Pith. sign in

REVIEW 3 major objections 5 minor 81 references

ToxBench: A Binding Affinity Prediction Benchmark with AB-FEP-Calculated Labels for Human Estrogen Receptor Alpha

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read ToxBench gives machine learning 8,770 AB-FEP-computed binding free energies for estrogen receptor alpha, and the DualBind model predicts them with 2.39 kcal/mol RMSE at a small fraction of the physics-based cost.

desk verdict ToxBench is a promising dataset and benchmark, but the 1.75 kcal/mol validation is a calibrated fit, not an independent check. read the letter →

arxiv 2507.08966 v1 pith:JJFIMPFE submitted 2025-07-11 cs.LG cs.AIphysics.chem-phq-bio.BM

Meng Liu , Karl Leswing , Simon K. S. Chu , Farhad Ramezanghorbani , Griffin Young , Gabriel Marques , Prerna Das , Anjali Panikar
show 14 more authors
Esther Jamir Mohammed Sulaiman Shamsudeen K. Shawn Watts Ananya Sen Hari Priya Devannagari Edward B. Miller Muyun Lihan Howook Hwang Janet Paulsen Xin Yu Kyle Gion Timur Rvachov Emine Kucukbenli Saee Gopal Paliwal
This is my paper
classification cs.LGcs.AIphysics.chem-phq-bio.BM
keywords bindingaffinitypredictionabsolutefreeenergyperturbationestrogenreceptoralphabenchmarkdatasetdenoisingscorematchingvirtualscreeningprotein–ligandinteractionsDualBind
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

ToxBench aims to fix a bottleneck in machine-learned binding affinity prediction: the lack of large, reliable training labels. The paper builds 8,770 ERα–ligand complexes with binding free energies computed by absolute binding free energy perturbation (AB-FEP), a simulation-based method whose accuracy is calibrated against 67 curated experimental affinities at 1.75 kcal/mol RMSE. Because the benchmark is dense (one target, 699 ligands) and split so that no ligand appears in more than one part, the authors argue models cannot hide behind ligand-only shortcuts. On this benchmark they propose DualBind, which combines a supervised mean-squared-error loss with an unsupervised denoising score matching loss, and report the best test performance: Pearson $R_p = 0.844$, $R^2 = 0.704$, Spearman $\rho = 0.786$, $\mathrm{RMSE} = 2.392$ kcal/mol. If correct, this gives the community a reusable ERα testbed and shows that a well-designed ML model can approximate AB-FEP at a fraction of the computational cost.

What carries the argument

The load-bearing object is ToxBench itself: 8,770 ERα–ligand complexes spanning 699 ligands across three PDB templates (1ERE, agonist-bound; 3ERT, antagonist-bound; 1SJ0, a third conformation), each labeled by a 1 ns AB-FEP simulation. To convert raw AB-FEP energies into experimental-like affinities, the authors apply empirical per-PDB structural reorganization penalties (+8.69 kcal/mol for 1ERE, +6.11 kcal/mol for 3ERT) and then take the lowest free energy across sampled conformations; this calibrated pipeline reaches 1.754 kcal/mol RMSE against 67 curated experimental values. The second mechanism is DualBind's dual loss: a mean-squared-error term anchors the predicted scalar energy to the AB-FEP label, while a denoising score matching (DSM) term adds Gaussian noise to ligand atom coordinates and forces the energy gradient to point back toward the unperturbed structure, shaping local minima in the learned energy landscape. The model is an SE(3)-invariant frame-averaging network with attention layers that reads the 50 residues closest to the ligand and has about 1.02 million parameters.

What would settle it

Take ToxBench ligands outside the 67 calibration compounds, obtain fresh experimental binding affinities for 50–100 of them, and compare those measurements against the corrected AB-FEP labels; if the RMSE substantially exceeds 1.75 kcal/mol or the per-PDB penalties do not reproduce on this fresh subset, the claim that all 8,770 labels are reliable is falsified.

Watch

Extended reading notes

Core claim

The central claim is that AB-FEP-derived labels can form a large-scale, reliable benchmark for a single pharmaceutically critical target, and that a dual-loss model can learn the AB-FEP binding energy function. To validate the labels, the authors compare corrected AB-FEP values against 67 manually curated experimental binding affinities, obtaining $\mathrm{RMSE}=1.754$ kcal/mol and Pearson $R_p = 0.692$, and they attribute the modest correlation to the narrow dynamic range of the validation set. On the ToxBench test split, DualBind outperforms both a ligand-only baseline (Chemprop) and an interaction-aware baseline (AEV-PLIG) on every metric, with $\mathrm{RMSE}=2.392$ kcal/mol. The paper also claims that the clear gap between ligand-only and interaction-aware models shows the dense single-target design forces models to learn genuine protein–ligand interactions, and reports roughly a million-fold inference speed-up over AB-FEP (126 ms per complex unbatched, 33 ms batched, versus about 35 hours per complex for the simulation).

Load-bearing premise

The labels are only as reliable as the assumption that the two per-PDB structural reorganization penalties, calibrated on 67 experimentally measured compounds, transfer to all other ligands and to the 1SJ0 template; if that calibration is overfit or does not transfer, every downstream model inherits a systematic bias in the labels.

Editorial extensions

If this is right

  • ToxBench provides a reusable, non-overlapping-ligand benchmark for ERα affinity prediction, so future models can be compared on the same AB-FEP-derived labels.
  • The performance gap between ligand-only and interaction-aware models indicates that dense single-target data pushes models to encode protein–ligand interactions rather than dataset-specific shortcuts.
  • DualBind's test RMSE of 2.392 kcal/mol suggests fast ML screening can rank ERα binders near AB-FEP quality, enabling high-throughput virtual screening of large chemical libraries.
  • The dataset construction protocol, combining ChEMBL and DUD-E ligands with three ERα conformations, 1 ns AB-FEP, and experimental calibration, is proposed as a blueprint for similar benchmarks on other targets.
  • Labels capped at -3.0 kcal/mol mark very weak or non-binding interactions, giving toxicity screeners a practical binder/non-binder boundary.

