Pith. sign in

REVIEW 4 major objections 5 minor 37 references

Active Learning on Synthons for Molecular Design

T0 review · 4 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read By scoring fragments instead of whole molecules, active learning screens trillion-compound spaces and finds 95% of top hits early.

desk verdict Factored synthon-level active learning is a genuine extension of pool-based AL to non-enumerable spaces, with an honest 1M ground-truth benchmark; the main gap is the narrowly tested independence assumption. read the letter →

arxiv 2505.12913 v1 pith:XPB7D2MK submitted 2025-05-19 cs.LG q-bio.QM

Tom George Grigg , Mason Burlage , Oliver Brook Scott , Adam Taouil , Dominique Sydow , Liam Wilbraham This is my paper · ORCID
classification cs.LGq-bio.QM
keywords activelearningsynthonmulti-vectorexpansionvirtualscreeningThompsonsamplingultra-largechemicalspacesmoleculardesignmulti-parameteroptimization
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tries to extend pool-based active learning to chemical libraries too large to enumerate. Its proposal is to model and acquire one fragment at each reaction site, treating each site as an independent bandit, so the cost of ranking a space grows with the number of fragments rather than the number of molecules. On a benchmark of one million molecules, this factored approach recovers about 95% of the best thousand compounds after scoring a small fraction of the space, and it scales to roughly two trillion compounds. The authors also show that molecules selected this way match or exceed a generative baseline on multi-objective design tasks for three protein targets while producing more diverse scaffolds.

What carries the argument

The load-bearing object is the factored synthon-level surrogate model with independent acquisition. A synthon is a fragment or R-group attached at one vector position; the paper treats each position as a separate multi-armed bandit whose arms are synthons. A message-passing neural network predicts a mean and variance for each synthon's contribution to the molecular score, acquisition samples from those Gaussian predictions via Thompson sampling, and the top synthons across vectors are combined into candidate molecules. This factorization, reducing a product of pool sizes to a sum, is what makes non-enumerable spaces tractable; the roughly additive structure of shape-based scores is what makes the factorization accurate enough to rank well.

What would settle it

Take a small enumerated space and define a score that is the sum of per-synthon terms plus a large cooperation bonus when two particular synthons co-occur; run SALSA on this objective and check whether recall of the true top-1,000 combinations falls far below the 95% observed on additive objectives.

Watch

Extended reading notes

Core claim

The central claim is that multi-vector molecular expansion can be screened without enumerating the product space. The algorithm attributes each scored molecule's objective value to its constituent synthons, trains a surrogate per vector, and uses Thompson sampling to pick synthons whose predicted score distributions look promising; assembling those synthons yields the next batch of molecules. This changes inference-time complexity from the product of synthon-pool sizes to their sum, so spaces with trillions of combinations become practical. In an enumerated one-million-molecule space the method identifies 96.5% of the top-1,000 compounds for a shape-based objective and 94.5% for docking, close to full-molecular active learning, and its top-scoring molecules improve as the space is enlarged to roughly 910K by 2.4M synthons, about two trillion molecules. For multi-parameter objectives combining docking or shape similarity with drug-likeness, the method produces molecules with drug-like property distributions and greater scaffold diversity than a generative baseline.

Load-bearing premise

The method assumes a molecule's objective score is close to the sum of independent contributions from its R-groups, so that a model trained on individual synthons can rank full molecules without ever seeing them together.

Editorial extensions

If this is right

  • Ultra-large make-on-demand libraries, with billions to trillions of compounds, become accessible to active learning with expensive scoring functions such as docking.
  • Every candidate molecule comes with an explicit synthesis route, because each synthon is tied to a reaction; this eases the design-to-make transition.
  • Practitioners can steer the search by filtering synthon sets for physicochemical properties or practical procurement concerns before screening begins.
  • Factored acquisition degrades minimally relative to full-molecular active learning on objectives that are roughly additive over fragments, and can learn faster in early rounds.
  • On multi-parameter objectives, SALSA matches or improves over a generative approach while yielding substantially more unique high-scoring scaffolds.

