BiBLDR: Bidirectional Behavior Learning for Drug Repositioning

Jianxin Wang; Mengyun Yang; Qichang Zhao; Renye Zhang

REVIEW 3 major objections 6 minor 53 references

BiBLDR: Bidirectional Behavior Learning for Drug Repositioning

T0 review · 3 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read BiBLDR recasts drug repositioning as bidirectional behavior-sequence learning and reports state-of-the-art AUROC and AUPRC on three benchmarks, with the largest gains on cold-start drugs.

desk verdict Interesting idea undermined by dataset inconsistencies and test-set hyperparameter tuning; the SOTA claim is not supported as written. read the letter →

arxiv 2505.23861 v1 pith:RXCPZ4HE submitted 2025-05-29 cs.LG cs.AI

Renye Zhang , Mengyun Yang , Qichang Zhao , Jianxin Wang This is my paper

classification cs.LGcs.AI

keywords drugrepositioningrepurposingbehaviorsequencerecommendationsystemTransformercold-startprototypespacedrug-diseaseassociation

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 tries to establish that drug repositioning can be treated as a behavior-sequence recommendation problem instead of a graph link-prediction problem, and that doing so yields the best reported performance on three standard benchmarks. The proposed model, BiBLDR, reads a drug's known indications and a disease's known treatments as two complementary behavioral sequences, combines them with similarity-derived prototype vectors, and passes them through a Transformer. On Gdataset, Cdataset, and LRSSL it reports AUROC values of 0.9941, 0.9978, and 0.9950 and AUPRC values of 0.9950, 0.9982, and 0.9964, beating seven baselines. The paper's distinctive claim is cold-start competence: for drugs with all known associations masked, BiBLDR reaches an AUPRC of 0.6194 on Gdataset while the next-best method reaches 0.3484.

What carries the argument

The load-bearing object is the bidirectional behavioral sequence: for a candidate pair, the drug-side sequence collects the diseases that the drug is known to treat in the training set, and the disease-side sequence collects the drugs that the disease is known to respond to, with the candidate itself excluded. These two sequences make the association matrix readable as a pair of complementary recommendation histories. Two additional mechanisms carry the method: stage I's prototype spaces, built by Siamese encoders that map similarity-matrix rows into vectors aligned by cosine similarity, and stage II's Transformer layer, adopted from behavior-sequence recommendation, which applies multi-head self-attention to the fused sequence features and is followed by an MLP association predictor. The paper also uses the temperature-weighted rating factor $e^{T\cdot A_{ki}}$ to amplify positive samples in the sequences.

What would settle it

Re-run the released implementation on the canonical LRSSL dataset described in Section III-A (763 drugs, 681 diseases, 3,051 associations) and on the canonical Cdataset, and compare against the same baselines; if BiBLDR no longer tops AUROC and AUPRC on those datasets, the central claim fails.

Watch

Extended reading notes

Core claim

The central claim is that replacing graph-based representations with bidirectional behavior sequences lets a model capture drug-disease interaction patterns that carry over to unseen drugs. BiBLDR defines, for each candidate pair (drug $u_k$, disease $v_m$), a drug-side sequence of diseases $u_k$ is known to treat and a disease-side sequence of drugs known to treat $v_m$, then trains a Transformer with multi-head self-attention over both sequences. Before that, stage I builds separate drug and disease prototype spaces by training Siamese encoders on similarity matrices with a cosine-similarity contrastive loss; stage II fuses prototypes, explicit similarity vectors, and learned ID embeddings, weights items in the sequence by a log-temperature factor $e^{T\cdot A_{ki}}$, and predicts the association with a multilayer network. The paper reports that this architecture outperforms prior non-deep and deep learning baselines on all three benchmark datasets, and it argues that the disease-side signal and pre-trained prototypes are what keep predictions robust when a drug has no known associations.

Load-bearing premise

The central claim assumes the experiments ran on the standard benchmark datasets whose sizes are reported, yet the text and the statistics table disagree about the Cdataset and LRSSL counts, so it is unclear which datasets produced the numbers.

Editorial extensions

If this is right

BiBLDR reports the best AUROC and AUPRC on Gdataset, Cdataset, and LRSSL, surpassing both non-deep and GCN-based baselines on every metric.
In the cold-start setting on Gdataset, BiBLDR reaches AUPRC 0.6194 versus the second-best 0.3484, indicating that disease-side sequences and prototypes carry usable signal for drugs with no known associations.
Under sparse training with only 10% of known associations available, the method keeps AUROC above 0.94 on all three datasets and AUPRC near 0.97 on Gdataset.
Ablation results show that removing either side of the bidirectional sequence, the prototype space, the similarity fusion, or the multi-head attention each lowers performance, with the disease-side sequence being the most critical.
Case studies on lung cancer and hypertension validate 8 of 10 and 9 of 10 top-ranked candidate drugs respectively, with molecular docking energies provided for the unvalidated candidates.

