Pith. sign in

REVIEW 4 major objections 6 minor 28 references

NeuralPLexer3: Accurate Biomolecular Complex Structure Prediction with Flow Models

T0 review · 4 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read NeuralPLexer3 claims state-of-the-art biomolecular complex prediction using flow-based generative modeling, beating AlphaFold3 on combined PoseBusters accuracy while running in roughly 30 seconds per prediction.

desk verdict Serious systems paper with useful benchmarks, but the AF3 beat rests on a ranking postprocessor and unmatched baselines, so the SOTA claim is provisional. read the letter →

arxiv 2412.10743 v2 pith:7LPZRULG submitted 2024-12-14 cs.LG physics.chem-phq-bio.BM

Zhuoran Qiao , Feizhi Ding , Thomas Dresselhaus , Mia A. Rosenfeld , Xiaotian Han , Owen Howell , Aniketh Iyengar , Stephen Opalenski
show 5 more authors
Anders S. Christensen Sai Krishna Sirumalla Frederick R. Manby Thomas F. Miller III Matthew Welborn
This is my paper · ORCID
classification cs.LGphysics.chem-phq-bio.BM
keywords protein–ligandstructurepredictionflowmatchingcontinuousnormalizingflowsbiomolecularcomplexPoseBustersbenchmarkconformationalchangestructure-baseddrugdesignphysicalvalidity
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

NeuralPLexer3 (NP3) is a flow-based generative model that predicts the 3D structure of biomolecular complexes—proteins, nucleic acids, ligands, ions, and post-translational modifications—from sequence and molecular topology alone. The paper's central claim is that NP3 outperforms AlphaFold3 on the combined PoseBusters protein–ligand benchmark, with a 78.4% success rate requiring both sub-2 Å RMSD and physical validity against AF3's 73.1%, while running a prediction in roughly 30 seconds on one L40S GPU rather than several A100-minutes. The work also introduces two new evaluation resources: NPBench, for low-homology recent PDB structures covering diverse interaction types and stoichiometry, and ConfBench, for apo/holo ligand-induced conformational changes. A sympathetic reader should care because these are the capabilities drug discovery actually needs: physically plausible poses, generalization to novel targets, conformational sensitivity, and speed.

What carries the argument

The central object is a continuous normalizing flow (CNF) trained by flow matching, which maps a simple prior distribution to the distribution of all heavy-atom coordinates of a biomolecular complex by learning a velocity field. Three mechanisms carry the argument: a physics-inspired globular polymer prior (random atom configurations relaxed by a short Langevin dynamics with harmonic connectivity and confinement terms) that starts sampling from chemically sensible structures; a simulation-free optimal-transport symmetry correction that permutes equivalent entities and atoms so the conditional flow trajectories between prior and ground truth are straightened; and a vector-field reparameterization that predicts denoised coordinates and applies optimal rigid alignment, reducing the number of integrator steps to 40. The architecture itself is an encoder–decoder transformer with anchor-level conditioning, MSA and language-model embeddings, and Flash-TriangularAttention, a kernel that avoids explicit bias broadcasting in the QK^T + bias operation and cuts peak memory by about 5x so training can use larger structure crops.

What would settle it

A head-to-head rerun of AF3 and NP3 on the same PoseBusters-V2 targets with identical code for pocket-aligned RMSD, PB-valid checks, and conformer ranking would settle the claim; if AF3's combined success rate under that protocol exceeds 78.4% or NP3's falls below 73.1%, the stated advantage is reversed.

Watch

Extended reading notes

Core claim

On its own terms, the paper establishes that a conditional flow-matching model with a physics-informed prior can be the best current sequence-only predictor of biomolecular interactions: NP3 achieves 78.4% combined PoseBusters success versus 73.1% for AF3 and 98.8% ligand stereochemistry accuracy, matches AF3 on raw coordinate error (80.2% vs 80.4%), and outperforms AlphaFold2-Multimer on protein–peptide interfaces and on apo/holo conformational-change prediction while remaining competitive on monomers, PPIs, and CASP15 RNA. The authors attribute this to flow matching rather than diffusion: an informative globular-polymer prior relaxed by Langevin dynamics, optimal-transport-based symmetry correction that straightens conditional flows, prediction of denoised coordinates with rigid alignment, and a 40-step sampler that removes expensive diffusion rollouts. They further report compute-optimal scaling behavior with an encoder/decoder FLOP ratio near 10 and 20 decoder replicas, and describe Flash-TriangularAttention as the memory optimization that allows large-crop training.

