Pith. sign in

REVIEW 3 major objections 4 minor 39 references

A Hybrid Supervised and Self-Supervised Graph Neural Network for Edge-Centric Applications

T0 review · 3 major / 4 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read The paper claims that a single graph neural network with node-and-edge attention and a hybrid supervised/self-supervised loss matches or beats specialized predictors for protein-protein interaction, gene function, and compound similarity.

desk verdict A genuinely useful edge-centric GNN with solid PPI/GO experiments, but the compound-similarity headline rests on an undefined ground truth for unknown-structure pairs. read the letter →

arxiv 2501.12309 v2 pith:AS65WWKB submitted 2025-01-21 cs.LG q-bio.MN

Eugenio Borzone , Leandro Di Persia , Matias Gerard This is my paper · ORCID
classification cs.LGq-bio.MN
keywords graphneuralnetworksedge-centricpredictionnode-edgeattentionhybridlossself-supervisedlearningprotein-proteininteractiongenefunctioncompoundsimilarity
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper argues that one graph architecture can handle edge-centric tasks — predicting properties of pairs of nodes — across biology. The architecture is a message-passing layer that attends over both node and edge features, followed by an MLP predictor, trained end-to-end with a hybrid loss that joins a supervised prediction term with two cosine-similarity terms. This lets the model learn from unlabeled subgraphs and from sparse one-hot node features, which is why the authors claim it can score compounds whose structures are unknown. A reader should care because edge-centric predictions are common in bioinformatics, and most graph networks under-use edge attributes and require dense labels.

What carries the argument

The central object is the NodeEdgeAttentionConv (NEAConv) layer, a message-passing update in which the attention weight is $w = \mathrm{sum}(Q \odot K)$ across the feature dimension, with $Q$ formed from the target node and edge features and $K,V$ formed from source node features; the update concatenates the weighted aggregation with the original tokenized features and applies a Tanh activation. The second load-bearing object is the loss $L = \alpha L_{\text{supervised}} + \beta L_{\text{cosine}} + \gamma L_{\text{cosine pred}}$. The cosine term pushes the cosine similarity of the two central-node embeddings toward the label, and the self-supervised prediction term aligns the model's output with that embedding similarity, which is what lets one-hot placeholder nodes and unlabeled pairs still organize the latent space.

What would settle it

Recompute the compound-similarity mean absolute error after removing the 5,390 pairs that involve at least one compound with unknown structure; if the reported 0.013 disappears or degrades, the claim that the model solves unknown-structure similarity is not supported by the stated experiment. A second check is to train the same model without the self-supervised term and compare the embedding organization of unknown-structure compounds; if the PCA map is essentially unchanged, that loss component is not doing the work the paper attributes to it.

Watch

Extended reading notes

Core claim

The central claim is that the proposed NEAConv architecture with the hybrid loss of Eq. 5 is a competitive general-purpose predictor for relations between node pairs. NEAConv computes attention weights as the summed Hadamard product of query and key vectors, builds the query from the target node together with edge features, and updates a central node by concatenating the aggregated neighbor messages with its tokenized features. The hybrid loss couples supervised prediction error, error between the cosine similarity of the two node embeddings and the label, and a self-supervised term that aligns the network's prediction with that embedding similarity. Empirically, the model achieves the highest average rank across five protein-protein interaction datasets, ties with exp2GO on gene function prediction with no statistically significant difference, and reaches a compound-similarity mean absolute error of 0.013 including pairs whose compounds lack structures.

Load-bearing premise

The broadest compound-similarity claim assumes that a fingerprint-based similarity label exists for every pair in the 21,528-pair dataset, including the 5,390 pairs where at least one compound has no usable structure; the paper itself says those fingerprints cannot be created, so the target for those pairs is undefined.

Editorial extensions

If this is right

  • On the five PPI benchmarks, the model ranks first among the four compared methods, and the gap to the second-place method is statistically significant.
  • On gene-function prediction, the model is statistically tied with exp2GO and ahead of two other deep learners, indicating that expression-similarity graphs can drive function transfer.
  • On compound similarity, the model reaches a mean absolute error of 0.013 and produces a PCA embedding map in which similarity to a reference compound fades smoothly with distance.
  • The ablation study shows that the node-and-edge attention and the hybrid loss each reduce error on their own, and together they tighten the relationship between embedding distance and predicted similarity.
  • Because the self-supervised term is active for unlabeled subgraphs, the training procedure can potentially exploit pairs that have no ground-truth label.