Reading between the lines

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

  • If objective scores contain strong interactions between specific R-groups, such as cooperative binding or steric clashes between two substituents, the independent-synthon surrogate can misrank combinations; testing SALSA on a deliberately coupled objective would quantify this failure mode.
  • Because acquisition is linear in the number of synthons, the same factoring could extend to more than two vectors or to scaffold hopping by treating the core as an additional synthon vector.
  • The per-synthon score distributions learned by SALSA could be reused to estimate full-library score distributions, guiding which sub-libraries to enumerate rather than only which molecules to sample.
  • The method's success on shape-based scores suggests fragment-additive objectives are the natural first targets; structure-based scores that depend on global geometry may need joint models to reach the same efficiency.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper introduces SALSA (Scalable Active Learning via Synthon Acquisition), an active learning method for multi-vector (R-group) combinatorial molecular spaces. SALSA trains a separate surrogate model per synthon vector on scores of complete molecules, decomposing each molecule's objective score into per-synthon contributions, and acquires new molecules by combining independently top-scored synthons. The authors report that SALSA identifies approximately 95% of the top-1K molecules in a fully enumerated 1M-molecule CDK2 space with a small scoring budget, that it scales to spaces of approximately 2 trillion molecules, and that it achieves higher scores and greater scaffold diversity than the LibINVENT generative baseline on linear multi-parameter objectives for CDK2, BACE1, and DRD2. The paper is published at the GEM workshop, ICLR 2025, and includes appendices with algorithm pseudocode, hyperparameters, ablations, and runtime tables.

Significance. If the reported results hold, SALSA offers a practical way to perform active learning over non-enumerable combinatorial spaces with expensive molecular objectives, which is a real bottleneck in early-stage drug discovery. The paper's strongest evidence is the honest ground-truth evaluation in a fully enumerated 1M-molecule space, with comparisons to a full-molecular active learning baseline (MolPAL) and tabular Thompson sampling, plus ablations of surrogate models and acquisition strategies. The method also provides explicit synthetic routes for proposed molecules, an advantage over many generative approaches. The principal risk is that the method's correctness rests on an additivity assumption across synthon vectors that is tested on only two objective families and acknowledged by the authors as fragile for more complex objective functions.