Reading between the lines

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

It follows from the architecture, though the paper does not isolate it, that the disease-side behavior sequence does most of the cold-start work; an experiment that ablates only the disease side while holding the drug side fixed would test whether all bidirectional information is necessary or mainly the disease history is.
If the reported near-ceiling AUROC values are reproducible on canonical splits, the practical implication is that similarity matrices plus recommendation-style sequence modeling may be a cheaper route to repurposing candidates than building large heterogeneous graphs.
The conflicting benchmark statistics in Table I mean that the released implementation should be re-run on the datasets described in Section III-A before the state-of-the-art comparison is used for downstream decisions.
A natural extension is to replace the pairwise similarity matrices with learned or text-based similarities, which would let the prototype stage generalize to drugs and diseases absent from precomputed similarity data.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

Dear colleague,

Here's the short version: the paper's core idea—treating drug repositioning as bidirectional behavior sequence learning with a Transformer—is genuinely interesting, but the experimental evidence for the SOTA claim falls apart under inspection. The dataset table contradicts the text for two of the three benchmarks, and the hyperparameters are tuned on the test folds. As written, the headline numbers can't be trusted.

What's new: instead of graph link prediction, they convert the drug-disease matrix into behavioral sequences from both the drug side and the disease side, and feed them through a BST-style Transformer. They add a Siamese prototype pre-training stage that uses drug-drug and disease-disease similarity to build entity prototypes, which are then fused with the sequence representations. That combination is novel as far as I know. The code is released, and the cold-start experiment—masking all associations for a drug—is a real attempt to test the practical scenario where graph methods struggle. The case studies with molecular docking add a bit of external validation, though they're anecdotal.

Now the problems. First, the dataset inconsistency is severe. Section III-A says Cdataset has 633 drugs, 409 diseases, and 2,352 associations; Table I reports 663, 409, and 2,532. For LRSSL, the text says 763 drugs, 681 diseases, 3,051 associations, but Table I lists 269 drugs, 598 diseases, and 18,416 associations (labeled 'LRRSL'). The Table I sparsity values match its own counts, so this isn't a simple typo. If the authors ran on the Table I versions, then the baselines aren't on the standard benchmarks and the SOTA claim is void. If they ran on the text versions, the table is wrong and the reader can't tell which dataset produced the LRSSL and Cdataset rows in Table II. Either way, the foundation is shaky.

Second, hyperparameters d0 and T are selected by evaluating on the test folds (Section III-G). That's test-set tuning. With no held-out validation, the reported numbers are optimistically biased. No error bars either, which matters when differences are as large as 0.9978 vs 0.9566.

Third, the cold-start experiment only uses Gdataset, and the paper doesn't explain how the baselines were adapted to a setting where the target drug has no training associations. The AUPRC gap is large, but without that detail, the comparison may be unfair.

Overall, the idea has merit and deserves referee time, but the current version should not be accepted as is. A serious editor should send it for peer review with clear instructions to fix the dataset reporting, redo hyperparameter selection on a validation set, report error bars, and describe the cold-start baseline setup. If the authors can do that, this could become a solid contribution. As it stands, the SOTA claim doesn't hold.

Best regards.

Referee Report

3 major / 6 minor

Summary. The paper proposes BiBLDR, a drug repositioning method that recasts drug-disease association prediction as a bidirectional behavior-sequence learning task. The method has two stages: first, drug and disease prototype spaces are learned from similarity matrices using Siamese networks; second, bidirectional behavior sequences (drug-side and disease-side) are combined with the prototypes and processed through a Transformer with multi-head self-attention to predict associations. Experiments are reported on Gdataset, Cdataset, and LRSSL under 10-fold cross-validation, with AUROC/AUPRC comparisons against seven baselines, ablation studies, cold-start experiments on masked drugs, sparse-environment tests, parameter analysis, and case studies with molecular docking. The central claims are state-of-the-art AUROC/AUPRC performance and significantly superior cold-start performance.

Significance. If the reported results are valid, BiBLDR would be a meaningful contribution to computational drug repositioning: the bidirectional behavior-sequence framing is a novel alternative to graph-based link prediction, and the cold-start and sparse-environment results target practically important scenarios. The paper also releases code and includes ablation studies and docking-based case studies, which are useful for reproducibility. However, the significance is currently contingent on experimental integrity: the dataset statistics contain a load-bearing inconsistency, and the hyperparameters are selected on the test folds. These issues prevent the state-of-the-art claim from being accepted at face value.

major comments (3)

