Pith. sign in

REVIEW 4 major objections 6 minor 55 references

Generative Model for Synthesizing Ionizable Lipids: A Monte Carlo Tree Search Approach

T0 review · 4 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read The paper claims that a policy network trained on its own search data lets Monte Carlo tree search generate predicted-ionizable two-tail lipids at 74% of unique test products, versus 27% for the unguided baseline.

desk verdict Clean MCTS-for-lipids application with a useful dataset, but the synthesis-path claim is contradicted by the paper's own Table 1 and the evaluation is partly circular. read the letter →

arxiv 2412.00807 v1 pith:WYB5KQYU submitted 2024-12-01 cs.LG cs.AIq-bio.BMq-bio.QM

Jingyi Zhao , Yuxuan Ou , Austin Tripp , Morteza Rasoulianboroujeni , José Miguel Hernández-Lobato This is my paper · ORCID
classification cs.LGcs.AIq-bio.BMq-bio.QM
keywords ionizablelipidslipidnanoparticlesmRNAdeliveryMonteCarlotreesearchpolicynetworkgenerativechemistrysynthesizabilityretrosynthesis
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

Ionizable lipids are the delivery molecules that let mRNA therapies reach cells, but designing new ones by hand is slow and generative models often propose structures that cannot be made. The paper aims to show that a Monte Carlo tree search, nudged by a learned policy network, can generate two-tail lipid candidates that are predicted to be ionizable and come with concrete synthesis pathways. It builds a dataset of synthetically accessible lipid-head and lipid-tail building blocks, trains a lipid-likeness classifier and an ionizability predictor to score candidates, and runs an iterative loop in which the tree search's choices train the policy that guides the next search. In test runs the guided search predicts 74% of unique generated products to be ionizable lipids, against 27% for the unguided baseline and 15% for random combination. If the in-silico predictors hold up, this would make the search for new mRNA-delivery lipids faster and more targeted.

What carries the argument

The load-bearing mechanism is the guided MCTS loop. The search builds a tree whose nodes are molecules: the root is expanded with lipid-head building blocks, each head is expanded with compatible tails, and chemical reactions combine them into intermediates until a two-tail lipid is formed. Every edge stores a visit count $N(s,a)$, a total value $W(s,a)$, and a prior probability $P(s,a)=f_\theta(s)$ supplied by the policy network, and selection follows the UCB score $Q(s,a)+U(s,a)$ with $Q=W/N$ and $U=c\,P(s,a)\sqrt{\sum_b N(s,b)}/(1+N(s,a))$. When a terminal two-tail lipid is reached, it is scored by the lipid classifier and the ionizability predictor, and that value is backpropagated. The visit counts of all state-action pairs are then converted into training targets, and the updated policy guides the next iteration.

What would settle it

Take the 545 unique test-phase products that the predictors call ionizable lipids, select a random sample of 50, and attempt to synthesize them using the building-block reactions the generative model proposes; then measure protonation between pH 7.4 and pH 5, the paper's own definition of ionizability. If the measured fraction of genuinely ionizable, synthesizable products lands near the 15% random-combination rate rather than near the predicted 74%, the central outperformance claim is falsified. A cheaper in-silico version of the same test is to re-score the generated set with an independent pKa calculation; if the high predicted rate disappears, the reported rates are artifacts of the specific predictors.

Watch

Extended reading notes

Core claim

The paper's central claim is that a policy-network-guided MCTS generative model outperforms the SyntheMol baseline at producing ionizable-lipid candidates that come with available synthesis paths. Concretely, the guided model generates 5,058 unique training-phase products with an ionizable-lipid rate of 0.3319 and 545 unique test-phase products with a rate of 0.7372, compared with 0.2739 for SyntheMol and 0.1547 for random combination. The same products receive average SA scores of 4.24 to 4.62, close to the 4.12 average of experimentally published ionizable lipids, and Syntheseus retro-valid rates of 0.4881 on training products and 0.2679 on test products. The authors present the approach not as a finished wet-lab validation but as a generative pipeline whose output is explicitly designed to be synthesizable from the curated building-block set.

Load-bearing premise

The paper's reported ionizable-lipid rates assume that the lipid classifier and the ionizability predictor, which were trained and validated mostly on known lipids, give correct answers for newly generated structures outside that distribution.

Editorial extensions

If this is right

  • Training the policy on the tree search's own visit data raises the fraction of unique generated products predicted to be ionizable lipids from 0.27 for unguided MCTS to 0.74 for guided MCTS on the testing head set.
  • The average SA scores of the generated products (4.24 to 4.62) are close to the 4.12 average of published ionizable lipids, indicating that the method does not sacrifice synthetic accessibility as measured by that score.
  • The curated building-block datasets, with over 2.7 million head candidates and 5,310 tail candidates, are a reusable resource for lipid generation tasks beyond this paper.
  • Syntheseus retro-valid rates of 0.49 on training products and 0.27 on test products show that at least some generated candidates can be decomposed back to molecules in the building-block dataset, giving concrete synthesis paths rather than abstract structures.
  • Because the largest quality gain appears after the first policy-training iteration, the guided method delivers its main benefit with relatively little additional computation.