major comments (4)
  1. [Section 2 (Surrogate models) and Section 3 (Sample efficiency; Multi-parameter objectives)] The central claim that SALSA generalizes to arbitrary objective functions rests on the assumption that f(s0,s1) ≈ g0(s0)+g1(s1), but the paper validates this on only two objective families—ROCS-TC and Hybrid Docking—and the MPO experiments are linear combinations of these same scores with QED. Appendix A.4 itself attributes the aleatoric variance of a synthon's score to the unobserved complementary synthon, confirming that interaction effects exist and are treated as noise. Please test SALSA on an objective with explicit R-group interactions (for example, an MPO containing a product or cross-term of the two vector contributions, or an activity model trained on both R-groups jointly) to determine whether the method remains effective when additivity is violated. If the method fails on such a task, the scope of the central claim should be narrowed, and the paper should state the conditions under which SALSA is expected to work.
  2. [Section 3, Sample efficiency (Figure 2)] The headline comparison to MolPAL (96.5% vs 98.5% for ROCS-TC; 94.5% vs 95.4% for docking) is reported as point means over 5 trials with no error bars, confidence intervals, or significance testing. The claimed 'minimal degradation' may be within trial-to-trial noise. Please report the per-trial variation (standard deviation, interquartile range, or confidence intervals) and, ideally, a paired test; if the difference is not statistically significant, state explicitly that the degradation is negligible rather than minimal. The same issue applies to Figure 3 (3 trials) and Figure 7.
  3. [Section 3, Scaling beyond enumerable spaces (Figure 3)] The 2T-space experiments have no ground-truth ranking and no baseline (not even random acquisition), so the statement that 'SALSA consistently finds better scoring molecules with increasing space size' only demonstrates that the scores of the molecules discovered improve as the space grows, which could reflect the larger space containing intrinsically better molecules rather than SALSA's search efficiency. The reference to Lyu et al. (2023) is suggestive, but a comparison to random or frequency-based acquisition in the 10B and 2T spaces would strengthen the scaling claim; without such a control, the scaling results remain uncontrolled and should be described as proof-of-concept rather than validation.
  4. [Appendix A.8 and Figure 8] The ADMET comparison relies on 'internal models' to predict several properties (LogD, hERG, PXR, fraction unbound, CACO2) with no description of these models, their training data, or their accuracy. Since Figure 8 is used to claim that SALSA molecules have 'comparable chemical property profiles to known bioactives,' either report the provenance and validation of these predictors (including error bars or scatter plots against measured values) or restrict the property comparison to metrics computed directly from the molecular graph (MW, TPSA, HBA, HBD, AROM, ROTB, ALERTS).
minor comments (5)
  1. [Section 4, Conclusion] Typo: 'directly combarable' should be 'directly comparable'.
  2. [Figure 2 and text around it] The text refers to 'Fig. 2B', but the figure panels are not lettered. Please add panel labels (a) and (b) for the ROCS-TC and docking rows, or reword the reference.
  3. [Appendix A.1, Algorithm 1] The early-termination condition 'if n<N and ρ≤ρ_max' is not fully consistent with the while-loop pseudo-code that uses an undefined variable 'count'. Please clarify how ρ_max and the sampling rejection threshold interact, and remove the undefined variable.
  4. [Section 3, Multi-parameter objectives] The sentence '10K objective function calls were again budgeted for each of 10 rounds' is ambiguous about the total budget; state explicitly that SALSA uses 100K calls total, matching the LibINVENT allocation.
  5. [Appendix A.7] The docking score normalization is described as dividing by 'a factor of -24', -17, and -24; please clarify whether this means multiplying by -1/24 etc., and state the sign convention (whether lower raw docking scores are better).

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: SALSA's performance is measured against externally computed objective functions and an enumerated ground truth, and the independence assumption is a stated modeling assumption rather than a hidden input.

full rationale

The central derivation chain is empirical. In Section 3, the 1M-molecule space is exhaustively enumerated and scored with ROCS-TC and Hybrid Docking to obtain ground truth; SALSA's recall of the top-1K molecules (96.5% and 94.5%) is compared with random screening, MolPAL, and tabular Thompson sampling. These numbers are not implied by the algorithm's construction: the surrogate models are trained on synthon-score pairs derived from externally scored molecules (Section 2, Surrogate models), and acquisition is evaluated against the same external objective functions. The scaling experiments on 100M to 2T spaces lack a ground-truth top set, and the paper's conclusions there are based on scores of SALSA's own top-1K candidates; that is a limitation in evidential strength, not circularity, because the objective scores are still external. The load-bearing additive-decomposition assumption is explicitly acknowledged in the Conclusion ('the implicit, naive independence assumptions that enable SALSA to scale but risk breaking down against more complex objective functions') and in Appendix A.4, where the unobserved complementary synthon is identified as a source of aleatoric variance. This is a modeling assumption that could fail, not a term defined in terms of the predicted quantity. No load-bearing self-citation or imported uniqueness theorem appears; the cited works serve as external baselines or implementation choices. Therefore no reduction of a prediction to its inputs by construction is present.

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

The ledger contains no invented entities. The numeric experimental settings such as rounds, budget, MPO weights, and normalization are hand-chosen protocol parameters rather than fitted scientific constants; they affect the reported numbers but are not fitted to the benchmark outcome. The model parameters of the chemprop MPNN are learned from the objective values at each round, which is the algorithm itself and is not counted here as free parameters. The key domain assumptions are the additive synthon independence assumption, Gaussianity of synthon score distributions, and the definition of the synthesizable space via SMIRKS and Mcule. These are stated in the paper, with the independence assumption explicitly acknowledged as a limitation.