Reading between the lines

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

  • An extension not explored in the paper: the same architecture should transfer to non-biological edge-regression problems such as drug-target affinity or social tie strength, because neither the attention update nor the cosine-alignment loss is biology-specific.
  • One testable extension would be to replace cosine similarity in the loss with other pairwise measures, such as Euclidean distance or dot product, and see whether the latent-space organization and error behavior change accordingly.
  • A further test the paper leaves open: mask a random subset of known compound structures during training and ask whether their one-hot embeddings still land near structural neighbors in the PCA map, which would directly quantify the claimed solution for unknown structures.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper proposes NEAConv, a message-passing graph neural network for edge-centric tasks. The architecture consists of a tokenizer MLP, a single attention-based convolution layer that uses both node and edge features, and a prediction MLP; training uses a hybrid loss that combines a supervised prediction term, a supervised cosine-similarity term on the central node embeddings, and a self-supervised term that aligns the prediction with the embedding cosine similarity. Experiments are reported on PPI prediction for five organisms, GO term prediction for three species, and compound similarity on six KEGG pathways. The authors report state-of-the-art or competitive results on PPI and GO, an MAE of 0.013 on compound similarity, and claim that the model provides a solution for predicting similarity between compounds with unknown structures.

Significance. If the results hold, the paper makes a useful contribution: a single architecture with attention over node and edge features plus a hybrid supervised/self-supervised loss is competitive across three bioinformatics benchmarks. The PPI evaluation is a strength, using 10x5-fold cross-validation, repeated runs, and statistical tests; the GO comparison uses standard CAFA-style evaluation; and the ablation study on attention and loss components directly probes the contribution of each design element. However, the strongest advertised claim, namely quantitative similarity prediction for compounds with unknown structures, is not supported by the reported evaluation because the ground-truth Tanimoto labels for the 5,390 pairs involving at least one unknown compound are never defined. The architecture and the PPI/GO results may still be defensible, but the compound-similarity experiment needs to be substantially clarified or its claims restricted.

major comments (3)
  1. [III-C and IV-C] The headline MAE of 0.013 is not a valid validation of the unknown-structure claim as reported. Section III-C states that edge weights are Tanimoto coefficients computed from MACCS fingerprints, while Section IV-C and Figure 9 state that for compounds with generic '-R' substituents 'it is impossible to create a fingerprint.' The dataset contains 21,528 pairs, including 5,390 pairs involving at least one unknown compound, but the paper never defines what ground-truth label is used for those pairs, nor whether they are excluded from the MAE calculation. If they are excluded, the MAE covers only known-known pairs and the unknown-structure claim rests on the qualitative Figure 9 examples and the PCA plot; if they are included with placeholder or imputed labels, the target is undefined and the number is not meaningful. The authors should state the labeling rule, report MAE separately for known-known and unknown-involved pairs, or temper the abstract claim accordingly.
  2. [IV-B] The GO model-selection rule is stated as 'Five runs were performed with different initialization seeds, and the run with the lowest validation F1 score was selected.' If literal, this is the opposite of standard model selection and would select the worst validation result, which would make the reported F1max comparisons difficult to interpret. Please clarify whether this is a typo (e.g., lowest validation loss or highest validation F1 was intended), report the criterion actually used, and provide the variance across the five seeds so the Nemenyi test in Figure 7 can be meaningfully assessed.
  3. [IV-B] Baseline provenance is unstated for the GO comparison. It is not clear whether exp2GO, deepGOplus, NMFGO, and BLAST were rerun under the same training/validation folds and the same 2016/2017 GOA files, or whether their F1max values were taken directly from the original publications. No error bars or seed-level variability are reported for any GO method. Without this information, the claim of no statistically significant difference between the proposed model and exp2GO cannot be independently checked, because the Friedman/Nemenyi procedure requires comparable samples from the same experimental protocol.
minor comments (4)
  1. [II-A] The notation in 'Let G = {v, e} be a graph, where v is the set of |v| nodes' reuses the same symbol for a set and its cardinality; using uppercase V for the node set and E for the edge set would avoid ambiguity.
  2. [II-C] In Equation (3), the message vector m_{k,j} is used before it is formally defined; please define it explicitly in the text that introduces Equation (2), or state that it is the value vector V of node k weighted by the attention scalar w.
  3. [IV-A] The caption of Figure 4 says 'the line inside the box denotes the mean F1 of each k-fold,' but the boxplots aggregate 10 repetitions of 5-fold cross-validation; the line should more precisely be described as the mean over repetitions for each fold, and the text would benefit from reporting the numeric mean and standard deviation of F1 for each model.
  4. [II-E] There are small language and consistency issues: 'the proposed loss function is build' should be 'is built'; 'exp2go' is sometimes written with lowercase 'go' (e.g., in the Section IV-B text) while the reference and figure use 'exp2GO'; and the claim that the model 'outperforms state-of-the-art methods' should be phrased more carefully for the GO task, where the authors themselves report no significant difference with exp2GO.