Reading between the lines

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

  • The test-phase rate of 0.74 comes from a single fixed set of 200 head building blocks, so the headline number should be read as conditional on that set; repeated sampling of different head sets would give a more reliable performance estimate.
  • The retro-valid rates (0.27 to 0.49) are lower than the ionizable-lipid rates, so the synthesis-path part of the claim is the weaker link; it would be settled by attempting the proposed reactions in the laboratory.
  • An experimental synthesis of a small random sample of the top-scoring test products would directly test whether the predictor reward is selecting genuine ionizable lipids or exploiting blind spots in the in-silico scorers.
  • The guided-MCTS loop is a general recipe for any building-block-and-reaction molecular family whose bottleneck is synthesizability, provided a reliable property predictor for the target class exists.
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 / 6 minor

Summary. The paper proposes a policy-network-guided Monte Carlo Tree Search (MCTS) generative model for designing ionizable lipids by assembling lipid head and tail building blocks, using a Chemprop-based lipid classifier and a MolGpKa-based ionizability predictor as the reward. It describes the construction of a lipid building-block dataset from ZINC20, adapts the SyntheMol MCTS approach to lipid generation, and adds an AlphaZero-style policy network that is trained on visit counts from prior searches. The experimental section compares the guided MCTS against SyntheMol and random combination baselines, reporting ionizable lipid rates, SA scores, and Syntheseus retro-valid rates. The central claim is that this method produces high-quality ionizable lipids with available synthesis pathways and outperforms the SyntheMol baseline.

Significance. If the claimed performance were fully supported, the paper would be a useful demonstration of applying policy-guided MCTS to a drug-delivery-relevant chemical space, and the compiled lipid building-block datasets could facilitate future generative lipid design. The paper is transparent about the algorithmic workflow, includes pseudocode, and reports comparisons against meaningful baselines on the same objective. However, the significance is substantially weakened by three issues: the evaluation metric is the same objective being optimized; the compute budget is not matched across methods; and the synthesis-path claim is contradicted by the authors' own quantitative retrosynthesis results. These issues need to be addressed before the central claims can be accepted.

major comments (4)
  1. [§5.2.3, Table 1] The central claim in the abstract and Section 6 that the model generates ionizable lipids 'with available synthesis paths' is not supported by the paper's own quantitative evaluation. For the guided MCTS test products, the Syntheseus retro-valid rate is 0.2679, which is below the random-combination rate of 0.3103 and essentially tied with the SyntheMol baseline of 0.2719; the paper also notes that no case exceeds 50%. Moreover, the validation protocol only checks whether some retrosynthetically proposed reactants appear in the building-block dataset rather than validating the forward route actually recorded by the generator. The 'available synthesis paths' claim therefore rests almost entirely on two hand-picked examples in Appendix E. Please report route-level validation, for example the fraction of generated products whose recorded MCTS route is accepted by Syntheseus, or substantially temper the claim.
  2. [§4.4, §5.2.2] The primary evaluation metric, the ionizable lipid rate, is computed with the same property predictors that serve as the MCTS reward, making the absolute rates in Table 1 and Figure 4 circular. The predictors are validated only on known ionizable lipids (Section 5.2.1), whereas the generated molecules are novel and may fall outside the training distribution, so the reported rates could overstate the true quality of the generated molecules. The relative improvements over random generation and SyntheMol are still informative as comparisons on the same objective, but the paper should not present the absolute rates as evidence of 'high-quality ionizable lipids' without an independent evaluation, a distribution-shift analysis, or an explicit caveat that these are proxy scores.
  3. [§5.1] The comparison with the SyntheMol baseline is not compute-matched. The guided MCTS runs 10 MCTS instances per iteration, each with 10,000 simulations, for 10 iterations, totaling up to 1,000,000 simulations, whereas the SyntheMol baseline is a single MCTS run of 10,000 simulations. The paper's claim in the introduction of Section 5 that the method demonstrates 'enhanced efficiency' is therefore not supported by the reported experiments. Please provide comparisons at matched simulation budgets or report compute-normalized curves, especially because the headline improvement over SyntheMol may be partly attributable to a 100-fold larger search budget.
  4. [§5.1] The selection of the 200 testing head building blocks is not described. Since the main improvement over the baseline (test ionizable lipid rate of 0.7372 versus 0.2739 for SyntheMol) depends entirely on this test set, the paper should specify how these heads were chosen, confirm that they were held out from policy-network training data, and show results across multiple random head subsets to rule out selection bias.
minor comments (6)
  1. [§5.1] The equation numbering is inconsistent: the text cites 'Equation 5' and 'Equation 2' for the exploration weight, but the main text equations are numbered (1)-(3) and Appendix B repeats (4)-(5). Please renumber or reference consistently.
  2. [Appendix C, Algorithm 1] The pseudocode declares `SearchProbability()` as a required function, but this function is never called in any of the presented algorithms, and the `Product()` reaction predictor is not defined. Please clarify the roles of these functions and make the algorithm self-contained.
  3. [Appendix D] The custom policy loss is specified only as 'MAE or MSE may be applied'; please state the exact loss function, the temperature τ used in the search-probability calculation, and the regularization constant λ if it is used.
  4. [§5.2.3] There is a typo: 'the generate products' should be 'the generated products'. More substantively, the statement that the authors 'do not strictly adhere to those suggested by the generative model' should be justified, since this loosens the synthesis-path validation and makes the reported retro-valid rates harder to interpret.
  5. [§5.2.1] Reporting ROC-AUC and PR-AUC above 0.9999 after a single epoch is surprisingly high; please report the training/validation split, class balance, and confusion matrix, and check for potential label leakage between the generated lipid samples and the non-lipid PubChem samples.
  6. [§3.3] The description of the Syntheseus validation is incomplete: it is not stated how candidate retrosynthetic pathways are generated, how 'valid' is defined, or how many pathways are considered per product. Please provide these details for reproducibility.