Reading between the lines

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

  • Editorial inference: the 1.75 kcal/mol validation RMSE is computed on 67 compounds that also set the per-PDB penalties, so it is not an independent accuracy guarantee for the other labels; a held-out experimental set would test whether the corrections transfer.
  • Editorial inference: the reported million-fold speed-up compares AB-FEP on a T4 GPU with DualBind inference on an A100, so a same-hardware comparison would give a fairer speed-up factor.
  • Editorial inference: the paper does not ablate the DSM loss, so how much of DualBind's gain comes from the dual-loss design rather than from the SE(3)-invariant architecture or the dense labels remains untested.
  • Editorial inference: a natural stress test is to train DualBind on ToxBench and evaluate on external ERα activity measurements, since the learned energy function should transfer beyond the AB-FEP label distribution if it has captured real interactions.
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

3 major / 5 minor

Summary. The paper introduces ToxBench, a dataset of 8,770 ERα-ligand complexes with binding free energies computed via 1 ns absolute binding free energy perturbation (AB-FEP), using three protein templates (1ERE, 3ERT, 1SJ0) and ligands from ChEMBL and DUD-E. The labels are adjusted with empirical per-PDB structural reorganization penalties (+8.69 kcal/mol for 1ERE, +6.11 kcal/mol for 3ERT) and the minimum free energy across conformations is selected. A subset of 67 ChEMBL compounds is compared with experimental affinities, reporting an RMSE of 1.754 kcal/mol and Pearson correlation 0.692. The paper also proposes DualBind, a 3D structure-based model trained with a combined MSE and denoising score matching loss, and benchmarks it against Chemprop and AEV-PLIG on non-overlapping ligand splits. DualBind achieves the best performance (RMSE 2.392 kcal/mol, Rp 0.844 on the test set).

Significance. If the ToxBench labels are reliable, the dataset would be a valuable dense single-target benchmark for structure-based binding affinity prediction, directly addressing the known shortcut-learning problem in sparse multi-target datasets. The paper makes the dataset and code publicly available, uses three-seed training, and reports non-overlapping ligand splits, which are good practices. However, the central claim of label validation is weakened by circularity: the per-PDB penalties and the min-selection rule are calibrated on the same 67 experimental values that are then used to report the validation RMSE. The absence of convergence diagnostics for the 1 ns AB-FEP runs further limits current evidence for label accuracy. With an independent or cross-validated assessment of label error, the dataset could become a significant community resource.

major comments (3)
  1. [Section 3, 'Agreement of Calculations with Experimental Data'] The reported validation RMSE of 1.754 kcal/mol is not an independent measure of label accuracy. The per-PDB structural reorganization penalties (+8.69 kcal/mol for 1ERE, +6.11 kcal/mol for 3ERT) and the rule of taking the lowest free energy across conformations are described as an 'algorithmic strategy to correlate' the AB-FEP output with the same 67 experimental values against which the RMSE is then computed. This makes 1.754 kcal/mol a training-set error for a two-parameter calibration, not a prediction error on unseen complexes. The abstract's statement that the dataset is 'validated against experimental affinities at 1.75 kcal/mol RMSE' is therefore overstated. Please provide a leave-one-out or held-out assessment of the calibration, or clearly label the 1.754 kcal/mol as an in-sample fit.
  2. [Section 3, 'Dataset Generation' and 'Agreement of Calculations with Experimental Data'] The transfer of the fitted offsets to the full dataset is unexamined. The penalties are derived from 67 ChEMBL compounds and are applied to all 8,770 complexes, including DUD-E-derived ligands and the 1SJ0 template, which receives no offset and does not appear in the 67-complex validation. If the offsets absorb ligand-series-specific or template-specific bias rather than a genuine physical reorganization free energy, the labels for the majority of the dataset could be systematically shifted by an amount comparable to the claimed accuracy. Please either provide physical justification for the offsets or validate them on a held-out set that includes DUD-E and 1SJ0 complexes.
  3. [Section 3, 'Dataset Generation'] No convergence diagnostics or uncertainty estimates are provided for the 1 ns AB-FEP simulations. Since the AB-FEP labels are the foundation of the benchmark, the paper should report block-averaging or bootstrap error estimates, or compare a subset of the computed values against longer simulations or multiple independent seeds. Without such information, the reader cannot determine how much of the 1.754 kcal/mol agreement is noise absorbed by the calibration, which is a necessary component of the dataset's reliability assessment.
minor comments (5)
  1. [Section 4, paragraph before Eq. (2)] The word 'struture' should be 'structure'.
  2. [References] Several references contain spacing artifacts, e.g., 'V olkov' and 'Schr ¨odinger'; please fix the author name formatting.
  3. [Figure 3 caption] The caption says 'kcal/mol' with a missing space; should be 'kcal/mol'.
  4. [Section 4, Eqs. (3)-(6)] The notation for the data distribution is unclear: pdata(X) is introduced without definition, and the role of q(X) is described only loosely. Please define all distributions explicitly, including how q(X|X) is obtained from the Gaussian perturbation in Eq. (2).
  5. [Section 5, 'Implementation Details'] The thresholding of labels at -3.0 kcal/mol is stated without justification. Please provide a citation or rationale for this cutoff, since it directly affects the training targets and the reported metrics.

Circularity Check

1 steps flagged · score 6.0 of 10

Validation RMSE is a calibration residual: the per-PDB offsets and lowest-energy rule are tuned on the same 67 experimental values used to report the 1.75 kcal/mol RMSE.