[Section III-A and Table I] The benchmark statistics are internally inconsistent. The text states that Cdataset contains 633 drugs and 2,352 associations, while Table I reports 663 drugs and 2,532 associations. More seriously, the text places LRSSL at 763 drugs, 681 diseases, and 3,051 associations, whereas Table I lists 'LRRSL' with 269 drugs, 598 diseases, and 18,416 associations. The LRSSL discrepancy is not a typo, since 18,416/(269×598) ≈ 0.1145 matches the reported sparsity, whereas the standard LRSSL sparsity is about 0.0059. Because Table II reports BiBLDR's LRSSL AUROC/AUPRC, the reader cannot tell whether the state-of-the-art comparison was run on the standard LRSSL benchmark or on a different dense matrix. This must be resolved by stating which dataset was used, correcting Table I, and rerunning the comparison on the standard benchmark if necessary.
[Section III-G] Hyperparameters d0 and T are selected by direct evaluation on the benchmark test folds. The parameter analysis samples d0 and T and reports performance on Gdataset, Cdataset, and LRSSL, then fixes d0 = 2^10 and per-dataset T = {2, 2, 3}. Since the same folds are used to select these values and to produce Table II, the reported numbers are not independent predictions; they are the result of test-set model selection. A validation split or nested cross-validation must be introduced, and Table II and the ablation table should be regenerated under that protocol.
[Section III-B and Table II] The baseline comparison lacks implementation and tuning details. The paper does not state which code or version is used for SCMFDD, iDrug, BNNR, NRLMF, NIMCGCN, DRWBNCF, and PSGCN, how their hyperparameters were chosen, or whether the same 10 train/test folds were used. Without this information, the large margins in Table II (e.g., AUROC 0.9978 vs 0.9566 on Cdataset) cannot be attributed to the proposed method rather than to asymmetric tuning. Reporting the baseline parameter settings and exact data splits is necessary for the state-of-the-art claim.

minor comments (6)

[Table I] The dataset name 'LRRSL' should be 'LRSSL'.
[Equation (6)] The sentence 'W U_i and W V_i represent the i-th row of W U and the i-th row of W U' contains a typo; the second occurrence of W U should be W V.
[Algorithm 1] In the disease-side sequence process, 'I b_k ← Get disease-side behavior sequence of vm' uses the subscript k instead of m; this appears to be a copy-paste error.
[Section III-F] The word 'bahevioral' should be 'behavioral'.
[Section III-H and Figure 7] The text says the cyclosporine molecular docking target is Hexameric (PDB Code: 1P9M), but the caption of Figure 7(c) lists PDB code 1H1B; these should be reconciled.
[Section IV] The final paragraph contains a duplicated sentence fragment: 'Although cosine similarity reflects directional differences between vectors, incorporating.' This should be cleaned up.

Circularity Check

1 steps flagged · score 6.0 of 10

Evaluation is partially circular: the hyperparameters d0 and T are selected by running on the same benchmark test folds, so the reported state-of-the-art AUROC/AUPRC numbers are fitted rather than independent predictions; the LRSSL dataset inconsistency in Table I is a separate correctness risk, not a circularity.

fitted input called prediction [Section III-G (Parameter Analysis) and Table II]
"We sample d0 in the range of {2^6, 2^7, 2^8, 2^9, 2^10}, T in the range of {1, 2, 3, 4, 5}. We conduct experiments using these parameters on the benchmark datasets, and the results are shown in Fig. 6. ... For Gdataset, Cdataset, and LRSSL, the optimal T parameters are {2, 2, 3}."

The paper's only evaluation protocol, described in Section III-B, is 10-fold cross-validation with folds rotated as the test set; no validation split is ever introduced. Section III-G then searches d0 and T by running on the benchmark datasets and selecting the values with the highest AUROC/AUPRC in Fig. 6. These same datasets and folds produce the final metrics in Table II. Thus the reported SOTA performance is not an independent estimate: the configuration was chosen to maximize the very test-fold scores that are later presented as predictions, making the headline numbers partly fitted to the test data.

full rationale

The method itself is not circular in derivation: the bidirectional behavioral sequences exclude the target pair, the prototype spaces are trained on similarity matrices, and the final prediction is a learned function of those inputs. There is no self-citation chain or imported uniqueness theorem carrying the argument. The one substantive circularity is in the evaluation: Section III-G selects d0 and T by measuring performance on the benchmark datasets, and the selected values are then used for the reported 10-fold cross-validation numbers in Table II, with no held-out validation set described. That makes the central SOTA claim partially reduce to a test-set fit. Separately, Table I reports 'LRRSL' as 269 drugs, 598 diseases, and 18,416 associations (sparsity 0.1145), while Section III-A describes the LRSSL dataset as 763 drugs, 681 diseases, and 3,051 associations; this is a serious data-provenance and reproducibility concern, but it is a correctness issue rather than a circularity, so it does not raise the circularity score beyond the hyperparameter-fitting issue.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The method introduces no new entities. Its central claim depends on several fitted hyperparameters and on the assumption that the benchmark datasets and similarities are correctly curated. No entity beyond the bidirectional behavior sequence representation is postulated.

free parameters (4)

d0 (prototype dimension) = 1024 (2^10)
Tuned on test folds via Fig. 6(a,c,e); the reported SOTA uses 2^10.
T (temperature coefficient in Eq. 6) = 2 (Gdataset, Cdataset), 3 (LRSSL)
Selected per dataset by evaluating AUROC/AUPRC on the same test folds (Fig. 6).
Negative sample ratio = 1:1 (equal number of negatives to positives)
Chosen by hand; affects all AUC/PRC values and comparability across baselines.
Stage II learning rate = 0.0001
Set without stated validation; may affect performance.