Circularity Check

1 steps flagged · score 5.0 of 10

The headline ionizable-lipid rate is the MCTS reward itself, so the absolute quality claim is self-referential; relative baseline comparisons remain informative.

  1. self definitional [Section 4.4 (rollout/backpropagation) and Section 5.2.2 / Table 1 (evaluation)]
    "This randomly generated product will be evaluated by the property predictor and this property score will act as the value of the selected leaf node. ... The property score combines a lipid classifier score, which typically hovers near 0 or 1, and a binary ionizability score. A property score approaching 2 typically identifies the molecule as a likely ionizable lipid. ... only 4513 of the 16 477 generated products are predicted to be ionizable lipids, providing an ionizable lipid rate at 0.2739."

    The function PropertyScore() is both the reward that guides selection and backpropagation in MCTS (and the signal on which the policy network is trained via visit counts) and the definition of the evaluation metric 'unique ionizable lipid rate' in Table 1 and Figure 4. A generated product is counted as a 'predicted ionizable lipid' exactly when its property score exceeds the same threshold whose maximization drives generation. Hence the high reported test rate (0.7372) is a value of the optimized objective, not an independent measurement of lipid quality. The predictor was validated on 2,500+ published ionizable lipids, so the metric has some external anchor, but for novel generated structures the absolute 'high-quality ionizable lipids' claim is self-referential.

full rationale

The paper's only clear circular element is the identity between the reward used during generation and the metric used to report success: the lipid classifier and ionizability property score is the MCTS rollout value and the 'ionizable lipid rate' of Table 1. This makes the absolute rate an optimization outcome rather than an independent prediction. The relative improvement over random and SyntheMol baselines is still informative, and the property predictors were trained on separate data and validated against published ionizable lipids, so the circularity is partial, not total. I found no load-bearing self-citation: Syntheseus [Maziarz et al., 2023] is co-authored by a member of this author team, but it is an open-source package and is used to report unfavorable retro-valid rates, so it does not force the paper's conclusion. The larger correctness problem, rather than circularity, is that Table 1's Syntheseus retro-valid rate for guided-MCTS test products (0.2679) is below the random-combination rate (0.3103), and the conclusion claims 'available synthesis paths' without requiring the recorded forward route to match the retro route. The appendix also notes that the Lipid Analyzer toolkit used for dataset construction is unpublished, which limits reproducibility without being circular.

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

The central result rests on the property predictors and the reaction templates, neither of which is externally validated on the generated molecules. The headline rates are therefore conditional on the predictor assumptions holding for out-of-distribution structures.

free parameters (5)
  • Exploration weight c (guided MCTS) = 20
    Hand-set in Section 5.1; controls the UCB exploration term in Equation 3. No sensitivity analysis is reported, and the baseline uses c=10, so the comparison mixes hyperparameters with method changes.
  • Head building block subset and test set = ~12,000 heads sampled from 2.7 million; 200 test heads
    Section 5.1 states the subset is extracted without describing the sampling procedure. The composition of the subset and test set directly determines the achievable ionizable lipid rates.
  • Property score composition = lipid classifier score + binary ionizability score
    Section 5.2.2 defines the property score as the sum of the lipid classifier output and the binary ionizability label. This ad hoc reward is used both to guide MCTS and to compute the reported rates.
  • Tail similarity thresholds = GED <= 1 and unspecified fingerprint similarity
    Appendix A describes filtering lipid tails from ZINC20 by graph edit distance and Daylight/ECFP4 similarity, but does not give similarity cutoffs or justify them.
  • Policy network training schedule = 10 iterations, 20 epochs, lr 0.001, dropout 0.5
    Section 5.1 fixes these hyperparameters without sensitivity analysis; the reported improvement in test ionizable lipid rate depends on them.
assumptions (5)
  • domain assumption The 13 reaction templates used in SyntheMol remain applicable to the lipid building block dataset.
    Section 4.3 adopts the SyntheMol reaction templates with the claim that they are widely applicable, but lipid-specific reactions with long tails may be underrepresented, which the low Syntheseus retro valid rates in Table 1 suggest.
  • domain assumption MolGpKa pKa predictions and the Henderson-Hasselbalch net charge calculation correctly determine ionizability for novel lipid-like molecules.
    Section 4.2 relies on MolGpKa to decide ionizability at pH 7.4 and 5; the paper cites MolGpKa's validation but does not independently validate it on generated structures.
  • domain assumption The lipid classifier generalizes from its training distribution to the generated products.
    Section 4.2 trains a Chemprop classifier on known lipids, PubChem small molecules, and synthetic lipid-like molecules; Section 5.2.1 validates on known ionizable lipids, not on novel generated molecules.
  • ad hoc to paper The property score, defined as the lipid classifier score plus binary ionizability, is a valid proxy for a useful ionizable lipid.
    Section 5.2.2 uses this combination as the objective and as the success metric; no evidence connects it to transfection efficiency or biological performance.
  • domain assumption A two-tail lipid product is the correct termination condition for a useful ionizable lipid.
    The MCTS terminates when a two-tail lipid is generated, as specified in Algorithm 1, and the paper uses this as the definition of a final product; real ionizable lipids in LNPs often have more complex structures including linkers.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Generative Model for Synthesizing Ionizable Lipids: A Monte Carlo Tree Search Approach." pith.science (2026). https://pith.science/paper/WYB5KQYU