free parameters (4)
  • Samples per round K = 1K or 5K per round in benchmark; 10K per round in scaling and MPO runs
    Hand-chosen evaluation budget; recall and top-1K results are reported at these budgets.
  • Number of rounds N = 10
    Hand-chosen stopping point for all reported experiments; early stopping also uses a rejection threshold.
  • MPO component weights = ROCS-TC+QED weighted 2:1; Docking+QED weighted 1:1
    User-chosen objective weights; different weights would change the top-1K sets and the comparison with LibINVENT.
  • Docking score normalization divisors = -24 (CDK2), -17 (BACE1), -24 (DRD2)
    Per-target hand-selected divisors to scale docking scores roughly to [0,1]; affects MPO values and comparisons.
assumptions (4)
  • domain assumption Molecular objective scores are approximately additive over synthon choices, so per-vector surrogate models and independent Thompson sampling identify near-optimal molecules.
    Invoked by construction of per-vector models and acquisition in Section 2 and Algorithm 1; explicitly flagged in the conclusion as risking breakdown for more complex objectives. This is the main load-bearing assumption.
  • domain assumption For every synthon, the distribution of observed scores is approximately Gaussian with mean and variance predicted by the MPNN.
    The MVE loss in Section 2 is maximum-likelihood estimation of a Gaussian density for synthon-score pairs; Thompson sampling draws from this Gaussian. Heavy-tailed or multimodal score distributions would distort acquisition.
  • domain assumption The SMIRKS reaction set and Mcule building-block database adequately define the synthesizable target space, and pattern matching determines applicable synthons.
    Section 2 and A.2 construct the search space from custom SMIRKS and Mcule; missed reactions or building blocks are excluded from the space and can never be found by SALSA.
  • domain assumption Uncertainty estimates from mean-variance estimation are trustworthy enough for exploration decisions.
    A.4 shows dropout-based uncertainty performs drastically worse and is rejected; the method therefore relies on MVE variances being meaningful for Thompson sampling.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Active Learning on Synthons for Molecular Design." pith.science (2026). https://pith.science/paper/XPB7D2MK

@misc{pith2026250512913,
  author       = {Pith},
  title        = {Pith review of: Active Learning on Synthons for Molecular Design},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XPB7D2MK}},
  note         = {Machine review of arXiv:2505.12913}
}
read the original abstract

Exhaustive virtual screening is highly informative but often intractable against the expensive objective functions involved in modern drug discovery. This problem is exacerbated in combinatorial contexts such as multi-vector expansion, where molecular spaces can quickly become ultra-large. Here, we introduce Scalable Active Learning via Synthon Acquisition (SALSA): a simple algorithm applicable to multi-vector expansion which extends pool-based active learning to non-enumerable spaces by factoring modeling and acquisition over synthon or fragment choices. Through experiments on ligand- and structure-based objectives, we highlight SALSA's sample efficiency, and its ability to scale to spaces of trillions of compounds. Further, we demonstrate application toward multi-parameter objective design tasks on three protein targets - finding SALSA-generated molecules have comparable chemical property profiles to known bioactives, and exhibit greater diversity and higher scores over an industry-leading generative approach.

Figures

Figures reproduced from arXiv: 2505.12913 by the authors.