  1. fitted input called prediction [Section 3, 'Agreement of Calculations with Experimental Data' (Figure 2); abstract claims 'validated against experimental affinities at 1.75 kcal/mol RMSE'.]
    "An empirical, PDB-specific structural reorganization penalty was applied to the calculated free energies: +8.69 kcal/mol for the 1ERE structure and +6.11 kcal/mol for the 3ERT structure. Following these adjustments, the lowest (most favorable) free energy value across all sampled conformations for each ligand was designated as its final computational binding affinity. This workflow constitutes an initial algorithmic strategy to correlate the experimentally observed binding affinities with the predictions derived from conformational sampling and free energy calculations."

    The two per-PDB constants and the lowest-energy selection rule are the calibration procedure, and the 67 curated experimental values are the set on which that procedure is explicitly designed ('to correlate'). The resulting RMSE of 1.754 kcal/mol is then computed on those same 67 values after applying the fitted offsets and selection rule, so it is an in-sample residual of a two-constant calibration, not an independent validation. The abstract's phrase 'validated against experimental affinities at 1.75 kcal/mol RMSE' therefore presents a fitted quantity as a validated accuracy.

full rationale

The ML-benchmark portion is not circular: DualBind, Chemprop, and AEV-PLIG are evaluated on held-out ligand splits against AB-FEP-generated labels, so the learning comparison is self-contained and the test-set metrics are honest relative to those labels. The circularity is confined to the label-accuracy claim. Section 3 describes an 'empirical, PDB-specific structural reorganization penalty' (+8.69 kcal/mol for 1ERE, +6.11 kcal/mol for 3ERT) and a lowest-energy selection rule as 'an initial algorithmic strategy to correlate' the 67 experimental values, and then reports the resulting RMSE of 1.754 kcal/mol as validation. Because the offsets and selection rule are calibrated on the same 67 values used for the RMSE, the reported accuracy is a calibration residual rather than an independent prediction. The paper provides no held-out experimental check for 1SJ0-based complexes or DUD-E ligands and no uncertainty estimates for the 1 ns AB-FEP runs, so the abstract's validation language overstates the evidence. Score 6 reflects partial circularity in the central validation claim while acknowledging that the ML benchmark itself is independently executed.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The central claim rests on the validity of AB-FEP labels, which in turn depend on 1 ns simulations, three protein templates, docking-generated poses, and two empirical per-PDB penalties. The two penalties are the most concerning free parameters because they are applied before validation on the same experimental values used to report the RMSE. The ML model introduces no new physical entities.

free parameters (5)
  • PDB-specific structural reorganization penalty for 1ERE = +8.69 kcal/mol
    Added to AB-FEP energies for complexes built on 1ERE. Appears calibrated against the 67 experimental values used for validation, though the fitting procedure is not fully described in the paper.
  • PDB-specific structural reorganization penalty for 3ERT = +6.11 kcal/mol
    Added to AB-FEP energies for complexes built on 3ERT. Appears calibrated against the same 67 experimental values used for validation.
  • Affinity capping threshold = -3.0 kcal/mol
    Labels with values above -3.0 kcal/mol are capped to -3.0 before training and inference; this is a hand-chosen cutoff based on the assumption that values above it indicate non-binders.
  • DSM loss weight lambda = 2
    Weight balancing MSE and DSM losses in Eq. (7); chosen by the authors and directly affects DualBind's reported performance.
  • Noise scale sigma sampling range = [0.1, 1]
    Uniformly sampled noise scale for ligand atom coordinates in Eq. (2); chosen by hand and affects the DSM training objective.
assumptions (5)
  • domain assumption 1 ns AB-FEP simulations converge to accurate absolute binding free energies.
    Invoked in Section 3 'Dataset Generation' for all 8,770 complexes; no convergence analysis is provided.
  • domain assumption The 67 manually curated ChEMBL experimental affinities are reliable ground truth.
    Used in Section 3 'Agreement of Calculations with Experimental Data' to validate AB-FEP labels and to calibrate per-PDB penalties.
  • ad hoc to paper Selecting the lowest (most favorable) AB-FEP energy across sampled conformations after penalties yields the representative binding affinity.
    Section 3 'Agreement of Calculations with Experimental Data'; this is a post-processing choice, not a derived result.
  • domain assumption Three PDB templates (1ERE, 3ERT, 1SJ0) represent the relevant ERα conformational states for all 699 ligands.
    Section 3 'Dataset Generation'; poses are generated against these templates only.
  • domain assumption Docking and IFD-MD poses are near-native and adequate for AB-FEP.
    Section 3 'Dataset Generation'; no pose quality assessment is reported.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ToxBench: A Binding Affinity Prediction Benchmark with AB-FEP-Calculated Labels for Human Estrogen Receptor Alpha." pith.science (2026). https://pith.science/paper/JJFIMPFE

@misc{pith2026250708966,
  author       = {Pith},
  title        = {Pith review of: ToxBench: A Binding Affinity Prediction Benchmark with AB-FEP-Calculated Labels for Human Estrogen Receptor Alpha},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JJFIMPFE}},
  note         = {Machine review of arXiv:2507.08966}
}
abstract

Protein-ligand binding affinity prediction is essential for drug discovery and toxicity assessment. While machine learning (ML) promises fast and accurate predictions, its progress is constrained by the availability of reliable data. In contrast, physics-based methods such as absolute binding free energy perturbation (AB-FEP) deliver high accuracy but are computationally prohibitive for high-throughput applications. To bridge this gap, we introduce ToxBench, the first large-scale AB-FEP dataset designed for ML development and focused on a single pharmaceutically critical target, Human Estrogen Receptor Alpha (ER$\alpha$). ToxBench contains 8,770 ER$\alpha$-ligand complex structures with binding free energies computed via AB-FEP with a subset validated against experimental affinities at 1.75 kcal/mol RMSE, along with non-overlapping ligand splits to assess model generalizability. Using ToxBench, we further benchmark state-of-the-art ML methods, and notably, our proposed DualBind model, which employs a dual-loss framework to effectively learn the binding energy function. The benchmark results demonstrate the superior performance of DualBind and the potential of ML to approximate AB-FEP at a fraction of the computational cost.