@misc{pith2026241200807,
  author       = {Pith},
  title        = {Pith review of: Generative Model for Synthesizing Ionizable Lipids: A Monte Carlo Tree Search Approach},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WYB5KQYU}},
  note         = {Machine review of arXiv:2412.00807}
}
read the original abstract

Ionizable lipids are essential in developing lipid nanoparticles (LNPs) for effective messenger RNA (mRNA) delivery. While traditional methods for designing new ionizable lipids are typically time-consuming, deep generative models have emerged as a powerful solution, significantly accelerating the molecular discovery process. However, a practical challenge arises as the molecular structures generated can often be difficult or infeasible to synthesize. This project explores Monte Carlo tree search (MCTS)-based generative models for synthesizable ionizable lipids. Leveraging a synthetically accessible lipid building block dataset and two specialized predictors to guide the search through chemical space, we introduce a policy network guided MCTS generative model capable of producing new ionizable lipids with available synthesis pathways.

Figures

Figures reproduced from arXiv: 2412.00807 by the authors.

Figure 1
Figure 1. Roadmap of lipid building block datasets construction. The upper flowchart illustrates the [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Workflow of policy network training [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Policy network guided Monte Carlo tree search for lipid generation. [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Ionizable lipid rate of the generated products during guided MCTS simulations. [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: The first row presents examples with no carboxyl group, one hydroxyl group, and two [PITH_FULL_IMAGE:figures/full_fig_p015_5.png]
Figure 5
Figure 5. Figure 5: Selected examples of lipid head building blocks. [PITH_FULL_IMAGE:figures/full_fig_p016_5.png]
Figure 6
Figure 6. Figure 6: Selected examples of lipid tail building blocks. [PITH_FULL_IMAGE:figures/full_fig_p016_6.png]
Figure 7
Figure 7. Figure 7: SyntheMol for lipid generation. B SyntheMol for Lipid Generation [PITH_FULL_IMAGE:figures/full_fig_p017_7.png]
Figure 8
Figure 8. Figure 8: Example 1 of generated ionizable lipid with synthesis path. [PITH_FULL_IMAGE:figures/full_fig_p022_8.png]
Figure 9
Figure 9. Figure 9: Example 2 of generated ionizable lipid with synthesis path. [PITH_FULL_IMAGE:figures/full_fig_p023_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

55 extracted references · 35 canonical work pages

  1. [1]

    Abd Elwakil, Ryota Suzuki, Alaa M

    Mahmoud M. Abd Elwakil, Ryota Suzuki, Alaa M. Khalifa, Rania M. Elshami, Takuya Isono, Yaser H.A. Elewa, Yusuke Sato, Takashi Nakamura, Toshifumi Satoh, and Hideyoshi Harashima. Harnessing topology and stereochemistry of glycidylamine-derived lipid nanoparticles for in vivo mrna delivery to immune cells in spleen and their application for cancer vaccinati...

    work page doi:10.1002/adfm.202303795 2023

  2. [2]

    Lucila Aimo, Robin Liechti, Nevila Hyka-Nouspikel, Anne Niknejad, Anne Gleizes, Lou Götz, Dmitry Kuznetsov, Fabrice P. A. David, F. Gisou van der Goot, Howard Riezman, Lydie Bougueleret, Ioannis Xenarios, and Alan Bridge. The swisslipids knowledgebase for lipid biology. Bioinformatics, 31 0 (17): 0 2860--2866, 2015. doi:10.1093/bioinformatics/btv285

    work page doi:10.1093/bioinformatics/btv285 2015

  3. [3]

    Rim Assouel, Mohamed Ahmed, Marwin H. S. Segler, Amir Saffari, and Yoshua Bengio. Defactor: Differentiable edge factorization-based probabilistic graph generation. CoRR, abs/1811.09766, 2018. URL http://arxiv.org/abs/1811.09766

    arXiv 2018

  4. [4]

    Deep convolutional generative adversarial network ( dcGAN ) models for screening and design of small molecules targeting cannabinoid receptors

    Yuemin Bian, Junmei Wang, Jaden Jungho Jun, and Xiang-Qun Xie. Deep convolutional generative adversarial network ( dcGAN ) models for screening and design of small molecules targeting cannabinoid receptors. Molecular Pharmaceutics, 16 0 (11): 0 4451--4460, 2019. ISSN 1543-8384. doi:10.1021/acs.molpharmaceut.9b00500. URL https://doi.org/10.1021/acs.molphar...

