Pith. sign in

REVIEW 5 major objections 5 minor 14 references

Active Learning-Guided Seq2Seq Variational Autoencoder for Multi-target Inhibitor Generation

T0 review · 5 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 A two-level active-learning loop around a Seq2Seq variational autoencoder generated 960 structurally diverse molecules with predicted simultaneous affinity to the main proteases of three coronaviruses, where a 214-compound reference set…

desk verdict The workflow is a credible proof-of-concept, but the headline 960-vs-1 comparison is a sample-size artifact, and the missing no-AL baseline means the paper does not actually show that active learning helps. read the letter →

arxiv 2506.15309 v1 pith:JRQHYKLP submitted 2025-06-18 cs.LG cs.AIq-bio.BM

Júlia Vilalta-Mor , Alexis Molina , Laura Ortega Varga , Isaac Filella-Merce , Victor Guallar This is my paper · ORCID
classification cs.LGcs.AIq-bio.BM
keywords activelearningSeq2Seqvariationalautoencodermulti-targetdrugdesignpan-inhibitorscoronavirusmainproteasemoleculardockingSMILESgeneration
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

This paper is trying to establish that a generative model can be steered toward multi-target inhibitors, a problem normally dominated by sparse rewards and conflicting objectives, by interleaving two kinds of active learning. The first loop keeps only molecules that pass drug-likeness, synthetic accessibility, and diversity filters; the second refines the generator on molecules that dock well against all three coronavirus main proteases at once, with the docking threshold tightened on a fixed schedule. In the proof of concept, the workflow produced 960 generated molecules that pass the -8 kcal/mol global and per-target Glide docking thresholds, compared with one molecule in the 214-compound fixed specific set. The paper also argues that where the SMARTS-based medicinal chemistry filters sit matters: deferring them until after generation rather than baking them into the training loop roughly triples the count at the most stringent thresholds. The authors state that experimental validation is the necessary next step before any of these molecules can be called a drug candidate.

What carries the argument

The machinery is the two-level active learning loop wrapped around a Seq2Seq variational autoencoder. The VAE embeds SMILES strings into a 128-dimensional Gaussian latent space and decodes fresh molecules from samples of $\mathcal{N}(0,I)$; because the encoder and decoder are LSTMs, the latent space is continuous and densely populated, so nearby latent points near valid molecules generate plausible SMILES. The Chemical AL cycle filters generated molecules by substructure motifs, QED $\ge 0.8$, SA $\le 3$, and a Tanimoto similarity cutoff, then fine-tunes the VAE from the general weights on the accumulated survivors. The Affinity AL cycle docks the survivors with Glide SP to the three Mpro structures and keeps only molecules whose mean docking score and per-target scores beat thresholds that decay by $\delta=0.1$ kcal/mol per cycle while at least $N_{\min}=50$ molecules survive, stopping after $p=3$ cycles without decay. This moving threshold is what converts sparse rewards into a smooth selection pressure: each retraining biases the latent distribution toward the current affinity bar, so tightening the bar shifts the whole distribution of generated molecules toward lower docking scores.

What would settle it

Synthesize a sample of, say, 50 of the 960 candidates and 50 molecules that narrowly missed the -8 kcal/mol threshold, and measure their IC50 values against recombinant Mpro from all three viruses; the central claim predicts a much higher hit rate in the passing group. If the two groups inhibit equally often, the enrichment is an artifact of the docking filter rather than real multi-target affinity.