assumptions (3)

standard math Multi-head self-attention and backpropagation are assumed.
Used in Transformer layer, Eq. 7-9.
domain assumption The drug-disease association matrix A and similarity matrices S^U, S^V contain sufficient signal to predict unknown associations.
Section II-B; the entire method is built on this input representation.
domain assumption Drug and disease similarity matrices are reliable and pre-computed from 2D fingerprints and phenotypic features.
Section III-A; no uncertainty or noise in similarities is modeled.

how reviews work

0 comments

Cite this review

Pith. "Pith review of BiBLDR: Bidirectional Behavior Learning for Drug Repositioning." pith.science (2026). https://pith.science/paper/RXCPZ4HE

@misc{pith2026250523861,
  author       = {Pith},
  title        = {Pith review of: BiBLDR: Bidirectional Behavior Learning for Drug Repositioning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RXCPZ4HE}},
  note         = {Machine review of arXiv:2505.23861}
}

read the original abstract

Drug repositioning aims to identify potential new indications for existing drugs to reduce the time and financial costs associated with developing new drugs. Most existing deep learning-based drug repositioning methods predominantly utilize graph-based representations. However, graph-based drug repositioning methods struggle to perform effective inference in cold-start scenarios involving novel drugs because of the lack of association information with the diseases. Unlike traditional graph-based approaches, we propose a bidirectional behavior learning strategy for drug repositioning, known as BiBLDR. This innovative framework redefines drug repositioning as a behavior sequential learning task to capture drug-disease interaction patterns. First, we construct bidirectional behavioral sequences based on drug and disease sides. The consideration of bidirectional information ensures a more meticulous and rigorous characterization of the behavioral sequences. Subsequently, we propose a two-stage strategy for drug repositioning. In the first stage, we construct prototype spaces to characterize the representational attributes of drugs and diseases. In the second stage, these refined prototypes and bidirectional behavior sequence data are leveraged to predict potential drug-disease associations. Based on this learning approach, the model can more robustly and precisely capture the interactive relationships between drug and disease features from bidirectional behavioral sequences. Extensive experiments demonstrate that our method achieves state-of-the-art performance on benchmark datasets. Meanwhile, BiBLDR demonstrates significantly superior performance compared to previous methods in cold-start scenarios. Our code is published in https://github.com/Renyeeah/BiBLDR.

Figures

Figures reproduced from arXiv: 2505.23861 by the authors.

**Figure 2.** The proposed BiBLDR framework. (a) Utilize similarity data to construct prototype spaces for drugs and diseases separately. (b) Utilize prototypes and bidirectional behavioral sequence information to predict drug-disease associations. a Siamese-based network, and cosine similarity aligns the prototype spaces. 1) Generating Initial Representations: The initial representation for each drug or disease is derived from … view at source ↗

**Figure 3.** The structure of the Transformer layer includes the multihead self-attention mechanism, normalization layer, feedforward neural network, and residual connections. A Transformer layer has a total of n attention heads, the input matrix X will be projected into n different lowdimensional subspaces by multiplying learnable weight matrices and then scaled dot-product attention will be computed, [PITH_FULL_IMAGE:figur… view at source ↗

Figures from the paper (3 more)

**Figure 4.** Figure 4: Comparison of cold-start experimental performance of different methods on the Gdataset. formation carried by the behavioral sequences constructed from the association matrix also diminishes accordingly. The reason BiBLDR can still achieve robust performance when the ba…

**Figure 6.** Figure 6: Parameter anlysis experiments on Gdataset, Cdataset, and LRSSL. 1) Lung Cancer (211980 in OMIM): Lung cancer is one of the cancers with the highest incidence and mortality rates worldwide, feared for its high fatality rate and lack of noticeable early symptoms. The le…

**Figure 7.** Figure 7: The molecular docking results of ligand molecules with target proteins [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

53 extracted references · 46 canonical work pages

[1]

A foundation model for clinician-centered drug repurposing,

K. Huang, P. Chandak, Q. Wang, S. Havaldar, A. Vaid, J. Leskovec, G. N. Nadkarni, B. S. Glicksberg, N. Gehlenborg, and M. Zitnik, “A foundation model for clinician-centered drug repurposing,” Nature Medicine, pp. 1–13, 2024

work page 2024
[2]

Drug repurposing for antimicrobial discovery,

M. A. Farha and E. D. Brown, “Drug repurposing for antimicrobial discovery,” Nature microbiology, vol. 4, no. 4, pp. 565–577, 2019

work page 2019
[3]

Research progress on the application of aspirin and its deriva- tives in the treatment and rehabilitation of cardiovascular diseases,

S. Ning, “Research progress on the application of aspirin and its deriva- tives in the treatment and rehabilitation of cardiovascular diseases,” in E3S Web of Conferences , vol. 553. EDP Sciences, 2024, p. 04010

work page 2024
[4]

Culprits of medication- induced telogen effluvium, part 2

D. Zhang, C. LaSenna, and B. E. Shields, “Culprits of medication- induced telogen effluvium, part 2.” Cutis, vol. 113, no. 1, pp. 11–14, 2024

work page 2024
[5]