Load-bearing premise

The headline comparison assumes that AlphaFold3's published PoseBusters numbers were computed under the same benchmark protocol—same dataset version, same pocket-aligned RMSD, same physical-validity filters—as the numbers reported for NeuralPLexer3, because the paper does not rerun AlphaFold3 itself.

Editorial extensions

If this is right

  • If the 78.4% versus 73.1% PoseBusters result holds, sequence-only prediction has overtaken the previous best model on the combined accuracy-plus-physical-validity metric that matters for structure-based drug design.
  • A prediction in about 30 seconds on a single L40S GPU would make virtual screening of large compound libraries practical, since the cost per target-ligand pair drops by roughly two orders of magnitude relative to the reported AF3 timing.
  • The 98.8% ligand stereochemistry accuracy and PB-valid rates imply predicted poses need less post-hoc filtering before use in medicinal chemistry.
  • The ConfBench results imply the model can sometimes track ligand-induced conformational changes, including apo-to-holo transitions in kinases, which is a prerequisite for predicting allosteric effects and induced-fit selectivity.
  • NPBench's low-homology recent-PDB evaluation provides a reusable standard for testing generalization to unseen chains, ligands, and stoichiometries, and the paper's RNA results suggest MSA-free conditioning with language models can approach MSA-based accuracy.

Reading between the lines

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

  • Editorial inference: if the timing comparison is done on matched hardware and includes the full sampling and ranking pipeline, the reported seconds-per-target speed would enable enumerating hundreds of conformers per ligand, which current diffusion models cannot afford; the paper only reports a single-inference time, so the end-to-end gain is likely smaller but still substantial.
  • Editorial inference: the AF3 comparison rests on published benchmark tables rather than a local rerun; a protocol-matched head-to-head could shift the 5.3-point gap in either direction, so the durable claim is that NP3 is competitive and faster, not necessarily that it wins by that exact margin.
  • Editorial inference: the ConfBench scoring function is transferable: any structure prediction model could be scored on the same apo/holo pairs, making 'does this model see ligand-induced motion?' a standard quantitative question rather than a qualitative one.
  • Editorial inference: the success of RNA language-model conditioning suggests that for organisms with sparse MSA coverage, structure prediction may rely more on learned sequence embeddings, which could extend NP3-style models to non-model organisms.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. NeuralPLexer3 (NP3) is a flow-matching generative model for all-atom biomolecular complex structure prediction. The model uses a Langevin-relaxed globular polymer prior, optimal-transport entity and atom permutation, rigid-alignment vector-field reparameterization, an encoder-decoder architecture with Flash-TriangularAttention, and a confidence module. The authors report state-of-the-art protein-ligand accuracy on PoseBusters (78.4% combined success vs 73.1% for AlphaFold3, with coordinate-only accuracy 80.2% vs 80.4%), 98.8% ligand stereochemistry accuracy, roughly 30 seconds of inference on one L40S GPU versus about six A100-minutes reported for AF3, broad interaction coverage on a new NPBench suite, and a new ConfBench benchmark for ligand-induced conformational change, where NP3 outperforms AlphaFold2-Multimer. They also present a scaling-law analysis of encoder/decoder capacity and decoder-replica count, and they release the NPBench code.

Significance. The paper is a substantial engineering contribution to biomolecular structure prediction. The flow-matching formulation with informative priors, the detailed ablations in Figure 2C, the open NPBench code, and the explicit sampling and training algorithms in the SI are valuable. The speed and physical-validity improvements are plausible and practically important for drug discovery. However, the headline superiority over AF3 is not established as stated: the AF3 baseline is not protocol-matched, the entire margin is in a PB-valid component that the conformer-ranking step explicitly optimizes, and the new NPBench and ConfBench benchmarks are self-built. With those caveats, the value of the work lies more in an efficient, broadly applicable predictor with comparable coordinate accuracy and strong physical validity than in a demonstrated accuracy advantage over AF3.