Figure 1
Figure 1. A Construction of a 2-vector synthon space. B AL loop against scoring function f. 2 METHODS Search space SALSA consumes as input a target molecular space formed by pre-defined choices of synthons or fragments, as well as a molecular objective function f. Fig. 1A exemplifies construc￾tion of a target space for a simple 2-vector expansion scheme on a core with two R-groups. Given a set of SMIRKS-encoded reactions and … view at source ↗
Figure 2
Figure 2. Recall of top-1K compounds in the 1M target space for ROCS-TC (top) and docking (bottom) as a function of molecules acquired, smoothed over 5 trials. The heatmaps show the enu￾merated target space decomposed across synthon axes and coloured by score. For a given SALSA round, synthons are ordered by Monte Carlo-estimated acquisition probability, i.e. the likelihood of sampling increases moving up and right. Top-1K gr… view at source ↗
Figure 3
Figure 3. The large violin plots show the min, max, mean, and estimated score density for the top-1K molecules identified by SALSA as space size increases for shape- (left) and structure-based (right) objectives, smoothed over 3 trials. Subplots on the right show the evolution of the top-1K distribution over AL rounds – the final top-1K molecules are marked in red at their sample index [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: A. Top-1K molecules identified across three targets for SALSA, random acquisition, and LibINVENT using QED (Bickerton et al., 2012) plus ROCS-TC and Hybrid Docking objectives. Mean scores and top-20 pareto optimal molecules are denoted by triangles and stars, respectiv…
Figure 5
Figure 5. Figure 5: Core scaffolds mapped to synthetic intermediates with functionalized reaction handles. Synthon space construction For our experiments, we explicitly construct large multi-vector spaces. Given a core scaffold with R-group handles at desired vectors, we determine applica…
Figure 6
Figure 6. Figure 6: Ablation Acquisition of the top-1K scoring compounds on the 1M-space ROCS-TC design task. Each method was allocated a 1K objective budget per round for 10 rounds. A.5 SALSA WITH ONE MODEL [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]
Figure 7
Figure 7. Figure 7: One vs two-model SALSA for sample efficiency (5 trials) and scaling (3 trials) In [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]
Figure 8
Figure 8. Figure 8: ADMET properties for the top-1K molecules generated by SALSA and LibINVENT compared with the top-1K bioactive molecules from ChEMBL for each protein target, ranked by pChEMBL value. Dashed lines represent thresholds for these properties, and an up or down arrow represe…
Figure 9
Figure 9. Figure 9: Selected top molecules from SALSA MPO runs for all three protein targets. Highlighted atoms and bonds represent fixed cores. QED and MPO score are labeled below each molecule. A.11 MPOS IN CHEMBL FRAGMENT SPACE [PITH_FULL_IMAGE:figures/full_fig_p014_9.png]
Figure 10
Figure 10. Figure 10: Top: the top-1K molecules identified across three targets for SALSA and LibINVENT using QED (Bickerton et al., 2012) plus ROCS-TC and Hybrid Docking objectives. Bottom: the number of unique scaffolds scoring above a given score for molecules enumerated by SALSA and Li…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

37 extracted references · 14 canonical work pages

  1. [1]

    Topological Similarity Search in Large Combinatorial Fragment Spaces

    Louis Bellmann, Patrick Penner, and Matthias Rarey. Topological Similarity Search in Large Combinatorial Fragment Spaces . Journal of Chemical Information and Modeling, 61 0 (1): 0 238--251, 2021. ISSN 1549-9596. doi:10.1021/acs.jcim.0c00850

    work page doi:10.1021/acs.jcim.0c00850 2021

  2. [2]

    Richard Bickerton, Gaia V

    G. Richard Bickerton, Gaia V. Paolini, Jérémy Besnard, Sorel Muresan, and Andrew L. Hopkins. Quantifying the chemical beauty of drugs . Nature Chemistry, 4 0 (2): 0 90, 2012. ISSN 1755-4349. doi:10.1038/nchem.1243

    doi:10.1038/nchem.1243 2012

  3. [3]

    A Model to Search for Synthesizable Molecules

    John Bradshaw, Brooks Paige, Matt J Kusner, Marwin H S Segler, and José Miguel Hernández-Lobato. A Model to Search for Synthesizable Molecules . arXiv, 2019. doi:10.48550/arxiv.1906.05221

    arXiv doi:10.48550/arxiv.1906.05221 2019

  4. [4]

    Carhart, Dennis H

    Raymond E. Carhart, Dennis H. Smith, and R. Venkataraghavan. Atom pairs as molecular features in structure-activity studies: definition and applications. J. Chem. Inf. Comput. Sci., 25: 0 64--73, 1985. URL https://api.semanticscholar.org/CorpusID:37017771

    work page 1985

  5. [5]

    Shape-Aware Synthon Search (SASS) for virtual screening of synthon-based chemical spaces