Oral and intravenous 1’-cyano- substituted adenosine-like antivirals for early covid-19,

R. L. Gottlieb and R. Paredes, “Oral and intravenous 1’-cyano- substituted adenosine-like antivirals for early covid-19,” The Lancet Infectious Diseases, vol. 24, no. 2, pp. 108–110, 2024

work page 2024
[6]

Predicting drug-disease associations by using similarity constrained matrix factorization,

W. Zhang, X. Yue, W. Lin, W. Wu, R. Liu, F. Huang, and F. Liu, “Predicting drug-disease associations by using similarity constrained matrix factorization,” BMC bioinformatics, vol. 19, pp. 1–12, 2018

work page 2018
[7]

idrug: Integration of drug repositioning and drug-target prediction via cross-network embedding,

H. Chen, F. Cheng, and J. Li, “idrug: Integration of drug repositioning and drug-target prediction via cross-network embedding,” PLoS compu- tational biology, vol. 16, no. 7, p. e1008040, 2020

work page 2020
[8]

Drug repositioning based on bounded nuclear norm regularization,

M. Yang, H. Luo, Y . Li, and J. Wang, “Drug repositioning based on bounded nuclear norm regularization,” Bioinformatics, vol. 35, no. 14, pp. i455–i463, 2019

work page 2019

Show all 53 references

[9]

Neighborhood regular- ized logistic matrix factorization for drug-target interaction prediction,

Y . Liu, M. Wu, C. Miao, P. Zhao, and X.-L. Li, “Neighborhood regular- ized logistic matrix factorization for drug-target interaction prediction,” PLoS computational biology , vol. 12, no. 2, p. e1004760, 2016

2016
[10]

deepdr: a network-based deep learning approach to in silico drug repositioning,

X. Zeng, S. Zhu, X. Liu, Y . Zhou, R. Nussinov, and F. Cheng, “deepdr: a network-based deep learning approach to in silico drug repositioning,” Bioinformatics, vol. 35, no. 24, pp. 5191–5198, 2019

2019
[11]

A geometric deep learning framework for drug repositioning over heteroge- neous information networks,

B.-W. Zhao, X.-R. Su, P.-W. Hu, Y .-P. Ma, X. Zhou, and L. Hu, “A geometric deep learning framework for drug repositioning over heteroge- neous information networks,” Briefings in Bioinformatics, vol. 23, no. 6, p. bbac384, 2022

2022
[12]

Neural inductive matrix completion with graph convolutional networks for mirna-disease association prediction,

J. Li, S. Zhang, T. Liu, C. Ning, Z. Zhang, and W. Zhou, “Neural inductive matrix completion with graph convolutional networks for mirna-disease association prediction,” Bioinformatics, vol. 36, no. 8, pp. 2538–2546, 2020

2020
[13]

A weighted bilinear neural collaborative filtering approach for drug repositioning,

Y . Meng, C. Lu, M. Jin, J. Xu, X. Zeng, and J. Yang, “A weighted bilinear neural collaborative filtering approach for drug repositioning,” Briefings in bioinformatics , vol. 23, no. 2, p. bbab581, 2022

2022
[14]

Partner-specific drug reposi- tioning approach based on graph convolutional network,

X. Sun, B. Wang, J. Zhang, and M. Li, “Partner-specific drug reposi- tioning approach based on graph convolutional network,” IEEE Journal of Biomedical and Health Informatics , vol. 26, no. 11, pp. 5757–5765, 2022

2022
[15]

Drug repositioning with adaptive graph convolutional networks,

X. Sun, X. Jia, Z. Lu, J. Tang, and M. Li, “Drug repositioning with adaptive graph convolutional networks,” Bioinformatics, vol. 40, no. 1, p. btad748, 2024

2024
[16]

Mining drug–disease relationships as a complement to medical genetics-based drug reposition- ing: Where a recommendation system meets genome-wide association studies,

H. Wang, Q. Gu, J. Wei, Z. Cao, and Q. Liu, “Mining drug–disease relationships as a complement to medical genetics-based drug reposition- ing: Where a recommendation system meets genome-wide association studies,” Clinical Pharmacology & Therapeutics , vol. 97, no. 5, pp. 451–454, 2015

2015
[17]

Realizing drug repositioning by adapting a recommendation system to handle the process,

M. G. Ozsoy, T. ¨Ozyer, F. Polat, and R. Alhajj, “Realizing drug repositioning by adapting a recommendation system to handle the process,” BMC bioinformatics, vol. 19, pp. 1–14, 2018

2018
[18]

Deep learning based rec- ommender system: A survey and new perspectives,

S. Zhang, L. Yao, A. Sun, and Y . Tay, “Deep learning based rec- ommender system: A survey and new perspectives,” ACM computing surveys (CSUR), vol. 52, no. 1, pp. 1–38, 2019

2019
[19]