major comments (4)
  1. [Methods – Baselines; Table 1; Figure 1B] The central claim that NP3 improves protein-ligand binding structure prediction with greater accuracy than AF3 is not supported by a protocol-matched comparison. The Methods state that AF3 baseline numbers are taken from Abramson et al. Extended Data Table 1, and the paper does not rerun AF3 under its own evaluation pipeline. Because the coordinate-only success rate is essentially tied (80.2% vs 80.4% RMSD < 2 Å in Table 1), the reported 5.3-point margin in the combined PoseBusters metric rests entirely on the PB-valid component. To make the headline claim credible, the authors should either rerun AF3 with the identical PoseBusters-V2 dataset definition, the same pocket-aligned RMSD calculation, the same physical-validity filters, and the same sample-ranking protocol, or explicitly and prominently reframe the comparison as literature-based. In either case, confidence intervals or bootstrap error bars for the success rates are needed.
  2. [SI S.5; Figure 2C] The combined PoseBusters success rate is partly an artifact of the ranking procedure. SI S.5 defines Score = pLDDT(LG) – 1000 × (is_clash + is_chirality_violation), and Figure 2C explicitly credits the clash and chirality penalties for PoseBusters gains. Since PB-valid includes clash and chirality checks, selecting the top-ranked sample with this score directly optimizes the validity component of the combined metric. The paper should report per-sample or unranked PoseBusters accuracy, such as the accuracy of a randomly chosen sample or the full distribution over samples, and apply the same ranking protocol to both NP3 and AF3 before claiming that NP3's generative model is more accurate.
  3. [Results – Compute-optimal scaling; Eq. (3)] Equation (3) defines Compute (FLOPs) = (α · β · P) · D with α and β described as encoder and decoder FLOPs. As written, the units are not FLOPs: if α and β are per-sample FLOPs, the total should be (α + β) · P · D. This inconsistency affects the compute-optimal frontier analysis in Figure 2B and the relative capacity comparison in Figure 2A; please correct the formula and verify that the scaling conclusions are unchanged.
  4. [Methods – Conformational Benchmarks; Eq. (4); SI S.7.2 Eqs. (5)-(6)] The main-text ConfBench score in Eq. (4) differs from the SI formulas in S.7.2. With the main-text denominator sqrt(RMSD_alt^2 + RMSD_ref^2 + RMSD_altref^2) multiplied by 1/2, a query identical to the reference does not receive score 1, contradicting the stated interpretation; the SI version uses a denominator of sqrt((1/2)(...)), which does satisfy that boundary. This discrepancy changes the meaning of a ConfBench score greater than 0 and could affect the reported win rates of 51.9% versus 29.6% and the apo/holo correctness rates. Please reconcile the equations and re-verify the conformational statistics with the corrected scoring function.
minor comments (6)
  1. [Methods – Baselines] In the baselines paragraph, 'Abramsom' should be 'Abramson'.
  2. [NPBench description; Methods – Structure Prediction Benchmarks] The main text says NPBench contains structures released after 2023, while the Methods give a deposition window of 2022-05-01 to 2023-01-12; these dates should be reconciled.
  3. [Figure 2C caption; Eq. (3)] The Figure 2C caption uses 'C: Compute in GFLOPs' and 'P: number of decoder replicas', but the text and Eq. (3) use α and β for encoder and decoder FLOPs and P for replicas; the notation should be made consistent.
  4. [Results – Computational Efficiency] The inference-time comparison of roughly 30 seconds on one L40S GPU versus about six A100-minutes for AF3 is based on timing statistics reported in the AF3 paper on different hardware; this should be described as an approximate, non-protocol-matched comparison.
  5. [Results – NPBench and ConfBench] The paper should state more prominently that NPBench and ConfBench are newly introduced benchmarks built by the authors, and that independent external validation is pending; the release of NPBench code is helpful but does not by itself establish community acceptance of the benchmark.
  6. [Discussion – Limitations of ConfBench] The Discussion acknowledges limitations of ConfBench, namely imperfect scores and apo structures that are not always fully ligand-free; these limitations should also be stated next to the headline conformational win rates in the Results section.

Circularity Check

1 steps flagged · score 5.0 of 10

Headline PoseBusters margin over AF3 rests on a validity-aware ranking rule that directly optimizes the PB-valid component of the same benchmark; the coordinate-accuracy component is tied, and AF3 baselines are not protocol-matched.