Watch

Extended reading notes

Core claim

On its own terms, the paper's central claim is that a two-level active learning scheme turns sparse-reward, multi-objective drug design into a tractable generation problem. A Seq2Seq variational autoencoder is pretrained on a general SMILES corpus, fine-tuned on a fixed set of 214 known protease inhibitors, and then repeatedly fine-tuned from the general weights on an accumulating set of generated molecules: first those that survive chemical filters (QED, SA, Tanimoto, SMARTS), then those that pass both a global mean docking threshold and individual thresholds against the Mpro of SARS-CoV-2, SARS-CoV, and MERS-CoV. The thresholds decay by 0.1 kcal/mol per Affinity AL cycle as long as at least 50 molecules survive, with a patience of three cycles. Across 15 cycles in the ablated configuration, this produced 650 post-filter molecules at -8 kcal/mol global and individual thresholds, and 310 in the regular configuration, for 960 candidates total versus one fixed-set molecule; at the more stringent -9/-8 thresholds, the ablated, post-hoc-filtered pipeline yields three times as many molecules as the regular one (21 versus 7). The 960 candidates spread across many scaffold clusters, and the paper reports that 959 of the 960 were absent from the Enamine REAL DB, ZINC22, and WuXi libraries.

Load-bearing premise

The load-bearing premise is that the computer-simulated binding-affinity scores used to filter and count molecules are a valid proxy for real inhibition of the three viral proteases, even though each score comes from a single fixed protein structure and has not been tested experimentally.

Editorial extensions

If this is right

  • Multi-target generation can be driven by docking-score thresholds alone, without per-target reward shaping or an explicit multi-objective objective function.
  • Filter placement is a quantitative design choice: moving SMARTS-based filters out of the training loop and applying them once at the end roughly triples the number of post-filter candidates at the most stringent thresholds.
  • The generator maintains high uniqueness (98.5%) and novelty (99.5%) across cycles while still shifting toward lower docking scores, so exploration and affinity optimization do not have to trade off under this scheme.
  • Almost none of the 960 candidates appear in ultra-large, combinatorially constructed libraries, which suggests the generated chemical space is complementary to what those libraries cover.
  • In target contexts where SMARTS-unfavourable motifs are the ones that drive affinity, the regular configuration may be preferable to avoid propagating such motifs; the paper says this trade-off is context-dependent.

Reading between the lines

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

  • Because both the training filter and the final count use the same three receptor structures, the enrichment could partly reflect overfitting to those structures; re-docking the 960 candidates against the full set of 195 non-redundant Mpro conformations would test whether the claimed pan-inhibitor signal is structure-specific.
  • The same two-level schedule should transfer to other target triples with a conserved binding site, and the most informative control would be to run both configurations on a less homologous target set, where the ablated configuration's advantage may shrink or reverse.
  • A direct experimental readout, for example comparing enzymatic inhibition of candidates that just pass versus just miss the -8 kcal/mol threshold, would separate genuine chemical enrichment from systematic docking-score bias, since active learning amplifies whatever signal the filter uses.
  • The method's stated flexibility implies a cheap extension that the authors do not run: replace the three viral proteases with several conformations of one target and the same threshold decay should steer toward conformation-general binders.
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

5 major / 5 minor

Summary. The paper proposes a two-level active learning (AL) workflow around a Seq2Seq VAE for generating SMILES strings, targeting the main proteases of SARS-CoV-2, SARS-CoV, and MERS-CoV. A 'Chemical AL' cycle filters generated molecules by physicochemical property and SMARTS-based criteria, and an 'Affinity AL' cycle filters by Glide docking scores against three crystal structures with progressively decaying thresholds. Two configurations are compared: a regular one with SMARTS filters inside the Chemical AL loop, and an ablated one that defers SMARTS filtering until after generation. The authors report that the ablated configuration produces more molecules passing the docking thresholds, and identify 960 candidate pan-inhibitors passing -8/-8 kcal/mol Glide thresholds, compared to only 1 in the fixed 214-molecule training set. The paper includes diversity analyses (validity, uniqueness, novelty, scaffold clustering, UMAP) and a search of the candidates against ultra-large chemical libraries.