Figures

Figures reproduced from arXiv: 2507.08966 by the authors.

Figure 1
Figure 1. An illustration of the ToxBench task. ML models, such as DualBind, are trained on high-fidelity AB-FEP data to predict protein-ligand binding affinities several orders of magnitude faster than original AB-FEP calculations. to reliably estimate how tightly a small molecule binds to its target protein enables the prioritization of drug candi￾dates and the early identification of off-target interactions that could lead… view at source ↗
Figure 2
Figure 2. AB-FEP calculations vs. experimental affinities. This Compares AB-FEP calculated binding affinities with the corre￾sponding 67 curated experimental binding affinities. with reported assay values was obtained from ChEMBL. From this collection, 67 experimental binding affinity val￾ues were manually curated and standardized to units of kcal/mol. These specific values were prioritized due to their high data quality and … view at source ↗
Figure 3
Figure 3. Distribution of the AB-FEP calculated binding affin￾ity labels in ToxBench. These labels, representing binding free en￾ergies for the 8,770 ERα-ligand complexes, are in units of kcal/mol and span a range of approximately -26 to +9 kcal/mol. binding affinity towards ERα for each of the 67 manually verified experimental values, the ensemble of AB-FEP cal￾culations was systematically processed. An empirical, PDB￾specif… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: An illustration of the DualBind approach. DualBind employs a dual-loss framework that combines the MSE loss LMSE and the DSM loss LDSM. Specifically, LMSE anchors the predicted binding affinity of the original structure to its groundtruth label. Concurrently, LDSM shap…
Figure 5
Figure 5. Figure 5: Performance comparison on the ToxBench benchmark. The evaluation metrics include Rp (Pearson correlation coefficient), R 2 (coefficient of determination), ρ (Spearman’s rank correlation coefficient), and RMSE (root mean square error). ↑ (↓) represents that a higher (lo…
Figure 6
Figure 6. Figure 6: DualBind predictions vs. ground truth. This compares DualBind’s predicted binding affinities from the best performing of three runs against the corresponding ground truth values on the ToxBench test set. 6. Conclusion and Outlook In this work, we introduce ToxBench, a …

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

81 extracted references · 65 canonical work pages

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION format.date year duplicate empty "emp...

  2. [2]

    J., Knapp, S., and Biggin, P

    Aldeghi, M., Heifetz, A., Bodkin, M. J., Knapp, S., and Biggin, P. C. Accurate calculation of the absolute free energy of binding for drug molecules. Chemical science, 7 0 (1): 0 207--218, 2016

    work page 2016

  3. [3]

    The protein data bank

    Berman, H., Westbrook, J., Feng, Z., Gilliland, G., Bhat, T., Weissig, H., Shindyalov, I., and Bourne, P. The protein data bank. Nucleic Acids Research, 28 0 (1): 0 235--242, 2000. doi:10.1093/nar/28.1.235. URL https://doi.org/10.1093/nar/28.1.235

    doi:10.1093/nar/28.1.235

  4. [4]

    J., Ganichkin, O., Gendelev, L., Harris, S

    Beroza, P., Crawford, J. J., Ganichkin, O., Gendelev, L., Harris, S. F., Klein, R., Miu, A., Steinbacher, S., Klingler, F.-M., and Lemmen, C. Chemical space docking enables large-scale structure-based virtual screening to discover rock1 kinase inhibitors. Nature Communications, 13 0 (1): 0 6447, 2022

    work page 2022

  5. [5]

    Bishop, C. M. Training with noise is equivalent to tikhonov regularization. Neural computation, 7 0 (1): 0 108--116, 1995

    work page 1995

  6. [6]

    The development of a simple empirical scoring function to estimate the binding constant for a protein-ligand complex of known three-dimensional structure

    B \"o hm, H.-J. The development of a simple empirical scoring function to estimate the binding constant for a protein-ligand complex of known three-dimensional structure. Journal of computer-aided molecular design, 8: 0 243--256, 1994

    work page 1994

  7. [7]

    Prediction of binding constants of protein ligands: a fast method for the prioritization of hits obtained from de novo design or 3d database search programs

    B \"o hm, H.-J. Prediction of binding constants of protein ligands: a fast method for the prioritization of hits obtained from de novo design or 3d database search programs. Journal of computer-aided molecular design, 12: 0 309--309, 1998

    work page 1998

  8. [8]

    Absolute binding free energies: a quantitative approach for their calculation

    Boresch, S., Tettinger, F., Leitgeb, M., and Karplus, M. Absolute binding free energies: a quantitative approach for their calculation. The Journal of Physical Chemistry B, 107 0 (35): 0 9535--9551, 2003

    work page 2003

Show all 81 references
  1. [9]

    How attentive are graph attention networks? In International Conference on Learning Representations, 2022

    Brody, S., Alon, U., and Yahav, E. How attentive are graph attention networks? In International Conference on Learning Representations, 2022

    2022

  2. [10]

    The rise of deep learning in drug discovery

    Chen, H., Engkvist, O., Wang, Y., Olivecrona, M., and Blaschke, T. The rise of deep learning in drug discovery. Drug discovery today, 23 0 (6): 0 1241--1250, 2018

    2018

  3. [11]

    A., and Wang, L

    Chen, W., Cui, D., Jerome, S., Michino, M., Lenselink, E., Huggins, D., Beautrait, A., Vendome, A., Abel, R., Friesner, R. A., and Wang, L. Enhancing hit discovery in virtual screening through absolute protein--ligand binding free-energy calculations. J. Chem. Inf. Model., 63 ...