  1. fitted input called prediction [SI S.5 "Ranking Methodology"; main text Fig. 2C and "Protein Structure and Interaction Accuracy"]
    "Ranking Methodology: Sample ranking leverages adjusted pLDDT and pDockQ values. Scores are penalized for steric clashes, ensuring physically plausible configurations. • For a ligand of interest, LG: Score = pLDDT(LG) – 1000 * (is_clash + is_chirality_violation) (SI S.5); "Key components that are found impactful include ... the incorporation of clash and chirality penalties in conformer ranking" (Fig. 2C)."

    Table 1 shows the headline margin over AF3 is not in coordinate accuracy: NP3 gets 80.2% RMSD<2 Å vs AF3's 80.4%, while the combined PoseBusters success (78.4 vs 73.1) is carried entirely by the PB-valid component. PB-valid is defined as passing clash, chirality, and related physical plausibility checks. The inference-time ranking score is pLDDT minus 1000 times (is_clash + is_chirality_violation), i.e., it selects the sample that already satisfies the two main PB-valid criteria. Figure 2C credits these penalties for PoseBusters gains. The reported PB-valid 'prediction' is therefore substantially an artifact of a selection rule keyed to the evaluation metric; the coordinate-accuracy comparison is tied. Because AF3 baseline numbers are taken from Abramson et al.

full rationale

The paper's central quantitative comparison is the combined PoseBusters success rate, 78.4% for NP3 versus 73.1% for AF3. The paper itself reports that the RMSD<2 Å component is essentially tied (80.2% vs 80.4%), so the only component producing the headline margin is PB-valid. That component is directly addressed by the conformer ranking score in SI S.5, which subtracts 1000 times the clash and chirality violation flags — the same physical-validity checks that PB-valid applies. The manuscript also credits these penalties for PoseBusters gains in Figure 2C. Thus the 'prediction' of PB-valid superiority is partly a selection effect: among generated samples, the ranking preferentially picks the sample that passes the evaluated validity checks. This is not a fully self-definitional derivation, because PB-valid is an external benchmark and the generative model still must produce plausible coordinates, but it substantially weakens the claim that the margin reflects a more accurate generative model. Separately, the AF3 baseline is taken from the published Extended Data Table 1 of Abramson et al. rather than produced under the same benchmark protocol, so any difference in dataset version, conformer ranking, pocket-aligned RMSD calculation, or physical-validity filtering confounds the comparison. That protocol mismatch is a validity threat, not itself a circular step. No other significant circularity is present: the flow-matching formulation, informative prior, model ablations, and scaling curves are empirical developments rather than definitional equivalences, and the self-citations to prior NeuralPLexer work are routine and not load-bearing for the central benchmark claims. The overall score of 5 reflects one partially circular step that affects the headline numerical claim, while the underlying generative modeling content retains independent value.

Assumptions & free parameters 7 free parameters · 6 assumptions · 0 invented entities

The central claims rest on a large trained model, so the ledger is dominated by hand-set hyperparameters and benchmark assumptions rather than mathematical axioms. No new physical entities are introduced. The most fragile entries are the assumed comparability of literature AF3 numbers and the leakage-control assumptions.