    work page doi:10.1021/acs.molpharmaceut.9b00500 2019

  5. [5]

    Kusner, Marwin H

    John Bradshaw, Brooks Paige, Matt J. Kusner, Marwin H. S. Segler, and José Miguel Hernández-Lobato. Barking up the right tree: an approach to search over molecule synthesis dags, 2020

    work page 2020

  6. [6]

    Carrasco, Suman Alishetty, Mohamad-Gabriel Alameh, Hooda Said, Lacey Wright, Mikell Paige, Ousamah Soliman, Drew Weissman, Thomas E

    Manuel J. Carrasco, Suman Alishetty, Mohamad-Gabriel Alameh, Hooda Said, Lacey Wright, Mikell Paige, Ousamah Soliman, Drew Weissman, Thomas E. Cleveland, Alexander Grishaev, and Michael D. Buschmann. Ionization and structural properties of mRNA lipid nanoparticles influence expression in intramuscular and intravascular administration. Communications Biolo...

    work page doi:10.1038/s42003-021-02441-2 2021

  7. [7]

    Efficient selectivity and backup operators in monte-carlo tree search

    R \'e mi Coulom. Efficient selectivity and backup operators in monte-carlo tree search. In Computers and Games, 2006. URL https://api.semanticscholar.org/CorpusID:16724115

    work page 2006

  8. [8]

    Isabelle M. S. Degors, Cuifeng Wang, Zia Ur Rehman, and Inge S. Zuhorn. Carriers break barriers in drug delivery: Endocytosis and endosomal escape of gene delivery vectors. Accounts of Chemical Research, 52 0 (7): 0 1750--1760, 2019. ISSN 0001-4842. doi:10.1021/acs.accounts.9b00177. URL https://doi.org/10.1021/acs.accounts.9b00177. Publisher: American Che...

    work page doi:10.1021/acs.accounts.9b00177 2019

Show all 55 references
  1. [9]

    Machine learning-guided lipid nanoparticle design for mrna delivery, 2023

    Daisy Yi Ding, Yuhui Zhang, Yuan Jia, and Jiuzhi Sun. Machine learning-guided lipid nanoparticle design for mrna delivery, 2023. URL https://arxiv.org/abs/2308.01402

    2023 arXiv

  2. [10]

    Estimation of synthetic accessibility score of drug-like molecules based on molecular complexity and fragment contributions

    Peter Ertl and Ansgar Schuffenhauer. Estimation of synthetic accessibility score of drug-like molecules based on molecular complexity and fragment contributions. Journal of Cheminformatics, 1 0 (1): 0 8, 2009. ISSN 1758-2946. doi:10.1186/1758-2946-1-8. URL https://doi.org/10.1...

    2009 doi

  3. [11]

    Visual feature extraction by a multilayered network of analog threshold elements

    Kunihiko Fukushima. Visual feature extraction by a multilayered network of analog threshold elements. IEEE Trans. Syst. Sci. Cybern., 5: 0 322--333, 1969. URL https://api.semanticscholar.org/CorpusID:206799280

    1969

  4. [12]

    Wenhao Gao and Connor W. Coley. The synthesizability of molecules proposed by generative models, 2020. URL https://arxiv.org/abs/2002.07007

    2020 arXiv

  5. [13]

    Grygorenko, Dmytro S

    Oleksandr O. Grygorenko, Dmytro S. Radchenko, Igor Dziuba, Alexander Chuprina, Kateryna E. Gubina, and Yurii S. Moroz. Generating multibillion chemical space of readily accessible screening compounds. iScience, 23 0 (11): 0 101681, 2020. doi:10.1016/j.isci.2020.101681. Publish...

    2020

  6. [14]

    Wei, David Duvenaud, José Miguel Hernández-Lobato, Benjamín Sánchez-Lengeling, Dennis Sheberla, Jorge Aguilera-Iparraguirre, Timothy D

    Rafael Gómez-Bombarelli, Jennifer N. Wei, David Duvenaud, José Miguel Hernández-Lobato, Benjamín Sánchez-Lengeling, Dennis Sheberla, Jorge Aguilera-Iparraguirre, Timothy D. Hirzel, Ryan P. Adams, and Alán Aspuru-Guzik. Automatic chemical design using a data-driven continuous r...

    2018 doi

  7. [15]

    A multidimensional approach to modulating ionizable lipids for high-performing and organ-selective mrna delivery

    Zepeng He, Zhicheng Le, Yi Shi, Lixin Liu, Zhijia Liu, and Yongming Chen. A multidimensional approach to modulating ionizable lipids for high-performing and organ-selective mrna delivery. Angewandte Chemie International Edition, 62 0 (43): 0 e202310401, 2023. doi:https://doi.o...

    2023 doi

  8. [16]

    Huayamares, Melissa P

    Sebastian G. Huayamares, Melissa P. Lokugamage, Regina Rab, Alejandro J. Da Silva Sanchez , Hyejin Kim, Afsane Radmand, David Loughrey, Liming Lian, Yuning Hou, Bhagelu R. Achyut, Annette Ehrhardt, Jeong S. Hong, Cory D. Sago, Kalina Paunovska, Elisa Schrader Echeverri, Daryll...