An intelligent recommendation system in e- commerce using ensemble learning,

A. Shankar, P. Perumal, M. Subramanian, N. Ramu, D. Natesan, V . R. Kulkarni, and T. Stephan, “An intelligent recommendation system in e- commerce using ensemble learning,”Multimedia Tools and Applications, vol. 83, no. 16, pp. 48 521–48 537, 2024

2024
[20]

A survey of graph neural networks for social recommender systems,

K. Sharma, Y .-C. Lee, S. Nambi, A. Salian, S. Shah, S.-W. Kim, and S. Kumar, “A survey of graph neural networks for social recommender systems,” ACM Computing Surveys , vol. 56, no. 10, pp. 1–34, 2024

2024
[21]

Deepfm: a factorization- machine based neural network for ctr prediction,

H. Guo, R. Tang, Y . Ye, Z. Li, and X. He, “Deepfm: a factorization- machine based neural network for ctr prediction,” arXiv preprint arXiv:1703.04247, 2017

2017 arXiv
[22]

Contrastive learning for sequential recommendation,

X. Xie, F. Sun, Z. Liu, S. Wu, J. Gao, J. Zhang, B. Ding, and B. Cui, “Contrastive learning for sequential recommendation,” in 2022 IEEE 38th international conference on data engineering (ICDE). IEEE, 2022, pp. 1259–1273

2022
[23]

Attention is all you need,

A. Vaswani, “Attention is all you need,” Advances in Neural Information Processing Systems, 2017

2017
[24]

Bert: Pre-training of deep bidirectional transformers for language understanding,

J. D. M.-W. C. Kenton and L. K. Toutanova, “Bert: Pre-training of deep bidirectional transformers for language understanding,” in Proceedings of naacL-HLT, vol. 1, no. 2. Minneapolis, Minnesota, 2019

2019
[25]

Language mod- els are few-shot learners,

T. Brown, B. Mann, N. Ryder, M. Subbiah, J. D. Kaplan, P. Dhariwal, A. Neelakantan, P. Shyam, G. Sastry, A. Askell et al., “Language mod- els are few-shot learners,” Advances in neural information processing systems, vol. 33, pp. 1877–1901, 2020

1901
[26]

Gpt-4 technical report,

J. Achiam, S. Adler, S. Agarwal, L. Ahmad, I. Akkaya, F. L. Aleman, D. Almeida, J. Altenschmidt, S. Altman, S. Anadkat et al. , “Gpt-4 technical report,” arXiv preprint arXiv:2303.08774 , 2023

2023 arXiv
[27]

Bert4rec: Sequential recommendation with bidirectional encoder representations from transformer,

F. Sun, J. Liu, J. Wu, C. Pei, X. Lin, W. Ou, and P. Jiang, “Bert4rec: Sequential recommendation with bidirectional encoder representations from transformer,” in Proceedings of the 28th ACM international confer- ence on information and knowledge management , 2019, pp. 1441–1450

2019
[28]

Behavior sequence transformer for e-commerce recommendation in alibaba,

Q. Chen, H. Zhao, W. Li, P. Huang, and W. Ou, “Behavior sequence transformer for e-commerce recommendation in alibaba,” inProceedings of the 1st international workshop on deep learning practice for high- dimensional sparse data , 2019, pp. 1–4

2019
[29]

Personalized behavior-aware transformer for multi-behavior sequential recommenda- tion,

J. Su, C. Chen, Z. Lin, X. Li, W. Liu, and X. Zheng, “Personalized behavior-aware transformer for multi-behavior sequential recommenda- tion,” in Proceedings of the 31st ACM International Conference on Multimedia, 2023, pp. 6321–6331

2023
[30]

A survey on siamese network: Methodologies, applications, and opportunities,

Y . Li, C. P. Chen, and T. Zhang, “A survey on siamese network: Methodologies, applications, and opportunities,” IEEE Transactions on artificial intelligence, vol. 3, no. 6, pp. 994–1014, 2022

2022
[31]

Predict: a method for inferring novel drug indications with application to personalized medicine,

A. Gottlieb, G. Y . Stein, E. Ruppin, and R. Sharan, “Predict: a method for inferring novel drug indications with application to personalized medicine,” Molecular systems biology , vol. 7, no. 1, p. 496, 2011

2011
[32]

Drug repositioning based on comprehensive similarity measures and bi-random walk algorithm,

H. Luo, J. Wang, M. Li, J. Luo, X. Peng, F.-X. Wu, and Y . Pan, “Drug repositioning based on comprehensive similarity measures and bi-random walk algorithm,” Bioinformatics, vol. 32, no. 17, pp. 2664– 2671, 2016

2016
[33]

Lrssl: predict and interpret drug–disease associations based on data integration using sparse subspace learning,

X. Liang, P. Zhang, L. Yan, Y . Fu, F. Peng, L. Qu, M. Shao, Y . Chen, and Z. Chen, “Lrssl: predict and interpret drug–disease associations based on data integration using sparse subspace learning,” Bioinformatics, vol. 33, no. 8, pp. 1187–1196, 2017

2017
[34]