free parameters (7)
  • Flow integrator steps T = 40 (main text) or 100 (Algorithm S1)
    The number of ODE steps is a hand-set trade-off between sampling accuracy and the claimed 30-second inference time; the two values in the text are inconsistent.
  • Prior relaxation radii and schedule = res_r=4.0, ent_r=10.0, dt=0.25, 64 Langevin iterations
    Algorithm S3 defines the informative globular polymer prior; these constants are chosen by hand and the prior is a central claimed contribution.
  • Loss weights for auxiliary terms = w2, w3 not reported
    Algorithm S2 sums pseudo-Huber, SmoothLDDT, FAPE, and distance-geometry losses with weights w2 and w3, whose values are not given in the paper.
  • Alignment weights walign = ligand atoms 10.0, backbone C-alpha 1.0, side chains 0.0
    These weights in the weighted Kabsch alignment define the training target and are hand-set; the paper says they are important for late-stage training stability.
  • Timestep shift exponent = 1.15
    The polynomial timestep shift t* = t^1.15 is introduced in the SI to reduce exposure bias; the exponent is chosen empirically.
  • Conformer ranking penalty coefficient = 1000
    S.5 ranks samples by pLDDT or pDockQ minus 1000 for clashes or chirality violations; this directly influences the PB-valid success rate.
  • Training crop schedule and decoder replicas = P=32, 32, 24, 12 across crop sizes
    The production training stages in S.3 and the compute-optimal claim (alpha/beta=10, P=20) depend on these hand-tuned schedule choices.
assumptions (6)
  • domain assumption Experimental PDB structures are treated as ground truth for training and evaluation.
    All loss functions and benchmarks compare predictions to PDB-derived coordinates; errors in experimental structures propagate into reported accuracy.
  • domain assumption The September 2020 training cutoff plus clustering at 40% sequence identity and 0.6 Tanimoto similarity prevents training/evaluation leakage.
    S.3 and S.6 rely on this to argue NPBench and PoseBusters results reflect generalization rather than memorization.
  • ad hoc to paper AlphaFold3's published benchmark numbers are directly comparable to NP3's own evaluation pipeline.
    The headline 78.4% versus 73.1% comparison depends on this assumption because AF3 was not rerun; any protocol mismatch changes the conclusion.
  • domain assumption A 40-step discretization of the continuous normalizing flow ODE is an accurate approximation of the learned distribution.
    Algorithm S1 integrates the flow with a fixed number of steps; no convergence or error analysis is provided.
  • domain assumption ConfBench's score greater than 0 threshold and the 1.5 angstrom change filters define a meaningful measure of conformational prediction success.
    The benchmark excludes lock-and-key systems and uses the proposed scoring function to compare NP3 with AF2-Multimer.
  • domain assumption Synthetic training data generated by OpenFold, NP2, and OrbNet conformers do not bias generalization estimates.
    These distillation sets are used in training with probabilities up to 30%; their effect on benchmark accuracy is not separately controlled.

how reviews work

0 comments
Cite this review

Pith. "Pith review of NeuralPLexer3: Accurate Biomolecular Complex Structure Prediction with Flow Models." pith.science (2026). https://pith.science/paper/7LPZRULG

@misc{pith2026241210743,
  author       = {Pith},
  title        = {Pith review of: NeuralPLexer3: Accurate Biomolecular Complex Structure Prediction with Flow Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7LPZRULG}},
  note         = {Machine review of arXiv:2412.10743}
}
read the original abstract

Structure determination is essential to a mechanistic understanding of diseases and the development of novel therapeutics. Machine-learning-based structure prediction methods have made significant advancements by computationally predicting protein and bioassembly structures from sequences and molecular topology alone. Despite substantial progress in the field, challenges remain to deliver structure prediction models to real-world drug discovery. Here, we present NeuralPLexer3 -- a physics-inspired flow-based generative model that achieves state-of-the-art prediction accuracy on key biomolecular interaction types and improves training and sampling efficiency compared to its predecessors and alternative methodologies. Examined through newly developed benchmarking strategies, NeuralPLexer3 excels in vital areas that are crucial to structure-based drug design, such as physical validity and ligand-induced conformational changes.

Figures

Figures reproduced from arXiv: 2412.10743 by the authors.

Figure 1
Figure 1. NeuralPLexer3 (NP3) accurately predicts biomolecular structures with improved physical quality and prediction speed. (A) Schematics of the NP3 system. To perform a prediction, NP3 uses molecular topology extracted from input biopolymer sequences and small molecule graphs as primitive inputs, with additional conditioning signals from sequence language models, multiple sequence alignments (MSAs), and templates. NP3 ad… view at source ↗
Figure 2
Figure 2. NP3 pushes the frontier of structure prediction accuracy and training efficiency. (A) Comparison of relative encoder/decoder capacity in terms of floating-point operations per second (FLOPs) among different methods. The asterisk indicates an estimate based on our reproduction of AF3. (B) Model scaling behavior across molecule and interface types as reported by the relation between the average validation set local di… view at source ↗
Figure 3
Figure 3. Model confidence estimation. (A) Scatter plot of ligand RMSD against ligand pLDDT and on PoseBusters. (B) PoseBusters ligand RMSD success rate statistics grouped into NP3 confidence percentiles. Prediction success rates are consistently higher for higher confidence prediction bins. (C) Scatter plot of DockQ score against the pDockQ score on NPBench protein-protein and protein-nucleic acid interfaces. (D) PPI DockQ s… view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Illustration of Flash-TriangularAttention kernel and experimental comparison on the peak memory usage and the inference time. (A) The workflow for Flash-TriangularAttention. The main goal is to reduce peak memory usage, which is the bottleneck of triangular attention. …
Figure 5
Figure 5. Figure 5: ConfBench conformational prediction. (A) Schematic illustration of types of conformational changes of interest. (B) Benchmark target statistics. 57k global and 52k pocket rearrangement entries with meaningful motions and biologically relevant ligands are identified. (C…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