    2023

  4. [12]

    I., Hunt, J

    Davis, M. I., Hunt, J. P., Herrgard, S., Ciceri, P., Wodicka, L. M., Pallares, G., Hocker, M., Treiber, D. K., and Zarrinkar, P. P. Comprehensive analysis of kinase inhibitor selectivity. Nature Biotechnology, 29 0 (11): 0 1046--1051, Nov 2011. doi:10.1038/nbt.1990. URL https:...

    2011 doi

  5. [13]

    Durant, G., Boyles, F., Birchall, K., Marsden, B., and Deane, C. M. Robustly interrogating machine learning-based scoring functions: what are they learning? Bioinformatics, 41 0 (2): 0 btaf040, 2025

    2025

  6. [14]

    D., Murray, C

    Eldridge, M. D., Murray, C. W., Auton, T. R., Paolini, G. V., and Mee, R. P. Empirical scoring functions: I. the development of a fast empirical scoring function to estimate the binding affinity of ligands in receptor complexes. Journal of computer-aided molecular design, 11: ...

    1997

  7. [15]

    Falcon, W. A. Pytorch lightning. GitHub, 3, 2019

    2019

  8. [16]

    K., Given, J

    Gilson, M. K., Given, J. A., Bush, B. L., and McCammon, J. A. The statistical-thermodynamic basis for computation of binding affinities: a critical review. Biophysical journal, 72 0 (3): 0 1047--1069, 1997

    1997

  9. [17]

    Simple gnn regularisation for 3d molecular property prediction & beyond

    Godwin, J., Schaarschmidt, M., Gaunt, A., Sanchez-Gonzalez, A., Rubanova, Y., Veli c kovi \'c , P., Kirkpatrick, J., and Battaglia, P. Simple gnn regularisation for 3d molecular property prediction & beyond. arXiv preprint arXiv:2106.07971, 2021

    2021 arXiv

  10. [18]

    and Case, D

    Gohlke, H. and Case, D. A. Converging free energy estimates: MM-PB (GB) SA studies on the protein--protein complex Ras--Raf . Journal of computational chemistry, 25 0 (2): 0 238--250, 2004

    2004

  11. [19]

    Predicting binding modes, binding affinities andhot spots' for protein-ligand complexes using a knowledge-based scoring function

    Gohlke, H., Hendlich, M., and Klebe, G. Predicting binding modes, binding affinities andhot spots' for protein-ligand complexes using a knowledge-based scoring function. Perspectives in Drug Discovery and Design, 20: 0 115--144, 2000

    2000

  12. [20]

    Gohlke, H., Kiel, C., and Case, D. A. Insights into protein--protein binding by binding free energy calculation and free energy decomposition for the Ras--Raf and Ras--RalGDS complexes. Journal of molecular biology, 330 0 (4): 0 891--913, 2003

    2003

  13. [21]

    A., Murphy, R

    Halgren, T. A., Murphy, R. B., Friesner, R. A., Beard, H. S., Frye, L. L., Pollard, W. T., and Banks, J. L. Glide: A new approach for rapid, accurate docking and scoring. 2. Enrichment factors in database screening. J. Med. Chem., 47: 0 1750--1759, 2004

    2004

  14. [22]

    D., Smythe, M

    Head, R. D., Smythe, M. L., Oprea, T. I., Waller, C. L., Green, S. M., and Marshall, G. R. Validate: A new method for the receptor-based prediction of binding affinities of novel ligands. Journal of the American Chemical society, 118 0 (16): 0 3959--3969, 1996

    1996

  15. [23]

    P., Chung, Y., Li, S.-C., Graff, D

    Heid, E., Greenman, K. P., Chung, Y., Li, S.-C., Graff, D. E., Vermeire, F. H., Wu, H., Green, W. H., and McGill, C. J. Chemprop: a machine learning package for chemical property prediction. Journal of Chemical Information and Modeling, 64 0 (1): 0 9--17, 2023

    2023

  16. [24]

    Denoising diffusion probabilistic models

    Ho, J., Jain, A., and Abbeel, P. Denoising diffusion probabilistic models. Advances in neural information processing systems, 33: 0 6840--6851, 2020

    2020

  17. [25]

    and Gohlke, H

    Homeyer, N. and Gohlke, H. Free energy calculations by the molecular mechanics poisson- boltzmann surface area method. Molecular informatics, 31 0 (2): 0 114--122, 2012

    2012

  18. [26]

    InteractionGraphNet : a novel and efficient deep graph representation learning framework for accurate protein--ligand interaction predictions

    Jiang, D., Hsieh, C.-Y., Wu, Z., Kang, Y., Wang, J., Wang, E., Liao, B., Shen, C., Xu, L., Wu, J., et al. InteractionGraphNet : a novel and efficient deep graph representation learning framework for accurate protein--ligand interaction predictions. Journal of medicinal chemist...