Drugbank: a comprehensive resource for in silico drug discovery and exploration,

D. S. Wishart, C. Knox, A. C. Guo, S. Shrivastava, M. Hassanali, P. Stothard, Z. Chang, and J. Woolsey, “Drugbank: a comprehensive resource for in silico drug discovery and exploration,” Nucleic acids research, vol. 34, no. suppl 1, pp. D668–D672, 2006

2006
[35]

Online mendelian inheritance in man (omim), a knowledge- base of human genes and genetic disorders,

A. Hamosh, A. F. Scott, J. S. Amberger, C. A. Bocchini, and V . A. McKusick, “Online mendelian inheritance in man (omim), a knowledge- base of human genes and genetic disorders,” Nucleic acids research , vol. 33, no. suppl 1, pp. D514–D517, 2005

2005
[36]

The 12 IEEE TRANSACTIONS AND JOURNALS TEMPLA TE comparative toxicogenomics database: update 2017,

A. P. Davis, C. J. Grondin, R. J. Johnson, D. Sciaky, B. L. King, R. McMorran, J. Wiegers, T. C. Wiegers, and C. J. Mattingly, “The 12 IEEE TRANSACTIONS AND JOURNALS TEMPLA TE comparative toxicogenomics database: update 2017,” Nucleic acids research, vol. 45, no. D1, pp. D972–...

2017
[37]

Etoposide as a key therapeutic agent in lung cancer: Mechanisms, efficacy, and emerging strategies,

J. Y . Jang, D. Kim, E. Im, and N. D. Kim, “Etoposide as a key therapeutic agent in lung cancer: Mechanisms, efficacy, and emerging strategies,” International Journal of Molecular Sciences , vol. 26, no. 2, p. 796, 2025

2025
[38]

Felodipine in hypertension,

P. Mace, T. Stallard, and W. Littler, “Felodipine in hypertension,” European journal of clinical pharmacology, vol. 29, pp. 383–389, 1985

1985
[39]

Topotecan for relapsed small-cell lung cancer: systematic review and meta-analysis of 1347 patients,

N. Horita, M. Yamamoto, T. Sato, T. Tsukahara, H. Nagakura, K. Tashiro, Y . Shibata, H. Watanabe, K. Nagai, M. Inoue et al. , “Topotecan for relapsed small-cell lung cancer: systematic review and meta-analysis of 1347 patients,” Scientific reports, vol. 5, no. 1, p. 15437, 2015

2015
[40]

Hypotensive effects of a calcium antagonist, bepridil, in patients with essential hypertension,

K. AOKI and K. SATO, “Hypotensive effects of a calcium antagonist, bepridil, in patients with essential hypertension,” Japanese heart journal, vol. 27, no. 5, pp. 693–699, 1986

1986
[41]

Methotrexate in the chemotherapy of lung cancer

J. Smyth and H. Ford, “Methotrexate in the chemotherapy of lung cancer.” Cancer Treatment Reports, vol. 65, pp. 161–163, 1981

1981
[42]

Hypotensive effects of clonidine and chlorthalidone: Controlled clinical trial of drugs administered singly and in combination,

D. B. Toubes, T. J. McIntosh, W. M. Kirkendall, and W. R. Wilson, “Hypotensive effects of clonidine and chlorthalidone: Controlled clinical trial of drugs administered singly and in combination,” American Heart Journal, vol. 82, no. 3, pp. 312–318, 1971

1971
[43]

Hydroxychloroquine reduces hypertension and soluble fms-like kinase-1 in a n ω-nitro-l-arginine methyl ester-induced preeclampsia rat model,

M. Choi, J. R. Hwang, J.-H. Sung, N. Byun, Y . S. Seok, G. J. Cho, S.-J. Choi, J.-S. Kim, S.-y. Oh, and C.-R. Roh, “Hydroxychloroquine reduces hypertension and soluble fms-like kinase-1 in a n ω-nitro-l-arginine methyl ester-induced preeclampsia rat model,” Journal of Hyperten...

2022
[44]

Irinotecan as a second-line monotherapy for small cell lung cancer,

A. Sevinc, M. E. Kalender, M. Altinbas, M. Ozkan, M. Dikilitas, C. Camci, A. S. of Medical Oncology (ASMO) et al. , “Irinotecan as a second-line monotherapy for small cell lung cancer,” Asian Pac J Cancer Prev, vol. 12, no. 4, pp. 1055–1059, 2011

2011
[45]

Indapamide: is it the better diuretic for hypertension?

N. M. Kaplan, “Indapamide: is it the better diuretic for hypertension?” pp. 983–984, 2015

2015
[46]

Temozolomide in patients with advanced non-small cell lung cancer with and without brain metastases: a phase ii study of the eortc lung cancer group (08965),

R. Dziadziuszko, A. Ardizzoni, P. Postmus, E. Smit, A. Price, C. De- bruyne, C. Legrand, G. Giaccone, E. L. C. Group et al., “Temozolomide in patients with advanced non-small cell lung cancer with and without brain metastases: a phase ii study of the eortc lung cancer group (0...