    2023

  9. [17]

    Irwin, Khanh G

    John J. Irwin, Khanh G. Tang, Jennifer Young, Chinzorig Dandarchuluun, Benjamin R. Wong, Munkhzul Khurelbaatar, Yurii S. Moroz, John Mayfield, and Roger A. Sayle. Z I N C 20—a free ultralarge-scale chemical database for ligand discovery. Journal of Chemical Information and Mod...

    2020 doi

  10. [18]

    Jaakkola

    Wengong Jin, Regina Barzilay, and Tommi S. Jaakkola. Hierarchical generation of molecular graphs using structural motifs. CoRR, abs/2002.03230, 2020. URL https://arxiv.org/abs/2002.03230

    2002 arXiv

  11. [19]

    M. Kim, M. Jeong, S. Hur, Y. Cho, J. Park, H. Jung, Y. Seo, H. A. Woo, K. T. Nam, K. Lee, and H. Lee. Engineered ionizable lipid nanoparticles for targeted delivery of rna therapeutics into different types of cells in the liver. Science Advances, 7 0 (9): 0 eabf4398, 2021. doi...

    2021 doi

  12. [20]

    Thiessen, Evan E

    Sunghwan Kim, Paul A. Thiessen, Evan E. Bolton, Jie Chen, Gang Fu, Asta Gindulyte, Lianyi Han, Jane He, Siqian He, Benjamin A. Shoemaker, Jiyao Wang, Bo Yu, Jian Zhang, and Stephen H. Bryant. Pubchem substance and compound databases. Nucleic Acids Research, 44 0 (D1): 0 D1202-...

    2016 doi

  13. [21]

    Kingma and Jimmy Ba

    Diederik P. Kingma and Jimmy Ba. Adam: A method for stochastic optimization, 2017. URL https://arxiv.org/abs/1412.6980

    2017 arXiv

  14. [22]

    Bandit based monte-carlo planning

    Levente Kocsis and Csaba Szepesvari. Bandit based monte-carlo planning. In European Conference on Machine Learning, 2006. URL https://api.semanticscholar.org/CorpusID:15184765

    2006

  15. [23]

    mrna: A promising platform for cancer immunotherapy

    Byoungjae Kong, Yelee Kim, Eun Hye Kim, Jung Soo Suk, and Yoosoo Yang. mrna: A promising platform for cancer immunotherapy. Advanced Drug Delivery Reviews, 199: 0 114993, 2023. ISSN 0169-409X. doi:https://doi.org/10.1016/j.addr.2023.114993. URL https://www.sciencedirect.com/sc...

    2023

  16. [24]

    The future of tissue-targeted lipid nanoparticle-mediated nucleic acid delivery

    Ruvanthi Kularatne, Rachael Crist, and Stephan Stern. The future of tissue-targeted lipid nanoparticle-mediated nucleic acid delivery. Pharmaceuticals, 15: 0 897, 07 2022. doi:10.3390/ph15070897

    2022 doi

  17. [25]

    Kusner, Brooks Paige, and José Miguel Hernández-Lobato

    Matt J. Kusner, Brooks Paige, and José Miguel Hernández-Lobato. Grammar variational autoencoder, 2017

    2017

  18. [26]

    Combinatorial design of nanoparticles for pulmonary mRNA delivery and genome editing

    Bowen Li, Rajith Singh Manan, Shun-Qing Liang, Akiva Gordon, Allen Jiang, Andrew Varley, Guangping Gao, Robert Langer, Wen Xue, and Daniel Anderson. Combinatorial design of nanoparticles for pulmonary mRNA delivery and genome editing. Nature Biotechnology, 41 0 (10): 0 1410--1...

    2023 doi

  19. [27]

    Accelerating ionizable lipid discovery for mrna delivery using machine learning and combinatorial chemistry

    Bowen Li, Idris Raji, Akiva Gordon, Lizhuang Sun, Theresa Raimondo, Favour Oladimeji, Allen Jiang, Andrew Varley, Robert Langer, and Daniel Anderson. Accelerating ionizable lipid discovery for mrna delivery using machine learning and combinatorial chemistry. Nature Materials, ...

    2024 doi

  20. [28]

    Johnson, Lukas Farbiak, and Daniel J

    Shuai Liu, Qiang Cheng, Tuo Wei, Xueliang Yu, Lindsay T. Johnson, Lukas Farbiak, and Daniel J. Siegwart. Membrane-destabilizing ionizable phospholipids for organ-selective mRNA delivery and CRISPR –cas gene editing. Nature Materials, 20 0 (5): 0 701--710, 2021. ISSN 1476-4660....

    2021 doi

  21. [29]

    Predicting molecular conformation via dynamic graph score matching

    Shitong Luo, Chence Shi, Minkai Xu, and Jian Tang. Predicting molecular conformation via dynamic graph score matching. In M. Ranzato, A. Beygelzimer, Y. Dauphin, P.S. Liang, and J. Wortman Vaughan, editors, Advances in Neural Information Processing Systems, volume 34, pages 19...

    2021

  22. [30]

    Han Han Ly, Simon Daniel, Shekinah K. V. Soriano, Zoltán Kis, and Anna K. Blakney. Optimization of lipid nanoparticles for sarna expression and cellular activation using a design-of-experiment approach. Molecular Pharmaceutics, 19 0 (6): 0 1892--1905, 2022. doi:10.1021/acs.mol...