    2021

  19. [27]

    K deep: protein--ligand absolute binding affinity prediction via 3d-convolutional neural networks

    Jim \'e nez, J., Skalic, M., Martinez-Rosell, G., and De Fabritiis, G. K deep: protein--ligand absolute binding affinity prediction via 3d-convolutional neural networks. Journal of chemical information and modeling, 58 0 (2): 0 287--296, 2018

    2018

  20. [28]

    DSMBind : Se (3) denoising score matching for unsupervised binding energy prediction and nanobody design

    Jin, W., Chen, X., Vetticaden, A., Sarzikova, S., Raychowdhury, R., Uhler, C., and Hacohen, N. DSMBind : Se (3) denoising score matching for unsupervised binding energy prediction and nanobody design. bioRxiv, pp.\ 2023--12, 2023 a

    2023

  21. [29]

    Unsupervised protein-ligand binding energy prediction via neural euler's rotation equation

    Jin, W., Sarkizova, S., Chen, X., Hacohen, N., and Uhler, C. Unsupervised protein-ligand binding energy prediction via neural euler's rotation equation. Advances in Neural Information Processing Systems, 36, 2023 b

    2023

  22. [30]

    C., Yao, K., Kaplan, Z., Chelliah, M., Leswing, K., Seekins, S., Watts, S., Calkins, D., Chief Elk, J., Jerome, S

    Johnston, R. C., Yao, K., Kaplan, Z., Chelliah, M., Leswing, K., Seekins, S., Watts, S., Calkins, D., Chief Elk, J., Jerome, S. V., Repasky, M. P., and Shelley, J. C. Epik: pka and protonation state prediction through machine learning. J. Chem. Theory Comput., 19: 0 2380--2388, 2023

    2023

  23. [31]

    L., Buckner, J

    Jorgensen, W. L., Buckner, J. K., Boudon, S., and Tirado-Rives, J. Efficient computation of absolute free energies of binding by computer simulations. application to the methane dimer in water. The Journal of chemical physics, 89 0 (6): 0 3742--3746, 1988

    1988

  24. [32]

    Kingma, D. P. and Ba, J. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980, 2014

    2014 arXiv

  25. [33]

    A., Massova, I., Reyes, C., Kuhn, B., Huo, S., Chong, L., Lee, M., Lee, T., Duan, Y., Wang, W., et al

    Kollman, P. A., Massova, I., Reyes, C., Kuhn, B., Huo, S., Chong, L., Lee, M., Lee, T., Duan, Y., Wang, W., et al. Calculating structures and free energies of complex molecules: combining molecular mechanics and continuum models. Accounts of chemical research, 33 0 (12): 0 889...

    2000

  26. [34]

    Flag: Adversarial data augmentation for graph neural networks

    Kong, K., Li, G., Ding, M., Wu, Z., Zhu, C., Ghanem, B., Taylor, G., and Goldstein, T. Flag: Adversarial data augmentation for graph neural networks. 2020

    2020

  27. [35]

    A., Vandenberg, L

    La Merrill, M. A., Vandenberg, L. N., Smith, M. T., Goodson, W., Browne, P., Patisaul, H. B., Guyton, K. Z., Kortenkamp, A., Cogliano, V. J., Woodruff, T. J., et al. Consensus on the key characteristics of endocrine-disrupting chemicals as a basis for hazard identification. Na...

    2020

  28. [36]

    A tutorial on energy-based learning

    LeCun, Y., Chopra, S., and Hadsell, R. A tutorial on energy-based learning. 2006

    2006

  29. [37]

    C., and Choi, K.-C

    Lee, H.-R., Jeung, E.-B., Cho, M.-H., Kim, T.-H., Leung, P. C., and Choi, K.-C. Molecular mechanism (s) of endocrine-disrupting chemicals and their potent oestrogenicity in diverse cells and tissues that express oestrogen receptors. Journal of cellular and molecular medicine, ...

    2013

  30. [38]

    When attention meets fast recurrence: Training language models with reduced compute

    Lei, T. When attention meets fast recurrence: Training language models with reduced compute. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pp.\ 7633--7648, 2021

    2021

  31. [39]

    Leak proof PDBBind : A reorganized dataset of protein-ligand complexes for more generalizable binding affinity prediction

    Li, J., Guan, X., Zhang, O., Sun, K., Wang, Y., Bagni, D., and Head-Gordon, T. Leak proof PDBBind : A reorganized dataset of protein-ligand complexes for more generalizable binding affinity prediction. ArXiv, pp.\ arXiv--2308, 2024

    2024

  32. [40]

    T., et al

    Liu, X., Jiang, S., Duan, X., Vasan, A., Liu, C., Tien, C.-c., Ma, H., Brettin, T., Xia, F., Foster, I. T., et al. Binding affinity prediction: From conventional to machine learning-based approaches. arXiv preprint arXiv:2410.00709, 2024

  33. [41]

    Forging the basis for developing protein--ligand interaction scoring functions

    Liu, Z., Su, M., Han, L., Liu, J., Yang, Q., Li, Y., and Wang, R. Forging the basis for developing protein--ligand interaction scoring functions. Accounts of chemical research, 50 0 (2): 0 302--309, 2017

    2017

  34. [42]

    Tankbind: Trigonometry-aware neural networks for drug-protein binding structure prediction

    Lu, W., Wu, Q., Zhang, J., Rao, J., Li, C., and Zheng, S. Tankbind: Trigonometry-aware neural networks for drug-protein binding structure prediction. Advances in neural information processing systems, 35: 0 7236--7249, 2022

    2022

  35. [43]

    M., and Biggin, P

    Meli, R., Morris, G. M., and Biggin, P. C. Scoring functions for protein-ligand binding affinity prediction using structure-based deep learning: a review. Frontiers in bioinformatics, 2: 0 885983, 2022

    2022

  36. [44]

    Molecular docking: a powerful approach for structure-based drug discovery

    Meng, X.-Y., Zhang, H.-X., Mezei, M., and Cui, M. Molecular docking: a powerful approach for structure-based drug discovery. Current computer-aided drug design, 7 0 (2): 0 146--157, 2011

    2011

  37. [45]

    B., Murphy, R

    Miller, E. B., Murphy, R. B., Sindhikara, D., Borrelli, K. W., Grisewood, M. J., Ranalli, F., Dixon, S. L., Jerome, S., Boyles, N. A., Day, T., Ghanakota, P., Mondal, S., Rafi, S. B., Troast, D. M., Abel, R., and Friesner, R. A. Reliable and accurate solution to the induced fi...