2003
[47]

Fenoldopam—a selec- tive peripheral dopamine-receptor agonist for the treatment of severe hypertension,

M. B. Murphy, C. Murray, and G. D. Shorten, “Fenoldopam—a selec- tive peripheral dopamine-receptor agonist for the treatment of severe hypertension,” New England Journal of Medicine , vol. 345, no. 21, pp. 1548–1557, 2001

2001
[48]

Sunitinib benefits patients with small-cell lung cancer,

S. Tanday, “Sunitinib benefits patients with small-cell lung cancer,” The Lancet Oncology, vol. 16, no. 4, p. e164, 2015

2015
[49]

Inhibition of tumor metastasis by targeted daunorubicin and dioscin codelivery liposomes modified with pfv for the treatment of non-small-cell lung cancer,

Y . Wang, M. Fu, J. Liu, Y . Yang, Y . Yu, J. Li, W. Pan, L. Fan, G. Li, X. Li et al. , “Inhibition of tumor metastasis by targeted daunorubicin and dioscin codelivery liposomes modified with pfv for the treatment of non-small-cell lung cancer,” International journal of nanome...

2019
[50]

Ramipril: an updated review of its therapeutic use in essential hypertension and heart failure,

J. E. Frampton and D. H. Peters, “Ramipril: an updated review of its therapeutic use in essential hypertension and heart failure,” Drugs, vol. 49, pp. 440–466, 1995

1995
[51]

S. J. Sciegienka, S. R. Solst, K. C. Falls, J. D. Schoenfeld, A. R. Klinger, N. L. Ross, S. N. Rodman, D. R. Spitz, and M. A. Fath, “D-penicillamine combined with inhibitors of hydroperoxide metabolism enhances lung and breast cancer cell responses to radiation and carboplatin...

2017
[52]

Chlorothiazide in hypertensive and normotensive patients,

E. D. Freis, A. Wanko, I. M. Wilson, A. E. Parrish et al., “Chlorothiazide in hypertensive and normotensive patients,” Ann NY Acad Sci , vol. 71, no. 4, pp. 450–455, 1958

1958
[53]

Antihypertensive and metabolic effects of concomitant administration of terazosin and methyclothiazide for the treatment of essential hypertension,

H. R. Black, S. G. Chrysant, C. L. Curry, W. H. Frishman, R. H. Grimm, K. C. Lasseter, R. Okun, J. L. Pool, V . Raizada, N. D. Vlachakis et al., “Antihypertensive and metabolic effects of concomitant administration of terazosin and methyclothiazide for the treatment of essenti...

1992

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

A foundation model for clinician-centered drug repurposing,

K. Huang, P. Chandak, Q. Wang, S. Havaldar, A. Vaid, J. Leskovec, G. N. Nadkarni, B. S. Glicksberg, N. Gehlenborg, and M. Zitnik, “A foundation model for clinician-centered drug repurposing,” Nature Medicine, pp. 1–13, 2024

work page 2024

[2] [2]

Drug repurposing for antimicrobial discovery,

M. A. Farha and E. D. Brown, “Drug repurposing for antimicrobial discovery,” Nature microbiology, vol. 4, no. 4, pp. 565–577, 2019

work page 2019

[3] [3]

Research progress on the application of aspirin and its deriva- tives in the treatment and rehabilitation of cardiovascular diseases,

S. Ning, “Research progress on the application of aspirin and its deriva- tives in the treatment and rehabilitation of cardiovascular diseases,” in E3S Web of Conferences , vol. 553. EDP Sciences, 2024, p. 04010

work page 2024

[4] [4]

Culprits of medication- induced telogen effluvium, part 2

D. Zhang, C. LaSenna, and B. E. Shields, “Culprits of medication- induced telogen effluvium, part 2.” Cutis, vol. 113, no. 1, pp. 11–14, 2024

work page 2024

[5] [5]

Oral and intravenous 1’-cyano- substituted adenosine-like antivirals for early covid-19,

R. L. Gottlieb and R. Paredes, “Oral and intravenous 1’-cyano- substituted adenosine-like antivirals for early covid-19,” The Lancet Infectious Diseases, vol. 24, no. 2, pp. 108–110, 2024

work page 2024

[6] [6]

Predicting drug-disease associations by using similarity constrained matrix factorization,

W. Zhang, X. Yue, W. Lin, W. Wu, R. Liu, F. Huang, and F. Liu, “Predicting drug-disease associations by using similarity constrained matrix factorization,” BMC bioinformatics, vol. 19, pp. 1–12, 2018

work page 2018

[7] [7]

idrug: Integration of drug repositioning and drug-target prediction via cross-network embedding,

H. Chen, F. Cheng, and J. Li, “idrug: Integration of drug repositioning and drug-target prediction via cross-network embedding,” PLoS compu- tational biology, vol. 16, no. 7, p. e1008040, 2020

work page 2020

[8] [8]

Drug repositioning based on bounded nuclear norm regularization,

M. Yang, H. Luo, Y . Li, and J. Wang, “Drug repositioning based on bounded nuclear norm regularization,” Bioinformatics, vol. 35, no. 14, pp. i455–i463, 2019

work page 2019