Significance. If the central claim is substantiated, the two-level AL workflow is a useful contribution to multi-target generative design: it is well-motivated, combines a generative model with a realistic docking oracle, and includes interesting ablations, diversity metrics, and computational timing. The paper also makes a concrete, checkable prediction (960 docking-passing compounds) that could be followed up in vitro. However, the reported evidence does not currently support the headline enrichment claim: the comparison is based on absolute counts with unstated denominators, there is no control generation without AL, and the evaluation metric is the same function used as the selection oracle. These issues are central rather than cosmetic.

major comments (5)
  1. [Section 3.2, Table 1, Figure 2B] The claimed enrichment, 'only one molecule in the fixed specific set' versus 650 or 960 generated molecules, compares absolute counts across pools of very different and unreported sizes. The total number of generated molecules and the number of molecules that were actually docked in each configuration are not reported in the main text or in Table 1; Figure 5 gives per-cycle counts and percentages but no cumulative denominator. Without a normalized hit rate, the 650-vs-1 comparison is not evidence of enrichment even under the authors' own docking-score assumption. Reports of the total unique valid molecules and the cumulative docking-tested pool per configuration are needed before any enrichment claim can be evaluated.
  2. [Section 3.1, Section 2.1] The paper does not include a no-AL control. Both the regular and ablated configurations run the two-level AL loops; neither is compared to a VAE fine-tuned on the fixed 214-molecule specific set without any Chemical or Affinity AL cycles. Consequently, the core claim that AL is what drives the reported docking-hit counts is unsupported: the observed counts could in principle be obtained from the fine-tuned VAE alone, with the AL loops contributing little. This is a load-bearing missing baseline for a paper titled and framed around active learning.
  3. [Sections 2.4 and 3.2] The success metric is the same Glide SP docking function used as the AL selection oracle. Molecules are admitted into the fine-tuning set in Section 2.4 by satisfying the Glide thresholds, and the candidate count in Section 3.2 is defined by applying the same thresholds. The reported counts therefore measure how well the pipeline optimized a particular scoring function, not independent evidence of affinity. The Discussion's acknowledgment that 'further in silico and experimental validation will be essential' does not address the circularity of the headline number. Evaluation with a different scoring method (e.g., Glide XP or MM-GBSA) or against a held-out receptor structure would make the claim informative.
  4. [Section 3.2, Section 4, Table 1] There is a numerical inconsistency in the headline count. Section 3.2 states 'Among all generated candidates (969)', while the Discussion states 'selection of 960 pan-inhibitor candidates' and '959 out of 960 were absent'. It is not explained how these relate to the regular (310) and ablated-after-filtering (650) counts in Table 1, or whether the two configuration outputs were merged with duplicate removal. Since 960/969 is the central quantitative result, this inconsistency must be resolved.
  5. [Section 3.1 and Table 1] The choice of the ablated configuration for the rest of the analysis is made post hoc because it produces more candidates, but the paper reports only one run per configuration. The claimed 3-fold increase at the -9/-8 kcal/mol thresholds (21 versus 7 in Table 1) and the attendant conclusion that deferred SMARTS filtering is 'more effective' are therefore without uncertainty estimates or statistical support. At minimum, results from repeated runs with different random seeds, or a clear statement that the comparison is descriptive only, are required.
minor comments (5)
  1. [Table 1] The table caption reads 'Percentage identity derived from the structure-based sequence alignment just on the catalytic site...' but the table contains docking-threshold counts. The caption appears to be a copy-paste error from Table 3 and should be corrected.
  2. [Appendix G] There is a typographical error in the performance section: one entry refers to a 'GPU H1000', which should presumably be 'H100'.
  3. [Section 2.1] The sentence 'After completing then Chemical AL cycles' should read 'After completing the n Chemical AL cycles'.
  4. [Section 3.2] The sentence 'we consider as a candidate pan-inhibitor any generated molecule with an individual docking score threshold of -8 kcal/mol for each target' is ambiguous; the table and the text together suggest both a global and an individual threshold of -8 kcal/mol, but the wording here only mentions the individual threshold.
  5. [Figure 2B] The zoom-in panel is described qualitatively ('marked enrichment'); the text would benefit from stating how the histogram counts were normalized (e.g., counts per 0.1 kcal/mol bin or raw counts) and whether the generated pool size differs between cycles.