Circularity Check

1 steps flagged · score 6.0 of 10

Compound-similarity 'prediction' is trained and evaluated on the Tanimoto coefficient that is also supplied as an input edge attribute, so the MAE=0.013 result reduces to copying the input and does not validate the unknown-structure claim.

  1. self definitional [Section III-C (Metabolic Pathways) and Section IV-C (Compounds similarity prediction); Eq. (5)-(6)]
    "Edge weights are Tanimoto coefficients [30], and enzyme-family one-hot vectors (EC 1–7) supply additional edge attributes. ... Using this setup, we conducted a 5-fold cross-validation, obtaining a Mean Absolute Error (MAE) of 0.013. Given that the Tanimoto coefficient ranges from 0 to 1, this error represents only 1.3%, indicating a high degree of accuracy in our model’s predictions."

    The edge feature matrix passed to NEAConv contains the Tanimoto coefficient (Section III-C), and Section II-C states that the query Q is generated from the concatenation of the target node features and, when available, the edge features E'. The regression target Y in L_supervised (Eq. 5) is the same Tanimoto coefficient. For every known-known pair, the network input therefore contains the exact scalar it is trained to output; the reported MAE is consistent with a near-identity readout of that input edge attribute, not with a structure-free prediction.

full rationale

The PPI and GO comparisons are against external baselines and external benchmarks (Yang et al., CAFA, deepGOplus, NMFGO), so those results are not circular. Self-citations to exp2GO [22] and to the authors' earlier glycolysis papers [27], [28] provide protocol and dataset provenance but do not by themselves force the conclusions. The central circularity is in the compound-similarity experiment: the edge attributes supplied to the NEAConv attention mechanism are Tanimoto coefficients, and the supervised target Y is the same Tanimoto coefficient. For every known-known pair, the network is handed the answer as an input feature and trained to output it, so the MAE=0.013 largely measures how well the model copies an input, not how well it predicts from structure-free inputs. For unknown-structure pairs, the Tanimoto label is not computable (Section IV-C), so the reported aggregate MAE cannot validate the 'previously unsolved' claim. This is a partial, task-specific circularity; the architecture's independent merits rest on the PPI and GO results.

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

The model depends on standard GNN assumptions plus two paper-specific choices: the first-degree subgraph definition and the cosine-similarity loss alignment. The most fragile input is the assumption that Tanimoto ground truth exists for pairs involving unknown structures. Loss weights alpha=beta=gamma=1 and the KNNG neighborhood size k are chosen by hand or tuned on validation.