    2021 doi

  38. [46]

    Estrogen receptor signaling in breast cancer

    Miziak, P., Baran, M., B aszczak, E., Przybyszewska-Podstawka, A., Ka afut, J., Smok-Kalwat, J., Dmoszy \'n ska-Graniczka, M., Kie bus, M., and Stepulak, A. Estrogen receptor signaling in breast cancer. Cancers, 15 0 (19): 0 4689, 2023

    2023

  39. [47]

    Moon, S., Zhung, W., Yang, S., Lim, J., and Kim, W. Y. PIGNet : a physics-informed deep learning model toward generalized drug--target interaction predictions. Chemical Science, 13 0 (13): 0 3661--3673, 2022

    2022

  40. [48]

    B., Repasky, M

    Murphy, R. B., Repasky, M. P., Greenwood, J. R., Tubert-Brohman, I., Jerome, S. S., Annabhimoju, R., Boyles, N. A., Schmitz, C. D., Abel, R., Farid, R., and Friesner, R. A. WScore: A flexible and accurate treatment of explicit water molecules in ligand--receptor docking . J. M...

    2016 doi

  41. [49]

    M., Carchia, M., Irwin, J

    Mysinger, M. M., Carchia, M., Irwin, J. J., and Shoichet, B. K. Directory of useful decoys, enhanced (dud-e): Better ligands and decoys for better benchmarking. Journal of Medicinal Chemistry, 55 0 (14): 0 6582--6594, jul 2012. doi:10.1021/jm300687e. URL https://doi.org/10.102...

    2012 doi

  42. [50]

    P., Nguyen, T., Le, T

    Nguyen, T., Le, H., Quinn, T. P., Nguyen, T., Le, T. D., and Venkatesh, S. GraphDTA : predicting drug--target binding affinity with graph neural networks. Bioinformatics, 37 0 (8): 0 1140--1147, 2021

    2021

  43. [51]

    O zt \"u rk, H., \

    \"O zt \"u rk, H., \"O zg \"u r, A., and Ozkirimli, E. DeepDTA : deep drug--target binding affinity prediction. Bioinformatics, 34 0 (17): 0 i821--i829, 2018

    2018

  44. [52]

    S., Syed, K., and Tuszynski, J

    Pagadala, N. S., Syed, K., and Tuszynski, J. Software for molecular docking: a review. Biophysical reviews, 9 0 (2): 0 91--102, 2017

    2017

  45. [53]

    Automatic differentiation in pytorch

    Paszke, A., Gross, S., Chintala, S., Chanan, G., Yang, E., DeVito, Z., Lin, Z., Desmaison, A., Antiga, L., and Lerer, A. Automatic differentiation in pytorch. 2017

    2017

  46. [54]

    J., Misra, I., Grover, A., Ben-Hamu, H., and Lipman, Y

    Puny, O., Atzmon, M., Smith, E. J., Misra, I., Grover, A., Ben-Hamu, H., and Lipman, Y. Frame averaging for invariant and equivariant network design. In International Conference on Learning Representations, 2021

    2021

  47. [55]

    The maximal and current accuracy of rigorous protein-ligand binding free energy calculations

    Ross, G., Lu, C., Scarabelli, G., and others . The maximal and current accuracy of rigorous protein-ligand binding free energy calculations. Communications Chemistry, 6 0 (1): 0 222, oct 2023. doi:10.1038/s42004-023-01019-9. URL https://doi.org/10.1038/s42004-023-01019-9

    2023 doi

  48. [56]

    Sadybekov, A. V. and Katritch, V. Computational approaches streamlining drug discovery. Nature, 616 0 (7958): 0 673--685, 2023

    2023

  49. [57]

    M., Adzhigirey, M., Day, T., Annabhimoju, R., and Sherman, W

    Sastry, G. M., Adzhigirey, M., Day, T., Annabhimoju, R., and Sherman, W. Protein and ligand preparation: Parameters, protocols, and influence on virtual screening enrichments. J. Comput. Aid. Mol. Des., 27 0 (3): 0 221--234, 2013. doi:10.1007/s10822-013-9644-8

    2013 doi

  50. [58]

    E., Turnbull, O

    Scantlebury, J., Vost, L., Carbery, A., Hadfield, T. E., Turnbull, O. M., Brown, N., Chenthamarakshan, V., Das, P., Grosjean, H., Von Delft, F., et al. A small step toward generalizability: training a machine learning scoring function for structure-based virtual screening. Jou...

    2023

  51. [59]

    o dinger Release 2021-3: FEP+ . Schr \

    Schr \"o dinger Release 2021-3: FEP+ . Schr \"o dinger, LLC, New York, NY, 2021. Software release

    2021

  52. [60]

    Schrödinger Release 2024-4: LigPrep , 2024

    Schrödinger, LLC . Schrödinger Release 2024-4: LigPrep , 2024

    2024

  53. [61]

    S., Isayev, O., and Roitberg, A

    Smith, J. S., Isayev, O., and Roitberg, A. E. ANI-1 : an extensible neural network potential with DFT accuracy at force field computational cost. Chemical science, 8 0 (4): 0 3192--3203, 2017

    2017

  54. [62]

    and Ermon, S

    Song, Y. and Ermon, S. Generative modeling by estimating gradients of the data distribution. Advances in neural information processing systems, 32, 2019

    2019

  55. [63]

    and Kingma, D

    Song, Y. and Kingma, D. P. How to train your energy-based models. arXiv preprint arXiv:2101.03288, 2021

    2021 arXiv

  56. [64]

    Maximum likelihood training of score-based diffusion models

    Song, Y., Durkan, C., Murray, I., and Ermon, S. Maximum likelihood training of score-based diffusion models. Advances in neural information processing systems, 34: 0 1415--1428, 2021