    1905 doi

  23. [31]

    Re-evaluating retrosynthesis algorithms with syntheseus

    Krzysztof Maziarz, Austin Tripp, Guoqing Liu, Megan Stanley, Shufang Xie, Piotr Gai \'n ski, Philipp Seidl, and Marwin Segler. Re-evaluating retrosynthesis algorithms with syntheseus. In NeurIPS 2023 AI for Science Workshop, 2023. URL https://openreview.net/forum?id=W5U18rgtpg

    2023

  24. [32]

    Mitchell, Margaret M

    Michael J. Mitchell, Margaret M. Billingsley, Rebecca M. Haley, Marissa E. Wechsler, Nicholas A. Peppas, and Robert Langer. Engineering precision nanoparticles for drug delivery. Nature Reviews Drug Discovery, 20 0 (2): 0 101--124, 2021. ISSN 1474-1784. doi:10.1038/s41573-020-...

    2021 doi

  25. [33]

    Representations of lipid nanoparticles using large language models for transfection efficiency prediction

    Saeed Moayedpour, Jonathan Broadbent, Saleh Riahi, Michael Bailey, Hoa Thu, Dimitar Dobchev, Akshay Balsubramani, Ricardo Santos, Lorenzo Kogler-Anele, Alejandro Corrochano-Navarro, Sizhen Li, Fernando Montoya, Vikram Agarwal, Ziv Bar-Joseph, and Sven Jager. Representations of...

    2024 doi

  26. [34]

    Mol G pka: A web server for small molecule pka prediction using a graph-convolutional neural network

    Xiaolin Pan, Hao Wang, Cuiyu Li, John Zeng Hui Zhang, and Changge Ji. Mol G pka: A web server for small molecule pka prediction using a graph-convolutional neural network. Journal of chemical information and modeling, 2021. URL https://api.semanticscholar.org/CorpusID:235798026

    2021

  27. [35]

    Petrucci, William S

    Ralph H. Petrucci, William S. Harwood, and F. Geoffrey Herring. General Chemistry: Principles and Modern Applications. Prentice Hall, Upper Saddle River, NJ, 8th edition, 2002. ISBN 0-13-014329-4, 978-0-13-014329-7, 0-13-017677-X, 978-0-13-017677-6, 0-13-111673-8, 978-0-13-111...

    2002

  28. [36]

    Molecule edit graph attention network: Modeling chemical reactions as sequences of graph edits

    Mikolaj Sacha, Mikolaj Blaz, Piotr Byrski, Pawel Wlodarczyk - Pruszynski, and Stanislaw Jastrzebski. Molecule edit graph attention network: Modeling chemical reactions as sequences of graph edits. CoRR, abs/2006.15426, 2020. URL https://arxiv.org/abs/2006.15426

    2006 arXiv

  29. [37]

    mRNA -based therapeutics — developing a new class of drugs

    Ugur Sahin, Katalin Karikó, and Özlem Türeci. mRNA -based therapeutics — developing a new class of drugs. Nature Reviews Drug Discovery, 13 0 (10): 0 759--780, 2014. ISSN 1474-1784. doi:10.1038/nrd4278. URL https://doi.org/10.1038/nrd4278

    2014 doi

  30. [38]

    Development of a novel fingerprint for chemical reactions and its application to large-scale reaction classification and similarity

    Nadine Schneider, Daniel Lowe, Roger Sayle, and Gregory Landrum. Development of a novel fingerprint for chemical reactions and its application to large-scale reaction classification and similarity. Journal of chemical information and modeling, 55, 02 2015. doi:10.1021/acs.jcim.5b00046

    2015 doi

  31. [39]

    alphachem

    Marwin H. S. Segler, Mike Preu , and Mark P. Waller. Towards "alphachem": Chemical synthesis planning with tree search and deep neural network policies. CoRR, abs/1702.00020, 2017. URL http://arxiv.org/abs/1702.00020

    2017 arXiv

  32. [40]

    Marwin H. S. Segler, Mike Preuss, and Mark P. Waller. Planning chemical syntheses with deep neural networks and symbolic AI . Nature, 555 0 (7698): 0 604--610, 2018. ISSN 1476-4687. doi:10.1038/nature25978. URL https://doi.org/10.1038/nature25978

    2018 doi

  33. [41]

    Learning gradient fields for molecular conformation generation, 2021

    Chence Shi, Shitong Luo, Minkai Xu, and Jian Tang. Learning gradient fields for molecular conformation generation, 2021. URL https://arxiv.org/abs/2105.03902

    2021 arXiv

  34. [42]

    David Silver, Aja Huang, Chris J. Maddison, Arthur Guez, Laurent Sifre, George van den Driessche, Julian Schrittwieser, Ioannis Antonoglou, Veda Panneershelvam, Marc Lanctot, Sander Dieleman, Dominik Grewe, John Nham, Nal Kalchbrenner, Ilya Sutskever, Timothy Lillicrap, Madele...

