tomasz Kosciolek, Julia Koehler Ber, Daniel Berenberg, Tommi Vatez, Chris Chandler, Andre Taylor-Compston, Brendan J Frey, and Richard Bonneau. Structure-based protein function prediction using graph convolutional networks.Nature Communications, 12(1):3168, 2021

    2021

  17. [17]

    Equiv- ariant graph neural networks for 3d macromolecular structure.arXiv preprint arXiv:2106.03843, 2021

    Bowen Jing, Stephan Eismann, Patricia Suriana, Raphael JL Townshend, and Ron Dror. Equiv- ariant graph neural networks for 3d macromolecular structure.arXiv preprint arXiv:2106.03843, 2021

    arXiv 2021

  18. [18]

    Generating functional protein variants with variational autoencoders.PLOS Computational Biology, 17(2):e1008736, 2021

    Alex Hawkins-Hooker, Florence Depardieu, Adrian Baez-Ortega, Marie Touchon, Eduardo PC Rocha, Ilaria Granata, Michael PH Brown, and Michael A Savageau. Generating functional protein variants with variational autoencoders.PLOS Computational Biology, 17(2):e1008736, 2021

    2021

  19. [19]

    Ig-vae: Generative modeling of protein structure by direct 3d coordinate generation.PLOS Computational Biology, 18(6):e1010271, 2022

    Raphael R Eguchi, Christian A Choe, and Po-Ssu Huang. Ig-vae: Generative modeling of protein structure by direct 3d coordinate generation.PLOS Computational Biology, 18(6):e1010271, 2022

    2022

  20. [20]

    Generative models for graph-based protein design

    John Ingraham, Vikas Garg, Regina Barzilay, and Tommi Jaakkola. Generative models for graph-based protein design. InAdvances in Neural Information Processing Systems, volume 32, 2019

    2019

  21. [21]

    Se(3) diffusion model with application to protein backbone generation

    Jason Yim, Brian L Trippe, Valentin De Bortoli, Emile Mathieu, Arnaud Doucet, Regina Barzilay, and Tommi Jaakkola. Se(3) diffusion model with application to protein backbone generation. InInternational Conference on Machine Learning, pages 40001–40039. PMLR, 2023

    2023

  22. [22]

    Mvi-bench: A comprehensive benchmark for evaluating robustness to misleading visual inputs in lvlms.arXiv preprint arXiv:2511.14159, 2025

    Huiyi Chen, Jiawei Peng, Dehai Min, Changchang Sun, Kaijie Chen, Yan Yan, Xu Yang, and Lu Cheng. Mvi-bench: A comprehensive benchmark for evaluating robustness to misleading visual inputs in lvlms.arXiv preprint arXiv:2511.14159, 2025

    Pith/arXiv arXiv 2025

  23. [23]

    Gui agents for continual game generation.arXiv preprint arXiv:2605.28258, 2026

    Yixu Huang, Bo Li, Na Li, Zhe Wang, Kaijie Chen, Haonan Ge, Qingyi Si, Yuanzhe Shen, Ruihan Yang, Guangjing Wang, et al. Gui agents for continual game generation.arXiv preprint arXiv:2605.28258, 2026

    Pith/arXiv arXiv 2026

  24. [24]

    Superflow: Training flow matching models with rl on the fly.arXiv preprint arXiv:2512.17951, 2025

    Kaijie Chen, Zhiyang Xu, Ying Shen, Zihao Lin, Yuguang Yao, and Lifu Huang. Superflow: Training flow matching models with rl on the fly.arXiv preprint arXiv:2512.17951, 2025. 14

    arXiv 2025

  25. [25]

    Generating novel, designable, and diverse protein structures by equivariantly diffusing oriented residue clouds

    Yeqing Lin and Mohammed AlQuraishi. Generating novel, designable, and diverse protein structures by equivariantly diffusing oriented residue clouds. InInternational Conference on Machine Learning, pages 21312–21333. PMLR, 2023

    2023

  26. [26]

    Out of many, one: Designing and scaffolding proteins at the scale of the structural universe with Genie 2.arXiv preprint arXiv:2405.15489, 2024

    Yeqing Lin, Minji Lee, Zhao Zhang, and Mohammed AlQuraishi. Out of many, one: Designing and scaffolding proteins at the scale of the structural universe with Genie 2.arXiv preprint arXiv:2405.15489, 2024

    arXiv 2024

  27. [27]