free parameters (3)
  • Loss weights alpha, beta, gamma = 1, 1, 1
    Set by hand in Section II-E after preliminary experiments; the authors note they could be tuned per dataset.
  • KNNG neighborhood size k = tuned on validation split
    Used for PPI and GO graph construction (Sections III-A and III-B); affects graph topology and results.
  • Tokenizer and prediction MLP dimensions = not reported
    Hidden sizes and layer counts are not specified, yet are required to reproduce the reported scores.
assumptions (4)
  • domain assumption The first-degree neighborhood subgraph G'_{i,j} contains sufficient information to predict edge properties.
    Section II-B constructs subgraphs only from the central nodes and their direct neighbors; the model never sees higher-order structure.
  • ad hoc to paper Cosine similarity between node embeddings is a meaningful proxy for the target edge property.
    Section II-E defines Lcosine and Eq. (6); the loss assumes a linear relationship between embedding cosine similarity and ground truth.
  • domain assumption Tanimoto similarity is computable for every compound pair, including pairs with unknown structure.
    Section III-C uses Tanimoto coefficients as edge weights and targets, but Section IV-C says unknown compounds cannot have fingerprints.
  • domain assumption A k-nearest-neighbor graph built from sequence similarities is a valid graph representation for PPI and GO prediction.
    Section III-A and III-B construct KNNGs from ClustalO similarities and expression profiles; the model's predictions depend on this graph construction.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Hybrid Supervised and Self-Supervised Graph Neural Network for Edge-Centric Applications." pith.science (2026). https://pith.science/paper/AS65WWKB

@misc{pith2026250112309,
  author       = {Pith},
  title        = {Pith review of: A Hybrid Supervised and Self-Supervised Graph Neural Network for Edge-Centric Applications},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AS65WWKB}},
  note         = {Machine review of arXiv:2501.12309}
}
read the original abstract

This paper presents a novel graph-based deep learning model for tasks involving relations between two nodes (edge-centric tasks), where the focus lies on predicting relationships and interactions between pairs of nodes rather than node properties themselves. This model combines supervised and self-supervised learning, taking into account for the loss function the embeddings learned and patterns with and without ground truth. Additionally it incorporates an attention mechanism that leverages both node and edge features. The architecture, trained end-to-end, comprises two primary components: embedding generation and prediction. First, a graph neural network (GNN) transform raw node features into dense, low-dimensional embeddings, incorporating edge attributes. Then, a feedforward neural model processes the node embeddings to produce the final output. Experiments demonstrate that our model matches or exceeds existing methods for protein-protein interactions prediction and Gene Ontology (GO) terms prediction. The model also performs effectively with one-hot encoding for node features, providing a solution for the previously unsolved problem of predicting similarity between compounds with unknown structures.

Figures

Figures reproduced from arXiv: 2501.12309 by the authors.

Figure 3
Figure 3. , shows the message-passing process involved in the NEAConv layer. The information of each target node i, j ∈ G ′ i,j is updated through the attention mechanism described in [PITH_FULL_IMAGE:figures/full_fig_p002_3.png] view at source ↗
Figure 2
Figure 2. Model architecture overview. The input consists of the pattern subgraph [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Diagram illustrating the attention mechanism in the [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: Comparison of our model with SVGAE, Siamese Residual RCNN, and EnAmDNN on five PPI organisms. Each boxplot [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Critical-difference (CD) diagram of average ranks for [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: Bar plot comparing the maximum F1 scores of different models across each subontology (Cellular Component, Molecular [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 7
Figure 7. Figure 7: The critical difference (CD) diagram presents the [PITH_FULL_IMAGE:figures/full_fig_p006_7.png]
Figure 8
Figure 8. Figure 8: PCA plot of embeddings for the analyzed compounds. Colors represent similarity to a reference compound (glucose [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]
Figure 9
Figure 9. Figure 9: Partial structures of the compounds with KEGG ID [PITH_FULL_IMAGE:figures/full_fig_p007_9.png]
Figure 10
Figure 10. Figure 10: Mean Absolute Error (MAE) for the three models—baseline MPNN, attention experiment and loss experi [PITH_FULL_IMAGE:figures/full_fig_p008_10.png]
Figure 11
Figure 11. Figure 11: Predicted compound similarity (y-axis) versus cosine [PITH_FULL_IMAGE:figures/full_fig_p008_11.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

39 extracted references · 35 canonical work pages

  1. [1]

    Social Media Based Recommender System for E- Com- merce Platforms,

    S. P. Tiwari, “Social Media Based Recommender System for E- Com- merce Platforms,” International Journal of Research in Engineering and Science (IJRES), pp. 87–98, 2021

    work page 2021

  2. [2]