Circularity Check

1 steps flagged · score 6.0 of 10

The 960-candidate headline is scored with the same Glide docking oracle used to select the Affinity AL training set, so the enrichment is partially by construction; no independent baseline or orthogonal evaluation is provided.

  1. fitted input called prediction [Section 2.4 (Affinity AL cycle, Eqs. 4-5) and Section 3.2 (Candidate pan-inhibitors, Table 1)]
    "Affinity AL cycles are applied to guide molecular generation toward molecules with multi-target affinity. Ligand–protein affinities are predicted by docking molecules from the accumulated specific dataset to the multiple protein targets ... Molecules that satisfy all affinity thresholds are transferred to the updated specific dataset. ... we consider as a candidate pan-inhibitor any generated molecule with an individual docking score threshold of -8 kcal/mol for each target."

    The molecules that pass the Glide docking-score thresholds in Section 2.4 are transferred into the updated specific dataset used to fine-tune the VAE; the same Glide SP docking-score thresholds (e.g., -8 kcal/mol global and individual) are then used in Section 3.2 to define and count candidate pan-inhibitors. The reported 'enrichment' (310 regular and 650 ablated at -8/-8 vs 1 in the fixed set) is therefore the expected consequence of training the generator on molecules selected by the same scoring function it is later evaluated against, rather than an independent prediction of multi-target activity.

full rationale

The load-bearing quantitative claim is the count of generated molecules passing -8 kcal/mol global and individual Glide thresholds versus one molecule in the fixed set. That metric is the same Glide SP docking score used as the Affinity AL selection oracle in Section 2.4: molecules satisfying the thresholds are fed back into the VAE fine-tuning, so later 'candidates' are molecules optimized against that exact function. This is a partial circularity: the model is being evaluated on the score it was trained to improve. The comparison with the fixed set is not controlled, since the fixed set was selected at -5.9 kcal/mol and contains only 214 compounds, while the generated pool is much larger and was iteratively filtered at more stringent thresholds. No no-AL baseline (VAE fine-tuned on the fixed set without AL cycles) is reported, so the contribution of the AL loop itself is unestablished. The paper is transparent that these are computational candidates and states that further in silico and experimental validation is needed; the architecture citations (e.g., Gómez-Bombarelli, Gupta) are standard and not load-bearing self-citations. Because the central claim reduces in part to optimizing and then counting by the same oracle, the circularity score is 6 rather than 0-2.

Assumptions & free parameters 9 free parameters · 5 assumptions · 0 invented entities

No new physical entities are introduced. The workflow depends on many hand-set thresholds that jointly define both the training signal and the success metric; sensitivity to these thresholds is not analyzed. The main circularity burden is that Glide docking is both the selection oracle in Affinity AL and the evaluation metric for pan-inhibitor candidates, although the fixed-set and library comparisons provide partial external anchoring.