28 extracted references · 14 canonical work pages

  1. [1]

    Qiao, Z., Nie, W., Vahdat, A., Miller III, T. F. & Anandkumar, A. State-specific protein–ligand complex structure prediction with a multiscale deep generative model. Nature Machine Intelligence 6, 195–208. https: //www.nature.com/articles/s42256-024-00792-z (2024)

    work page 2024

  2. [2]

    Eddy, S. R. Accelerated Profile HMM Searches. PLOS Computational Biology 7, e1002195. https : / / journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1002195 (Oct. 2011)

    doi:10.1371/journal.pcbi.1002195 2011

  3. [3]

    Suzek, B. E. et al. UniRef clusters: a comprehensive and scalable alternative for improving sequence similarity searches. Bioinformatics 31, 926–932. https://doi.org/10.1093/bioinformatics/btu739 (Mar. 2015)

    doi:10.1093/bioinformatics/btu739 2015

  4. [4]

    Steinegger, M. et al. HH-suite3 for fast remote homology detection and deep protein annotation. BMC Bioinformatics 20, 473. https://doi.org/10.1186/s12859-019-3019-7 (Sept. 2019)

    work page doi:10.1186/s12859-019-3019-7 2019

  5. [5]

    & Söding, J

    Steinegger, M., Mirdita, M. & Söding, J. Protein-level assembly increases protein sequence recovery from metagenomic samples manyfold. Nature Methods 16, 603–606. https://www.nature.com/articles/ s41592-019-0437-4 (July 2019)

    work page 2019

  6. [6]

    Lin, Z. et al. Evolutionary-scale prediction of atomic-level protein structure with a language model. Science 379, 1123–1130. https://www.science.org/doi/10.1126/science.ade2574 (Mar. 2023)

    doi:10.1126/science.ade2574 2023

  7. [7]

    J., Vlaši´c, T., Huber, R

    Peni´c, R. J., Vlaši´c, T., Huber, R. G., Wan, Y . & Šiki´c, M. RiNALMo: General-Purpose RNA Language Models Can Generalize Well on Structure Prediction TasksNov. 2024. http://arxiv.org/abs/2403.00043

    arXiv 2024

  8. [8]

    & Elofsson, A

    Bryant, P., Pozzati, G. & Elofsson, A. Improved prediction of protein-protein interactions using AlphaFold2. Nature Communications 13, 1265. https://www.nature.com/articles/s41467- 022- 28865- w (Mar. 2022)

    work page 2022

Show all 28 references
  1. [9]

    Abramson, J. et al. Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature 630, 493–500. https://www.nature.com/articles/s41586-024-07487-w (June 2024)

    2024

  2. [10]

    Shirts, M. R. et al. Lessons learned from comparing molecular dynamics engines on the SAMPL5 dataset. Journal of Computer-Aided Molecular Design 31, 147–161. https://doi.org/10.1007/s10822- 016- 9977-1 (Jan. 2017)

    2017 doi

  3. [11]

    Esser, P. et al. Scaling Rectified Flow Transformers for High-Resolution Image Synthesis in International Conference on Machine Learning (June 2024). https://openreview.net/forum?id=FPnUhsQJ5B

    2024

  4. [12]

    & Fanti, G

    Lee, S., Lin, Z. & Fanti, G. Improving the Training of Rectified FlowsOct. 2024. http://arxiv.org/abs/ 2405.20320

    2024 arXiv

  5. [13]

    Touvron, H. et al. Llama 2: Open Foundation and Fine-Tuned Chat ModelsJuly 2023. http://arxiv.org/ abs/2307.09288