    Chen Cheng and Paul Beroza. Shape-Aware Synthon Search (SASS) for virtual screening of synthon-based chemical spaces . ChemRxiv, 2023. doi:10.26434/chemrxiv-2023-68jzh

    work page doi:10.26434/chemrxiv-2023-68jzh 2023

  6. [6]

    Synflownet: Design of diverse and novel molecules with synthesis constraints, 2024

    Miruna Cretu, Charles Harris, Ilia Igashov, Arne Schneuing, Marwin Segler, Bruno Correia, Julien Roy, Emmanuel Bengio, and Pietro Liò. Synflownet: Design of diverse and novel molecules with synthesis constraints, 2024. URL https://arxiv.org/abs/2405.01155

    arXiv 2024

  7. [7]

    REAL database, 2024

    Enamine. REAL database, 2024. URL https://enamine.net/compound-collections/real-compounds/real-database

    work page 2024

  8. [8]

    PyTorch Lightning , March 2019

    William Falcon and The PyTorch Lightning team . PyTorch Lightning , March 2019. URL https://github.com/Lightning-AI/lightning

    2019

Show all 37 references
  1. [9]

    Haloperidol bound D2 dopamine receptor structure inspired the discovery of subtype selective ligands

    Luyu Fan, Liang Tan, Zhangcheng Chen, Jianzhong Qi, Fen Nie, Zhipu Luo, Jianjun Cheng, and Sheng Wang. Haloperidol bound D2 dopamine receptor structure inspired the discovery of subtype selective ligands . Nature Communications, 11 0 (1): 0 1074, 2020. doi:10.1038/s41467-020-14884-y

    2020 doi

  2. [10]

    LibINVENT: Reaction-based Generative Scaffold Decoration for in Silico Library Design

    Vendy Fialkov\'a, Jiaxi Zhao, Kostas Papadopoulos, Ola Engkvist, Esben Jannik Bjerrum, Thierry Kogej, and Atanas Patronov. LibINVENT: Reaction-based Generative Scaffold Decoration for in Silico Library Design . Journal of Chemical Information and Modeling, 2021. ISSN 1549-9596...

    2021 doi

  3. [11]

    Dropout as a bayesian approximation: Representing model uncertainty in deep learning

    Yarin Gal and Zoubin Ghahramani. Dropout as a bayesian approximation: Representing model uncertainty in deep learning. In Proceedings of the 33nd International Conference on Machine Learning, ICML , volume 48, pp.\ 1050--1059. JMLR.org, 2016. URL http://proceedings.mlr.press/v...

    2016

  4. [12]

    Wenhao Gao and Connor W. Coley. The Synthesizability of Molecules Proposed by Generative Models . Journal of Chemical Information and Modeling, 60 0 (12): 0 5714--5723, 04 2020. ISSN 1549-9596. doi:10.1021/acs.jcim.0c00174. URL https://doi.org/10.1021/acs.jcim.0c00174

    2020 doi

  5. [13]

    Generative Artificial Intelligence for Navigating Synthesizable Chemical Space

    Wenhao Gao, Shitong Luo, and Connor W Coley. Generative Artificial Intelligence for Navigating Synthesizable Chemical Space . arXiv, 2024

    2024

  6. [14]

    Accelerating high-throughput virtual screening through molecular pool-based active learning

    David E Graff, Eugene I Shakhnovich, and Connor W Coley. Accelerating high-throughput virtual screening through molecular pool-based active learning . Chem Sci., 2021. ISSN 2021;12(22):7866-7881. doi:10.1039/d0sc06805e

    2021 doi

  7. [15]

    Francesca Grisoni, Berend J. H. Huisman, Alexander L. Button, Michael Moret, Kenneth Atz, Daniel Merk, and Gisbert Schneider. Combining generative artificial intelligence and on-chip synthesis for de novo drug design . Science Advances, 7 0 (24): 0 eabg3338, 2021. ISSN 2375-25...