free parameters (9)
  • Initial global and individual docking score thresholds = -7.5 and -7.0 kcal/mol
    Starting thresholds for Affinity AL; chosen by hand. They determine which molecules enter the fine-tuning set in the first Affinity cycles.
  • Docking threshold decay rate δ = 0.1 kcal/mol
    Linear decay applied each cycle if at least Nmin molecules pass; hand-chosen and never varied.
  • Minimum molecule count Nmin = 50
    Condition for continuing threshold decay; controls how aggressively thresholds tighten.
  • Stopping patience p = 3
    Number of consecutive cycles without threshold decay that stops the workflow; hand-chosen.
  • Tanimoto similarity threshold = 0.4 in first cycle, 0.6 later
    Diversity filter in Chemical AL; the first long cycle uses stricter 0.4 to maximize exploration.
  • QED threshold = 0.8
    Drug-likeness filter in Chemical AL; hand-chosen.
  • SA threshold = 3
    Synthetic accessibility filter in Chemical AL; hand-chosen.
  • Specific-set construction docking threshold = -5.9 kcal/mol
    Threshold used to retain known inhibitors for the fixed specific set; set as the average docking score of all molecules across all targets.
  • Candidate pan-inhibitor threshold = -8.0 kcal/mol global and individual
    Definition of candidate used for the headline 960 count; chosen by the authors without a stated selection rule.
assumptions (5)
  • domain assumption Glide SP docking score is a valid proxy for binding affinity to the three Mpro targets
    Invoked in Section 2.4 (Affinity AL filtering) and Section C (Docking Protocol); the entire evaluation and selection rests on this.
  • domain assumption The selected crystal structures 7RNW, 2GX4, and 7ENE are representative of their respective viral Mpro targets
    Section B describes the structural clustering and receptor selection; the docking scores depend on these structures.
  • domain assumption The 214 molecules in the fixed specific set are genuine multi-target-relevant inhibitors
    Section A; these come from literature IC50 data but are then filtered by docking, and their activity labels are taken on trust.
  • domain assumption The Seq2Seq VAE trained on the ChEMBL-derived set generates a chemically valid distribution of SMILES
    Section 2.2; the workflow relies on the VAE for molecular grammar, and validity averages only 59.62%.
  • domain assumption SMARTS-based catalogs (PAINS, Brenk, NIH, CHEMBL) identify undesirable chemical motifs
    Section 2.3; these filters define chemically relevant in both configurations and the post-generation filtering.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Active Learning-Guided Seq2Seq Variational Autoencoder for Multi-target Inhibitor Generation." pith.science (2026). https://pith.science/paper/JRQHYKLP

@misc{pith2026250615309,
  author       = {Pith},
  title        = {Pith review of: Active Learning-Guided Seq2Seq Variational Autoencoder for Multi-target Inhibitor Generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JRQHYKLP}},
  note         = {Machine review of arXiv:2506.15309}
}
read the original abstract

Simultaneously optimizing molecules against multiple therapeutic targets remains a profound challenge in drug discovery, particularly due to sparse rewards and conflicting design constraints. We propose a structured active learning (AL) paradigm integrating a sequence-to-sequence (Seq2Seq) variational autoencoder (VAE) into iterative loops designed to balance chemical diversity, molecular quality, and multi-target affinity. Our method alternates between expanding chemically feasible regions of latent space and progressively constraining molecules based on increasingly stringent multi-target docking thresholds. In a proof-of-concept study targeting three related coronavirus main proteases (SARS-CoV-2, SARS-CoV, MERS-CoV), our approach efficiently generated a structurally diverse set of pan-inhibitor candidates. We demonstrate that careful timing and strategic placement of chemical filters within this active learning pipeline markedly enhance exploration of beneficial chemical space, transforming the sparse-reward, multi-objective drug design problem into an accessible computational task. Our framework thus provides a generalizable roadmap for efficiently navigating complex polypharmacological landscapes.

Figures

Figures reproduced from arXiv: 2506.15309 by the authors.