    2023 arXiv

  6. [14]

    & Sennrich, R

    Zhang, B. & Sennrich, R. Root Mean Square Layer Normalization in Advances in Neural Information Processing Systems 32 (Curran Associates, Inc., 2019). https://proceedings.neurips.cc/paper/2019/ hash/1e8a19426224ca89e83cef47f1e7f53b-Abstract.html

    2019

  7. [15]

    L., Kiros, J

    Ba, J. L., Kiros, J. R. & Hinton, G. E. Layer Normalization July 2016. http://arxiv.org/abs/1607.06450

    2016 arXiv

  8. [16]

    Ahdritz, G. et al. OpenFold: retraining AlphaFold2 yields new insights into its learning mechanisms and capacity for generalization. Nature Methods 21, 1514–1524. https://www.nature.com/articles/s41592- 024-02272-z (Aug. 2024)

    2024

  9. [17]

    Christensen, A. S. et al. OrbNet Denali: A machine learning potential for biological and organic chemistry with semi-empirical cost and DFT accuracy. The Journal of Chemical Physics 155, 204103. https://doi. org/10.1063/5.0061990 (Nov. 2021)

    doi

  10. [18]

    R., Pawar, S

    Henry, A., Dachapally, P. R., Pawar, S. & Chen, Y .Query-Key Normalization for Transformers Oct. 2020. http://arxiv.org/abs/2010.04245. 30/31

    2020 arXiv

  11. [19]

    Jumper, J. et al. Highly accurate protein structure prediction with AlphaFold. Nature 596, 583–589. https: //www.nature.com/articles/s41586-021-03819-2 (Aug. 2021)

    2021

  12. [20]

    Zhao, J., Schaefer, F. T. & Anandkumar, A.ZerO initialization: Initializing residual networks with only zeros and ones in Transactions on Machine Learning Research (2022). https://openreview.net/forum?id= 1AxQpKmiTc

    2022

  13. [21]

    & Xie, S

    Peebles, W. & Xie, S. Scalable Diffusion Models with Transformers Mar. 2023. http://arxiv.org/abs/ 2212.09748

    2023 arXiv

  14. [22]

    A solution for the best rotation to relate two sets of vectors

    Kabsch, W. A solution for the best rotation to relate two sets of vectors. Acta Crystallographica Section A 32, 922–923. https://onlinelibrary.wiley.com/doi/abs/10.1107/S0567739476001873 (1976)

    1976 doi

  15. [23]

    Least-squares estimation of transformation parameters between two point patterns

    Umeyama, S. Least-squares estimation of transformation parameters between two point patterns. IEEE Transactions on Pattern Analysis and Machine Intelligence13, 376–380. https://ieeexplore.ieee.org/ document/88573 (Apr. 1991)

    1991

  16. [24]

    Ansel, J. et al. PyTorch 2: Faster Machine Learning Through Dynamic Python Bytecode Transformation and Graph Compilation in Proceedings of the 29th ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Volume 2 2 (Association fo...

    2024

  17. [25]

    & Vinyals, O.Representation Learning with Contrastive Predictive Coding 2018

    Van den Oord, A., Li, Y . & Vinyals, O.Representation Learning with Contrastive Predictive Coding 2018. http://arxiv.org/abs/1807.03748

    2018 arXiv

  18. [26]

    Pymol: An open-source molecular graphics tool

    DeLano, W. Pymol: An open-source molecular graphics tool. CCP4 Newsletter on Protein Crystallography 40, 82–92. http://www.ccp4.ac.uk/newsletters/newsletter40/11_pymol.pdf (2002)

    2002

  19. [27]

    RDKit: Open-Source Cheminformatics Software

    Landrum, G. RDKit: Open-Source Cheminformatics Software. https : / / github . com / rdkit / rdkit / releases/tag/Release_2024_09_1 (2024)

    2024

  20. [28]

    & Wallner, B.DockQ v2: Improved automatic quality measure for protein multimers, nucleic acids, and small molecules June 2024

    Mirabello, C. & Wallner, B.DockQ v2: Improved automatic quality measure for protein multimers, nucleic acids, and small molecules June 2024. https://www.biorxiv.org/content/10.1101/2024.05.28. 596225v1. 31/31

    2024 doi

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.