    2021 doi

  8. [16]

    Greenman, Yunsie Chung, Shih-Cheng Li, David E

    Esther Heid, Kevin P. Greenman, Yunsie Chung, Shih-Cheng Li, David E. Graff, Florence H. Vermeire, Haoyang Wu, William H. Green, and Charles J. McGill. Chemprop: A Machine Learning Package for Chemical Property Prediction . Journal of Chemical Information and Modeling, 64 0 (1...

    2024 doi

  9. [17]

    Patrick Walters

    Kathryn Klarich, Brian Goldman, Trevor Kramer, Patrick Riley, and W. Patrick Walters. Thompson Sampling An Efficient Method for Searching Ultralarge Synthesis on Demand Databases . Journal of Chemical Information and Modeling, 2024. ISSN 1549-9596. doi:10.1021/acs.jcim.3c01790

    2024 doi

  10. [18]

    Fast substructure search in combinatorial library spaces

    Thomas Liphardt and Thomas Sander. Fast substructure search in combinatorial library spaces. Journal of Chemical Information and Modeling, 63 0 (16): 0 5133--5141, 2023. doi:10.1021/acs.jcim.3c00290. URL https://doi.org/10.1021/acs.jcim.3c00290. PMID: 37221856

    2023 doi

  11. [19]

    Loeffler, Jiazhen He, Alessandro Tibo, Jon Paul Janet, Alexey Voronov, Lewis H

    Hannes H. Loeffler, Jiazhen He, Alessandro Tibo, Jon Paul Janet, Alexey Voronov, Lewis H. Mervin, and Ola Engkvist. Reinvent 4: Modern AI–driven generative molecule design . Journal of Cheminformatics, 16 0 (1): 0 20, 2024. ISSN 1758-2946. doi:10.1186/s13321-024-00812-5

    2024 doi

  12. [20]

    Coley, and Jianzhu Ma

    Shitong Luo, Wenhao Gao, Zuofan Wu, Jian Peng, Connor W. Coley, and Jianzhu Ma. Projecting molecules into synthesizable chemical spaces, 2024. URL https://arxiv.org/abs/2406.04628

    2024

  13. [21]

    Irwin, and Brian K

    Jiankun Lyu, John J. Irwin, and Brian K. Shoichet. Modeling the expansion of virtual screening libraries . Nature Chemical Biology, 19 0 (6): 0 712--718, 2023. ISSN 1552-4450. doi:10.1038/s41589-022-01234-w

    2023 doi

  14. [22]

    Mcule Database , accessed 18 Sep 2023

    Mcule. Mcule Database , accessed 18 Sep 2023. URL https://mcule.com/database/

    2023

  15. [23]

    David Mendez, Anna Gaulton, A. P. Bento, Jon Chambers, Marleen De Veij, Edwin Félix, María P. Magariños, Juan F. Mosquera, Prudence Mutowo, Michał Nowotka, Marta Gordillo-Marañón, Fiona Hunter, Leandro Junco, Grace Mugumbate, María Rodriguez-Lopez, Francis Atkinson, Nicolas Bo...

    2019 doi

  16. [24]

    OpenEye Scientific Software , OpenEye toolkits (2022.1.1)

    OpenEye . OpenEye Scientific Software , OpenEye toolkits (2022.1.1). http://www.eyesopen.com, 2022

    2022

  17. [25]

    Pytorch: An imperative style, high-performance deep learning library

    Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, et al. Pytorch: An imperative style, high-performance deep learning library. Advances in neural information processing systems, 32, 2019

    2019

  18. [26]

    Rajapakse, Philippe G

    Hemaka A. Rajapakse, Philippe G. Nantermet, Harold G. Selnick, Sanjeev Munshi, Georgia B. McGaughey, Stacey R. Lindsley, Mary Beth Young, Ming-Tain Lai, Amy S. Espeseth, Xiao-Ping Shi, Dennis Colussi, Beth Pietrak, Ming-Chih Crouthamel, Katherine Tugusheva, Qian Huang, Min Xu,...