    2021

  57. [65]

    C., Tempczyk, A., Hawley, R

    Still, W. C., Tempczyk, A., Hawley, R. C., and Hendrickson, T. Semianalytical treatment of solvation for molecular mechanics and dynamics. Journal of the American Chemical Society, 112 0 (16): 0 6127--6129, 1990

    1990

  58. [66]

    Comparative assessment of scoring functions: the CASF-2016 update

    Su, M., Yang, Q., Du, Y., Feng, G., Liu, Z., Li, Y., and Wang, R. Comparative assessment of scoring functions: the CASF-2016 update. Journal of chemical information and modeling, 59 0 (2): 0 895--913, 2018

    2016

  59. [67]

    Applications of machine learning in drug discovery and development

    Vamathevan, J., Clark, D., Czodrowski, P., Dunham, I., Ferran, E., Lee, G., Li, B., Madabhushi, A., Shah, P., Spitzer, M., et al. Applications of machine learning in drug discovery and development. Nature reviews Drug discovery, 18 0 (6): 0 463--477, 2019

    2019

  60. [68]

    N., Kaiser, ., and Polosukhin, I

    Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, ., and Polosukhin, I. Attention is all you need. Advances in neural information processing systems, 30, 2017

    2017

  61. [69]

    A connection between score matching and denoising autoencoders

    Vincent, P. A connection between score matching and denoising autoencoders. Neural computation, 23 0 (7): 0 1661--1674, 2011

    2011

  62. [70]

    Extracting and composing robust features with denoising autoencoders

    Vincent, P., Larochelle, H., Bengio, Y., and Manzagol, P.-A. Extracting and composing robust features with denoising autoencoders. In Proceedings of the 25th international conference on Machine learning, pp.\ 1096--1103, 2008

    2008

  63. [71]

    On the frustration to predict binding affinities from protein--ligand structures with deep neural networks

    Volkov, M., Turk, J.-A., Drizard, N., Martin, N., Hoffmann, B., Gaston-Math \'e , Y., and Rognan, D. On the frustration to predict binding affinities from protein--ligand structures with deep neural networks. Journal of medicinal chemistry, 65 0 (11): 0 7946--7958, 2022

    2022

  64. [72]

    and Dokholyan, N

    Wang, J. and Dokholyan, N. V. Yuel: improving the generalizability of structure-free compound--protein interaction prediction. Journal of chemical information and modeling, 62 0 (3): 0 463--471, 2022

    2022

  65. [73]

    Further development and validation of empirical scoring functions for structure-based binding affinity prediction

    Wang, R., Lai, L., and Wang, S. Further development and validation of empirical scoring functions for structure-based binding affinity prediction. Journal of computer-aided molecular design, 16: 0 11--26, 2002

    2002

  66. [74]

    The PDBbind database : Collection of binding affinities for protein- ligand complexes with known three-dimensional structures

    Wang, R., Fang, X., Lu, Y., and Wang, S. The PDBbind database : Collection of binding affinities for protein- ligand complexes with known three-dimensional structures. Journal of medicinal chemistry, 47 0 (12): 0 2977--2980, 2004

    2004

  67. [75]

    How to make machine learning scoring functions competitive with fep

    Warren, M., Deane, C., Magarkar, A., Morris, G., Biggin, P., et al. How to make machine learning scoring functions competitive with fep. 2024

    2024

  68. [76]

    Predicting or pretending: artificial intelligence for protein-ligand interactions lack of sufficiently large and unbiased datasets

    Yang, J., Shen, C., and Huang, N. Predicting or pretending: artificial intelligence for protein-ligand interactions lack of sufficiently large and unbiased datasets. Frontiers in pharmacology, 11: 0 69, 2020

    2020

  69. [77]

    Analyzing learned molecular representations for property prediction

    Yang, K., Swanson, K., Jin, W., Coley, C., Eiden, P., Gao, H., Guzman-Perez, A., Hopper, T., Kelley, B., Mathea, M., et al. Analyzing learned molecular representations for property prediction. Journal of chemical information and modeling, 59 0 (8): 0 3370--3388, 2019 a

    2019

  70. [78]

    Concepts of artificial intelligence for computer-assisted drug discovery

    Yang, X., Wang, Y., Byrne, R., Schneider, G., and Yang, S. Concepts of artificial intelligence for computer-assisted drug discovery. Chemical reviews, 119 0 (18): 0 10520--10594, 2019 b

    2019

  71. [79]

    Yuan, W., Chen, G., and Chen, C. Y.-C. FusionDTA : attention-based feature polymerizer and knowledge distillation for drug-target binding affinity prediction. Briefings in Bioinformatics, 23 0 (1): 0 bbab506, 2022

    2022

  72. [80]

    W., Sanchez-Gonzalez, A., Battaglia, P., Pascanu, R., and Godwin, J

    Zaidi, S., Schaarschmidt, M., Martens, J., Kim, H., Teh, Y. W., Sanchez-Gonzalez, A., Battaglia, P., Pascanu, R., and Godwin, J. Pre-training via denoising for molecular property prediction. In The Eleventh International Conference on Learning Representations, 2022

    2022

  73. [81]

    J., Blackshaw, J., Corbett, S., de Veij, M., Ioannidis, H., Lopez, D

    Zdrazil, B., Felix, E., Hunter, F., Manners, E. J., Blackshaw, J., Corbett, S., de Veij, M., Ioannidis, H., Lopez, D. M., Mosquera, J. F., Magarinos, M. P., Bosc, N., Arcila, R., Kizilören, T., Gaulton, A., Bento, A. P., Adasme, M. F., Monecke, P., Landrum, G. A., and Leach, A...

    2023 doi

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.