Figure 1
Figure 1. The multi-target generative workflow. The two-level AL workflow is illustrated with arrows of different colours: the first-level cycle, Chemical AL, is shown in orange, while the second-level cycle, Affinity AL, is shown in green. The specific training set stands for F: fixed, U: updated, and A: accumulated. µ = Wµh ′ + bµ, logσ2 = Wσh ′ + bσ (2) where Wµ, Wσ ∈ R 128×256 and bµ, bσ ∈ R 128. The latent vector z is sa… view at source ↗
Figure 2
Figure 2. A) Evolution of specific set size (logarithmic scale) after each Chemical AL cycle for both regular and ablated configurations. Vertical dotted lines mark the end of each Affinity AL cycle. The secondary vertical axis represents the global docking score threshold applied to each Affinity AL cycle. The initial point represents the fixed specific set size. Points between two dotted vertical lines represent the evoluti… view at source ↗
Figure 3
Figure 3. Global docking score cumulative histograms of generated molecules across all Affinity AL cycles for the regular and the ablated configurations, and also the fixed specific set. The zoomed-in section represents the molecules under the final defined threshold of -8 kcal/mol. E. Generation Statistics The validity, uniqueness, and novelty of the generated molecules at each Chemical AL cycle were calculated as follows: V… view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Validity, uniqueness, and novelty of the molecules generated at each Chemical AL cycle. The vertical dotted lines represent the division between Affinity AL cycles. We tracked the number of molecules that fulfilled the criteria of the Chemical and Affinity AL cycles at…
Figure 5
Figure 5. Figure 5: Number and percentage of generated molecules at each Affinity AL cycle fulfilling the Chemical AL filters and the Affinity AL filters. The docking score thresholds (global and individual) that were applied at each Affinity AL cycle are represented on the right-y axis. …
Figure 6
Figure 6. Figure 6: Evolution of scaffold clusters across Affinity AL cycles. Each connected line represents a DBSCAN clustering performed with a different epsilon parameter, as indicated in the legend. A) Evolution of scaffold clusters among generated molecules below the thresholds appli…
Figure 7
Figure 7. Figure 7: UMAP illustrating the generated molecules over the Affinity AL cycles, with marker styles and colours indicating different cycles as shown in the legend. Fucsin crosses represent the candidate pan-inhibitors, thus generated molecules that passed the docking score thres…

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

14 extracted references · 9 canonical work pages

  1. [11]

    scikit learn

    URL https://www.schrodinger.com/ platform/products/ligprep/. scikit learn. DBSCAN, 2025. URL https: //scikit-learn/stable/modules/ generated/sklearn.cluster.DBSCAN.html. Shahhamzehei, N., Abdelfatah, S., and Efferth, T. In Sil- ico and In Vitro Identification of Pan-Coronaviral Main Protease Inhibitors from a Large Natural Product Li- brary.Pharmaceutical...

    work page doi:10.3390/ph15030308 2025

  2. [13]

    Target structures were preprocessed by removing water molecules, ligands, and ions

    were collected and classified according to their viral origin: 479 from SARS-CoV-2, 32 from SARS-CoV , and 32 from MERS-CoV . Target structures were preprocessed by removing water molecules, ligands, and ions. In cases where structures contained multiple chains, reflecting both Mpro dimeric and monomeric states, individual chains were separated, and only ...

    work page 2025

  3. [14]

    Then we hierarchically clustered this matrix using Seaborn (Waskom et al., 2017)

    and constructed a pairwise catalytic site volume overlapping matrix. Then we hierarchically clustered this matrix using Seaborn (Waskom et al., 2017). This allowed us to identify clusters of very similar 3D catalytic site volumes from which we extract representatives, reducing the total number of Mpro structures from 543 to 195. Cross-docking of the fixed...

    work page 2017

  4. [2004]

    doi: 10.1021/jm030644s

    ISSN 0022-2623. doi: 10.1021/jm030644s. URL https://doi.org/10.1021/jm030644s. Pub- lisher: American Chemical Society. Huang, C., Shuai, H., Qiao, J., Hou, Y ., Zeng, R., Xia, A., Xie, L., Fang, Z., Li, Y ., Yoon, C., Huang, Q., Hu, B., You, J., Quan, B., Zhao, X., Guo, N., Zhang, S., Ma, R., Zhang, J., Wang, Y ., Yang, R., Zhang, S., Nan, J., Xu, H., Wan...

    doi:10.1021/jm030644s

  5. [2008]

    doi: 10.1002/cmdc.200700139

    ISSN 1860-7187. doi: 10.1002/cmdc.200700139. URL https://onlinelibrary.wiley.com/ doi/abs/10.1002/cmdc.200700139. Cicho´nska, A., Ravikumar, B., and Rahman, R. AI for targeted polypharmacology: The next frontier in drug discovery.Current Opinion in Struc- tural Biology, 84:102771, February 2024. ISSN 0959-440X. doi: 10.1016/j.sbi.2023.102771. URL https://...