    Se(3)-stochastic flow matching for protein backbone generation

    Avishek Joey Bose, Tara Akhound-Sadegh, Kilian Fatras, Guillaume Huguet, Jarrid Rector- Brooks, Cheng-Hao Liu, Andrei Cristian Nica, Maksym Korablyov, Michael Bronstein, and Alexander Tong. Se(3)-stochastic flow matching for protein backbone generation. InInterna- tional Conference on Learning Representations, 2024

    2024

  28. [28]

    Fast protein backbone generation with SE(3) flow matching.arXiv preprint arXiv:2310.05297, 2023

    Jason Yim, Andrew Campbell, Andrew YK Foong, Michael Gastegger, José Jiménez-Luna, Sarah Lewis, Victor Garcia Satorras, Bastiaan S Veeling, Regina Barzilay, Tommi Jaakkola, et al. Fast protein backbone generation with SE(3) flow matching.arXiv preprint arXiv:2310.05297, 2023

    arXiv 2023

  29. [29]

    Joint generation of protein sequence and structure with RoseTTAFold sequence space diffusion

    Sidney Lyayuga Lisanza, Jake M Gershon, Sam WK Tipps, Jerald A Arnesen, Chenlin Zhu, Samuel J Zandberg, Rishi Raman, Casper Bakker, W Sebastian Koska, Dustin Lehnert, et al. Joint generation of protein sequence and structure with RoseTTAFold sequence space diffusion. bioRxiv, pages 2023–05, 2023

    2023

  30. [30]

    Protgpt2 is a deep unsupervised language model for protein design.Nature Communications, 13(1):4348, 2022

    Noelia Ferruz, Steffen Schmidt, and Birte Höcker. Protgpt2 is a deep unsupervised language model for protein design.Nature Communications, 13(1):4348, 2022

    2022

  31. [31]

    Protein design with guided discrete diffusion.Advances in Neural Information Processing Systems, 36, 2024

    Nate Gruver, Samuel Stanton, Nathan C Frey, Tim GJ Rudner, Isidro Hotzel, Julien Lafrance- Vanasse, Arvind Rajpal, Kyunghyun Cho, and Andrew Gordon Wilson. Protein design with guided discrete diffusion.Advances in Neural Information Processing Systems, 36, 2024

    2024

  32. [32]

    E(n) equivariant graph neural networks

    Víctor Garcia Satorras, Emiel Hoogeboom, and Max Welling. E(n) equivariant graph neural networks. InInternational Conference on Machine Learning, pages 9323–9332. PMLR, 2021

    2021

  33. [33]

    Se(3)-transformers: 3d roto-translation equivariant attention networks

    Fabian Fuchs, Daniel Worrall, V olker Fischer, and Max Welling. Se(3)-transformers: 3d roto-translation equivariant attention networks. InAdvances in Neural Information Processing Systems, volume 33, pages 1970–1981, 2020

    1970

  34. [34]

    Equiformer: Equivariant graph attention transformer for 3d atomistic graphs

    Yi-Lun Liao and Tess Smidt. Equiformer: Equivariant graph attention transformer for 3d atomistic graphs. InInternational Conference on Learning Representations, 2023

    2023

  35. [35]

    The protein data bank.Nucleic Acids Research, 28(1):235–242, 2000

    Helen M Berman, John Westbrook, Zukang Feng, Gary Gilliland, Talapady N Bhat, Helge Weissig, Ilya N Shindyalov, and Philip E Bourne. The protein data bank.Nucleic Acids Research, 28(1):235–242, 2000

    2000

  36. [36]

    Mmseqs2 enables sensitive protein sequence searching for the analysis of massive data sets.Nature Biotechnology, 35(11):1026–1028, 2017

    Martin Steinegger and Johannes Söding. Mmseqs2 enables sensitive protein sequence searching for the analysis of massive data sets.Nature Biotechnology, 35(11):1026–1028, 2017

    2017

  37. [37]

    Alphafold protein structure database: massively expanding the structural coverage of protein-sequence space with high-accuracy models.Nucleic Acids Research, 50(D1):D419–D427, 2022