    2016 doi

  35. [43]

    Mastering the game of go without human knowledge

    David Silver, Julian Schrittwieser, Karen Simonyan, Ioannis Antonoglou, Aja Huang, Arthur Guez, Thomas Hubert, Lucas Baker, Matthew Lai, Adrian Bolton, Yutian Chen, Timothy Lillicrap, Fan Hui, Laurent Sifre, George van den Driessche, Thore Graepel, and Demis Hassabis. Masterin...

    2017 doi

  36. [44]

    Graphvae: Towards generation of small graphs using variational autoencoders, 2018

    Martin Simonovsky and Nikos Komodakis. Graphvae: Towards generation of small graphs using variational autoencoders, 2018

    2018

  37. [45]

    Stern and Scott E

    Stephan T. Stern and Scott E. McNeil . Nanotechnology safety concerns revisited. Toxicological Sciences, 101 0 (1): 0 4--21, 2007. ISSN 1096-6080. doi:10.1093/toxsci/kfm169. URL https://doi.org/10.1093/toxsci/kfm169. \_eprint: https://academic.oup.com/toxsci/article-pdf/101/1/...

    2007 doi

  38. [46]

    Alex Brown, Edward A

    Manish Sud, Eoin Fahy, Deirdre Cotter, H. Alex Brown, Edward A. Dennis, Christopher K. Glass, Alfred H. Jr. Merrill, Robert C. Murphy, Christian R. H. Raetz, David W. Russell, and Shankar Subramaniam. L M S D : Lipid maps® structure database. Nucleic Acids Research, 2007. PMID...

    2007

  39. [47]

    Reinforcement learning: An introduction

    Richard S Sutton and Andrew G Barto. Reinforcement learning: An introduction. MIT press, 2018

    2018

  40. [48]

    Catacutan, Autumn Arnold, James Zou, and Jonathan M

    Kyle Swanson, Gary Liu, Denise B. Catacutan, Autumn Arnold, James Zou, and Jonathan M. Stokes. Generative ai for designing and validating easily synthesizable and structurally novel antibiotics. Nature Machine Intelligence, 6: 0 338--353, 2024. doi:10.1038/s42256-024-00809-7

    2024 doi

  41. [49]

    Regularization of neural networks using dropconnect

    Li Wan, Matthew Zeiler, Sixin Zhang, Yann Le Cun, and Rob Fergus. Regularization of neural networks using dropconnect. In Sanjoy Dasgupta and David McAllester, editors, Proceedings of the 30th International Conference on Machine Learning, volume 28 of Proceedings of Machine Le...

    2013

  42. [50]

    Johnson, Melissa L

    Tuo Wei, Yehui Sun, Qiang Cheng, Sumanta Chatterjee, Zachary Traylor, Lindsay T. Johnson, Melissa L. Coquelin, Jialu Wang, Michael J. Torres, Xizhen Lian, Xu Wang, Yufen Xiao, Craig A. Hodges, and Daniel J. Siegwart. Lung SORT LNPs enable precise homology-directed repair media...

    2023 doi

  43. [51]

    Visualizing lipid-formulated siRNA release from endosomes and target gene knockdown

    Anders Wittrup, Angela Ai, Xing Liu, Peter Hamar, Radiana Trifonova, Klaus Charisse, Muthiah Manoharan, Tomas Kirchhausen, and Judy Lieberman. Visualizing lipid-formulated siRNA release from endosomes and target gene knockdown. Nature Biotechnology, 33 0 (8): 0 870--876, 2015....

    2015 doi

  44. [52]

    Mark Saltzman, and Alexandra S

    Emily Xu, W. Mark Saltzman, and Alexandra S. Piotrowski-Daspit. Escaping the endosome: assessing cellular trafficking mechanisms of non-viral vehicles. Journal of Controlled Release, 335: 0 465--480, 2021. ISSN 0168-3659. doi:https://doi.org/10.1016/j.jconrel.2021.05.038. URL ...

    2021 doi

  45. [53]

    AGILE platform: a deep learning powered approach to accelerate LNP development for mRNA delivery

    Yue Xu, Shihao Ma, Haotian Cui, Jingan Chen, Shufen Xu, Fanglin Gong, Alex Golubovic, Muye Zhou, Kevin Chang Wang, Andrew Varley, Rick Xing Ze Lu, Bo Wang, and Bowen Li. AGILE platform: a deep learning powered approach to accelerate LNP development for mRNA delivery. Nature Co...

    2024 doi

  46. [54]

    Analyzing learned molecular representations for property prediction

    Kevin Yang, Kyle Swanson, Wengong Jin, Connor Coley, Philipp Eiden, Hua Gao, Angel Guzman-Perez, Timothy Hopper, Brian Kelley, Miriam Mathea, Andrew Palmer, Volker Settels, Tommi Jaakkola, Klavs Jensen, and Regina Barzilay. Analyzing learned molecular representations for prope...

    2019 doi

  47. [55]

    Siegwart

    Xueliang Yu, Shuai Liu, Qiang Cheng, Tuo Wei, Sang Lee, Di Zhang, and Daniel J. Siegwart. Lipid-modified aminoglycosides for mrna delivery to the liver. Advanced Healthcare Materials, 9 0 (7): 0 1901487, 2020. doi:https://doi.org/10.1002/adhm.201901487. URL https://onlinelibra...

    2020 doi

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.