    Deep Learning for Recommender Systems: A Netflix Case Study,

    H. Steck, L. Baltrunas, E. Elahi, D. Liang, Y . Raimond, and J. Basilico, “Deep Learning for Recommender Systems: A Netflix Case Study,” AI Magazine, vol. 42, no. 3, pp. 7–18, Nov. 2021, number: 3

    work page 2021

  3. [3]

    Deep Neural Networks for YouTube Recommendations,

    P. Covington, J. Adams, and E. Sargin, “Deep Neural Networks for YouTube Recommendations,” in Proceedings of the 10th ACM Confer- ence on Recommender Systems . Boston Massachusetts USA: ACM, Sep. 2016, pp. 191–198

    work page 2016

  4. [4]

    A comprehensive survey on graph neural networks,

    Z. Wu, S. Pan, F. Chen, G. Long, C. Zhang, and P. S. Yu, “A comprehensive survey on graph neural networks,” IEEE transactions on neural networks and learning systems , vol. 32, no. 1, pp. 4–24, 2020

    2020

  5. [5]

    DeepWalk: Online Learning of Social Representations,

    B. Perozzi, R. Al-Rfou, and S. Skiena, “DeepWalk: Online Learning of Social Representations,” in Proceedings of the 20th ACM SIGKDD international conference on Knowledge discovery and data mining, Aug. 2014, pp. 701–710, arXiv:1403.6652 [cs]

    arXiv 2014

  6. [6]

    Line: Large-scale information network embedding,

    J. Tang, M. Qu, M. Wang, M. Zhang, J. Yan, and Q. Mei, “Line: Large-scale information network embedding,” in Proceedings of the 24th International Conference on World Wide Web , ser. WWW ’15, vol. 14. International World Wide Web Conferences Steering Committee, May 2015, p. 1067–1077

    work page 2015

  7. [7]

    node2vec: Scalable Feature Learning for Networks,

    A. Grover and J. Leskovec, “node2vec: Scalable Feature Learning for Networks,” in Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining . San Francisco California USA: ACM, Aug. 2016, pp. 855–864

    work page 2016

  8. [8]

    Representation learning on graphs: Methods and applications,

    W. L. Hamilton, R. Ying, and J. Leskovec, “Representation learning on graphs: Methods and applications,” 9 2017

    work page 2017

Show all 39 references
  1. [9]

    Graph neural networks: A review of methods and applications,

    J. Zhou, G. Cui, S. Hu, Z. Zhang, C. Yang, Z. Liu, L. Wang, C. Li, and M. Sun, “Graph neural networks: A review of methods and applications,” AI Open, vol. 1, pp. 57–81, 2020

    2020

  2. [10]

    Spectral networks and locally connected networks on graphs,

    J. Bruna, W. Zaremba, A. Szlam, and Y . LeCun, “Spectral networks and locally connected networks on graphs,” arXiv preprint arXiv:1312.6203, 2013

    2013 arXiv

  3. [11]

    Convolutional neural networks on graphs with fast localized spectral filtering,

    M. Defferrard, X. Bresson, and P. Vandergheynst, “Convolutional neural networks on graphs with fast localized spectral filtering,” in Advances in neural information processing systems , 2016, pp. 3844–3852

    2016

  4. [12]

    The emerging field of signal processing on graphs: Ex- tending high-dimensional data analysis to networks and other irregular domains,

    D. I. Shuman, S. K. Narang, P. Frossard, A. Ortega, and P. Van- dergheynst, “The emerging field of signal processing on graphs: Ex- tending high-dimensional data analysis to networks and other irregular domains,” IEEE Signal Processing Magazine , vol. 30, pp. 83–98, 2013

    2013

  5. [13]

    Approximating signals supported on graphs,

    X. Zhu and M. Rabbat, “Approximating signals supported on graphs,” ICASSP , IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings, pp. 3921–3924, 2012

    2012

  6. [14]

    Geometric deep learning: going beyond euclidean data,

    M. M. Bronstein, J. Bruna, Y . LeCun, A. Szlam, and P. Vandergheynst, “Geometric deep learning: going beyond euclidean data,” 11 2016

    2016

  7. [15]

    A comprehensive survey on graph neural networks,

    Z. Wu, S. Pan, F. Chen, G. Long, C. Zhang, and P. S. Yu, “A comprehensive survey on graph neural networks,” 2019

    2019

  8. [16]