    2006

  19. [27]

    Open-source cheminformatics

    RDKit . Open-source cheminformatics. (2023.09.6). https://www.rdkit.org, 2023

    2023

  20. [28]

    On failure modes in molecule generation and optimization

    Philipp Renz, Dries Van Rompaey, Jörg Kurt Wegner, Sepp Hochreiter, and Günter Klambauer. On failure modes in molecule generation and optimization . Drug Discovery Today: Technologies, 32: 0 55--63, 2019. ISSN 1740-6749. doi:10.1016/j.ddtec.2020.09.003

    2019 doi

  21. [29]

    Sadybekov, Anastasiia V

    Arman A. Sadybekov, Anastasiia V. Sadybekov, Yongfeng Liu, Christos Iliopoulos-Tsoutsouvas, Xi-Ping Huang, Julie Pickett, Blake Houser, Nilkanth Patel, Ngan K. Tran, Fei Tong, Nikolai Zvonok, Manish K. Jain, Olena Savych, Dmytro S. Radchenko, Spyros P. Nikas, Nicos A. Petasis,...

    2022

  22. [30]

    Maximum Common Substructure Searching in Combinatorial Make-on-Demand Compound Spaces

    Robert Schmidt, Raphael Klein, and Matthias Rarey. Maximum Common Substructure Searching in Combinatorial Make-on-Demand Compound Spaces . Journal of Chemical Information and Modeling, 2021. ISSN 1549-9596. doi:10.1021/acs.jcim.1c00640

    2021 doi

  23. [31]

    Automating drug discovery

    Gisbert Schneider. Automating drug discovery . Nature Reviews Drug Discovery, 17 0 (2): 0 97, 02 2017. ISSN 1474-1784. doi:10.1038/nrd.2017.232. URL http://www.nature.com/articles/nrd.2017.232

    2017 doi

  24. [32]

    Bronstein, and Renana Gershoni-Poranne

    Tomer Weiss, Eduardo Mayo Yanes, Sabyasachi Chakraborty, Luca Cosmo, Alex M. Bronstein, and Renana Gershoni-Poranne. Guided diffusion for inverse molecular design . Nature Computational Science, 3 0 (10): 0 873--882, 2023. doi:10.1038/s43588-023-00532-0

    2023 doi

  25. [33]

    Wood, Svitlana Korolchuk, Natalie J

    Daniel J. Wood, Svitlana Korolchuk, Natalie J. Tatum, Lan-Zhen Wang, Jane A. Endicott, Martin E.M. Noble, and Mathew P. Martin. Differences in the Conformational Energy Landscape of CDK1 and CDK2 Suggest a Mechanism for Achieving Selective CDK Inhibition . Cell Chemical Biolog...

    2019 doi

  26. [34]

    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 gl...

  27. [35]

    @esa (Ref

    \@ifxundefined[1] #1\@undefined \@firstoftwo \@secondoftwo \@ifnum[1] #1 \@firstoftwo \@secondoftwo \@ifx[1] #1 \@firstoftwo \@secondoftwo [2] @ #1 \@temptokena #2 #1 @ \@temptokena \@ifclassloaded agu2001 natbib The agu2001 class already includes natbib coding, so you should ...

  28. [36]

    \@lbibitem[] @bibitem@first@sw\@secondoftwo \@lbibitem[#1]#2 \@extra@b@citeb \@ifundefined br@#2\@extra@b@citeb \@namedef br@#2 \@nameuse br@#2\@extra@b@citeb \@ifundefined b@#2\@extra@b@citeb @num @parse #2 @tmp #1 NAT@b@open@#2 NAT@b@shut@#2 \@ifnum @merge>\@ne @bibitem@firs...

  29. [37]

    @open @close @open @close and [1] URL: #1 \@ifundefined chapter * \@mkboth \@ifxundefined @sectionbib * \@mkboth * \@mkboth\@gobbletwo \@ifclassloaded amsart * \@ifclassloaded amsbook * \@ifxundefined @heading @heading NAT@ctr thebibliography [1] @ \@biblabel @NAT@ctr \@bibset...

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.