    arXiv

  6. [2009]

    doi: 10.1021/ci800324m

    ISSN 1549-9596. doi: 10.1021/ci800324m. URL https://doi.org/10.1021/ci800324m. Pub- lisher: American Chemical Society. 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 Scor- ing. 2. Enrichment Factors in Database Screening.Jour- nal of Medicina...

    work page doi:10.1021/ci800324m

  7. [2010]

    doi: 10.1021/jm901070c

    ISSN 0022-2623. doi: 10.1021/jm901070c. URL https://doi.org/10.1021/jm901070c. Pub- lisher: American Chemical Society. Katoh, K. and Standley, D. M. MAFFT Multiple Sequence Alignment Software Version 7: Improvements in Per- formance and Usability.Molecular Biology and Evolu- tion, 30(4):772–780, April 2013. ISSN 0737-4038. doi: 10.1093/molbev/mst010. URL ...

    work page doi:10.1021/jm901070c

  8. [2014]

    doi: 10.1039/C4OB02287D

    ISSN 1477-0539. doi: 10.1039/C4OB02287D. URL https://pubs.rsc.org/en/content/ articlelanding/2015/ob/c4ob02287d. Publisher: The Royal Society of Chemistry. Edgar, R. C. MUSCLE: multiple sequence alignment with high accuracy and high throughput.Nucleic Acids Research, 32(5):1792–1797, 2004. ISSN 0305-1048. doi: 10.1093/nar/gkh340. URL https://www.ncbi. nlm...

    work page doi:10.1039/c4ob02287d 2015

Show all 14 references
  1. [2018]

    doi: 10.1002/pro.3290

    ISSN 1469-896X. doi: 10.1002/pro.3290. URL https://onlinelibrary.wiley. 10 AL-Guided Seq2Seq V AE for Multi-target Generation com/doi/abs/10.1002/pro.3290. eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1002/pro.3290. Swanson, K., Liu, G., Catacutan, D. B., Arnold, A., Zou...

    arXiv

  2. [2019]

    URL http://arxiv.org/abs/1907. 08956. arXiv:1907.08956 [cs]. Olivecrona, M., Blaschke, T., Engkvist, O., and Chen, H. Molecular de-novo design through deep reinforce- ment learning.Journal of Cheminformatics, 9(1):48, September 2017. ISSN 1758-2946. doi: 10.1186/ s13321-017-02...

    1907 arXiv

  3. [2024]

    URL https://www.medchemexpress. com/. Munson, B. P., Chen, M., Bogosian, A., Kreisberg, J. F., Licon, K., Abagyan, R., Kuenzi, B. M., and Ideker, T. De novo generation of multi-target compounds using deep generative chemistry.Nature Communications, 15 (1):3636, May 2024. ISSN ...

    2024

  4. [2025]

    doi: 10.1093/nar/gkae1059

    ISSN 1362-4962. doi: 10.1093/nar/gkae1059. URL https://doi.org/10.1093/nar/gkae1059. Knox, C., Wilson, M., Klinger, C., Franklin, M., Oler, E., Wilson, A., Pon, A., Cox, J., Chin, N. E., Straw- bridge, S., Garcia-Patino, M., Kruger, R., Sivakumaran, A., Sanford, S., Doshi, R.,...

  5. [2946]

    URL https: //doi.org/10.1186/1758-2946-1-8

    doi: 10.1186/1758-2946-1-8. URL https: //doi.org/10.1186/1758-2946-1-8. Ester, M., Kriegel, H.-P., and Xu, X. A Density-Based Algo- rithm for Discovering Clusters in Large Spatial Databases with Noise.KDD-96 Proceedings, 1996. Filella-Merce, I., Molina, A., Orzechowski, M., D ...

    arXiv

  6. [4951]

    URL https: //doi.org/10.1007/s10822-013-9644-8

    doi: 10.1007/s10822-013-9644-8. URL https: //doi.org/10.1007/s10822-013-9644-8. McInnes, L., Healy, J., and Melville, J. UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction, September 2020. URL http://arxiv. org/abs/1802.03426. arXiv:1802.03426 [stat]. ...

    2020 arXiv

Pith tools

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