    Message passing neural networks,

    J. Gilmer, S. S. Schoenholz, P. F. Riley, O. Vinyals, and G. E. Dahl, “Message passing neural networks,” Lecture Notes in Physics , vol. 968, pp. 199–214, 2020

    2020

  9. [17]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Łukasz Kaiser, and I. Polosukhin, “Attention is all you need,” Advances in Neural Information Processing Systems , vol. 2017-December, pp. 5999–6009, 6 2017

    2017

  10. [18]

    Graph-based prediction of protein-protein interactions with attributed signed graph embedding,

    F. Yang, K. Fan, D. Song, and H. Lin, “Graph-based prediction of protein-protein interactions with attributed signed graph embedding,” BMC Bioinformatics, vol. 21, pp. 1–16, 7 2020

    2020

  11. [19]

    Fast, scalable generation of high-quality protein multiple sequence alignments using clustal omega,

    F. Sievers, A. Wilm, D. Dineen, T. J. Gibson, K. Karplus, W. Li, R. Lopez, H. McWilliam, M. Remmert, J. S ¨oding, J. D. Thompson, and D. G. Higgins, “Fast, scalable generation of high-quality protein multiple sequence alignments using clustal omega,” Molecular Systems Biology,...

    2011

  12. [20]

    Practical con- struction of k-nearest neighbor graphs in metric spaces,

    R. Paredes, E. Ch ´avez, K. Figueroa, and G. Navarro, “Practical con- struction of k-nearest neighbor graphs in metric spaces,” Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) , vol. 4007 LNCS...

    2006

  13. [21]

    Predicting protein-protein interactions based only on sequences infor- mation,

    J. Shen, J. Zhang, X. Luo, W. Zhu, K. Yu, K. Chen, Y . Li, and H. Jiang, “Predicting protein-protein interactions based only on sequences infor- mation,” Proceedings of the National Academy of Sciences of the United States of America , vol. 104, pp. 4337–4341, 3 2007

    2007

  14. [22]

    exp2go: Improving prediction of functions in the gene ontology with expression data,

    L. D. Persia, T. Lopez, A. Arce, D. H. Milone, and G. Stegmayer, “exp2go: Improving prediction of functions in the gene ontology with expression data,” IEEE/ACM transactions on computational biology and bioinformatics, vol. 20, pp. 999–1008, 3 2023

    2023

  15. [23]

    Cluster analysis and display of genome-wide expression patterns,

    M. B. Eisen, P. T. Spellman, P. O. Brown, and D. Botstein, “Cluster analysis and display of genome-wide expression patterns,” Proceedings of the National Academy of Sciences of the United States of America , vol. 95, pp. 14 863–14 868, 12 1998

    1998

  16. [24]

    Interaction with diurnal and circadian regulation results in dynamic metabolic and transcriptional changes during cold acclimation in arabidopsis,

    C. Espinoza, T. Degenkolbe, C. Caldana, E. Zuther, A. Leisse, L. Willmitzer, D. K. Hincha, and M. A. Hannah, “Interaction with diurnal and circadian regulation results in dynamic metabolic and transcriptional changes during cold acclimation in arabidopsis,” PloS one, vol. 5, 2010

    2010

  17. [25]

    dictybase: a new dictyostelium discoideum genome database,

    L. Kreppel, P. Fey, P. Gaudet, E. Just, W. A. Kibbe, R. L. Chisholm, and A. R. Kimmel, “dictybase: a new dictyostelium discoideum genome database,” Nucleic acids research, vol. 32, 1 2004

    2004

  18. [26]

    The cafa challenge reports improved protein function prediction and new functional annotations for hundreds of genes through experimental screens,

    N. Zhou, Y . Jiang, T. R. Bergquist, A. J. Lee, B. Z. Kacsoh, A. W. Crocker, K. A. Lewis, G. Georghiou, H. N. Nguyen, M. N. Hamid, L. Davis, T. Dogan, V . Atalay, A. S. Rifaioglu, A. Dalklran, R. C. Atalay, C. Zhang, R. L. Hurto, P. L. Freddolino, Y . Zhang, P. Bhat, F. Supek,...