    Mihaly Varadi, Stephen Anyango, Mandar Deshpande, Sreenath Nair, Cyrus Natassia, Galabina Yordanova, David Yuan, Oana Stroe, Gemma Wood, Agata Laydon, et al. Alphafold protein structure database: massively expanding the structural coverage of protein-sequence space with high-accuracy models.Nucleic Acids Research, 50(D1):D419–D427, 2022

    2022

  38. [38]

    Robust deep learning–based protein sequence design using ProteinMPNN.Science, 378(6615):49–56, 2022

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

    2022

  39. [39]

    Diffdock: Diffusion steps, twists, and turns for molecular docking.International Conference on Learning Representations, 2023

    Gabriele Corso, Hannes Stärk, Bowen Jing, Regina Barzilay, and Tommi Jaakkola. Diffdock: Diffusion steps, twists, and turns for molecular docking.International Conference on Learning Representations, 2023. 15

    2023

  40. [40]

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

    2021

  41. [41]

    Deepsol: a deep learning framework for sequence-based protein solubility prediction.Bioinformatics, 34(15):2605–2613, 2018

    Sameer Khurana, Reda Rawi, Kumardeep Kuber, Shaomin Hadar, Ohad Manor, Christine Orengo, Douglas EV Pires, David B Ascher, Lenore Cowen, and Gaurav Bhardwaj. Deepsol: a deep learning framework for sequence-based protein solubility prediction.Bioinformatics, 34(15):2605–2613, 2018

    2018

  42. [42]

    Openmm 7: Rapid development of high performance algorithms for molecular dynamics.PLOS Computational Biology, 13(7):e1005659, 2017

    Peter Eastman, Jason Swails, John D Chodera, Robert T McGibbon, Yutong Zhao, Kyle A Beauchamp, Lee-Ping Wang, Andrew C Simmonett, Matthew P Harrigan, Chaya D Stern, et al. Openmm 7: Rapid development of high performance algorithms for molecular dynamics.PLOS Computational Biology, 13(7):e1005659, 2017

    2017

  43. [43]

    Macromolecular modeling and design in Rosetta: recent methods and frameworks.Nature Methods, 17(7):665–680, 2020

    Julia Koehler Leman, Brian D Weitzner, Steven M Lewis, Jared Adolf-Bryfogle, Nawsad Alam, Rebecca F Alford, Melanie Aprahamian, David Baker, Kyle A Barlow, Patrick Barth, et al. Macromolecular modeling and design in Rosetta: recent methods and frameworks.Nature Methods, 17(7):665–680, 2020

    2020

  44. [44]

    Tm-align: a protein structure alignment algorithm based on the tm-score.Nucleic Acids Research, 33(7):2302–2309, 2005

    Yang Zhang and Jeffrey Skolnick. Tm-align: a protein structure alignment algorithm based on the tm-score.Nucleic Acids Research, 33(7):2302–2309, 2005

    2005

  45. [45]

    Neural ordinary differential equations

    Ricky TQ Chen, Yulia Rubanova, Jesse Bettencourt, and David K Duvenaud. Neural ordinary differential equations. InAdvances in Neural Information Processing Systems, volume 31, 2018. 16 A Theoretical Analysis A.1 Convergence Guarantee for Hierarchical Flow Matching We provide theoretical justification for our hierarchical approach by analyzing the converge...

    2018

  46. [46]

    Design 8 sequences using ProteinMPNN with sampling temperature 0.1 19 Table 6: Model hyperparameters. Parameter Stage 1 Stage 2 Hidden dimension 384 256 Number of layers 12 8 Attention heads 12 8 Dropout 0.1 0.1 Learning rate3×10 −4 3×10 −4 Batch size 256 128 Training steps 500K 300K

  47. [47]

    Predict structures for all sequences using ESMFold

  48. [48]

    Compute scTM score between generated backbone and predicted structures

  49. [49]

    Report designability as fraction with max(scTM)>0.5 D Additional Results D.1 Per-Length Designability Breakdown Table 7: Designability breakdown by protein length. Method 50-100 100-150 150-200 200-250 250-300 RFDiffusion 0.912 0.867 0.798 0.723 0.651 Chroma 0.894 0.834 0.756 0.689 0.612 FoldFlow-2 0.923 0.889 0.834 0.778 0.712 ProHiFlo 0.967 0.945 0.912 ...