    2019

  19. [27]

    Neural model-based similarity prediction for compounds with unknown structures,

    E. Borzone, L. E. Di Persia, and M. Gerard, “Neural model-based similarity prediction for compounds with unknown structures,” in Ap- plied Informatics , H. Florez and H. Gomez, Eds. Cham: Springer International Publishing, 2022, pp. 75–87

    2022

  20. [28]

    Evaluaci ´on de un modelo neuronal para la estimaci ´on de similaridad entre compuestos a partir de representaciones one-hot,

    E. Borzone, L. Ezequiel, D. Persia, and M. Gerard, “Evaluaci ´on de un modelo neuronal para la estimaci ´on de similaridad entre compuestos a partir de representaciones one-hot,” 2022

    2022

  21. [29]

    Reopti- mization of mdl keys for use in drug discovery,

    J. L. Durant, B. A. Leland, D. R. Henry, and J. G. Nourse, “Reopti- mization of mdl keys for use in drug discovery,” Journal of chemical information and computer sciences , vol. 42, pp. 1273–1280, 11 2002

    2002

  22. [30]

    Why is tanimoto index an appropriate choice for fingerprint-based similarity calculations?

    D. Bajusz, A. R ´acz, and K. H ´eberger, “Why is tanimoto index an appropriate choice for fingerprint-based similarity calculations?” Journal of Cheminformatics, vol. 7, pp. 1–13, 12 2015

    2015

  23. [31]

    Multifaceted protein–protein interaction prediction based on siamese residual rcnn,

    M. Chen, C. J. Ju, G. Zhou, X. Chen, T. Zhang, K. W. Chang, C. Zaniolo, and W. Wang, “Multifaceted protein–protein interaction prediction based on siamese residual rcnn,” Bioinformatics, vol. 35, pp. i305–i314, 7 2019. [Online]. Available: https://dx.doi.org/10.1093/ bioinform...

    2019

  24. [32]

    Protein interaction network reconstruction through ensemble deep learning with attention mechanism,

    F. Li, F. Zhu, X. Ling, and Q. Liu, “Protein interaction network reconstruction through ensemble deep learning with attention mechanism,” Frontiers in Bioengineering and Biotechnology , vol. 8, p. 523015, 5 2020. [Online]. Available: www.frontiersin.org

    2020

  25. [33]

    Benchmark evaluation of protein–protein interaction prediction algorithms,

    B. Dunham and M. K. Ganapathiraju, “Benchmark evaluation of protein–protein interaction prediction algorithms,” Molecules, vol. 27, 1

  26. [34]

    A large-scale evaluation of computational protein function prediction,

    P. Radivojac, W. T. Clark, T. R. Oron, A. M. Schnoes, and T. Wittkop, “A large-scale evaluation of computational protein function prediction,” Nature Methods 2013 10:3 , vol. 10, pp. 221–227, 1 2013

    2013

  27. [35]

    Basic local alignment search tool,

    S. F. Altschul, W. Gish, W. Miller, E. W. Myers, and D. J. Lipman, “Basic local alignment search tool,” Journal of molecular biology , vol. 215, pp. 403–410, 1990. 10

    1990

  28. [36]

    Nmfgo: Gene function prediction via nonnegative matrix factorization with gene ontology,

    G. Yu, K. Wang, G. Fu, M. Guo, and J. Wang, “Nmfgo: Gene function prediction via nonnegative matrix factorization with gene ontology,” IEEE/ACM transactions on computational biology and bioinformatics , vol. 17, pp. 238–249, 1 2020

    2020

  29. [37]

    Deepgoplus: improved protein function prediction from sequence,

    M. Kulmanov and R. Hoehndorf, “Deepgoplus: improved protein function prediction from sequence,” Bioinformatics (Oxford, England) , vol. 36, pp. 422–429, 1 2020

    2020

  30. [38]

    Statistical comparisons of classifiers over multiple data sets,

    Dem ˇsarJanez, “Statistical comparisons of classifiers over multiple data sets,” The Journal of Machine Learning Research , vol. 7, pp. 1–30, 12 2006

    2006

  31. [2022]

    Available: https://pubmed.ncbi.nlm.nih.gov/35011283/

    [Online]. Available: https://pubmed.ncbi.nlm.nih.gov/35011283/

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.