All reviews of published articles are made public. This includes manuscript files, peer review comments, author rebuttals and revised materials. Note: This was optional for articles submitted before 13 February 2023.
Peer reviewers are encouraged (but not required) to provide their names to the authors when submitting their peer review. If they agree to provide their name, then their personal profile page will reflect a public acknowledgment that they performed a review (even if the article is rejected). If the article is accepted, then reviewers who provided their name will be associated with the article itself.
The reviewers are satisfied with the revisions and I concur; I recommend accepting this manuscript.
[# PeerJ Staff Note - this decision was reviewed and approved by Claudio Ardagna, a PeerJ Section Editor covering this Section #]
The paper employs clear and precise professional English throughout, while adequately referencing literature and providing ample background and context in the field. Good structure and detailed proofs.
no comment
The revised version answered my questions and enhanced the validity of the findings.
All previous comments addressed as shown in the rebuttal report.
All previous comments addressed as shown in the rebuttal report.
All previous comments addressed as shown in the rebuttal report.
All previous comments have been thoroughly and correctly addressed. No further concerns. Recommend for accept.
Reviewers have some minor concerns that need to be addressed. The authors should provide point-to-point responses to address all the concerns and provide a revised manuscript with the revised parts being marked in different color.
**PeerJ Staff Note:** Please ensure that all review, editorial, and staff comments are addressed in a response letter and that any edits or clarifications mentioned in the letter are also inserted into the revised manuscript where appropriate.
**Language Note:** PeerJ staff have identified that the English language needs to be improved. When you prepare your next revision, please either (i) have a colleague who is proficient in English and familiar with the subject matter review your manuscript, or (ii) contact a professional editing service to review your manuscript. PeerJ can provide language editing services - you can contact us at [email protected] for pricing (be sure to provide your manuscript number and title). – PeerJ Staff
no comment
no comment
no comment
The paper presents multi-view non-linear dimensionality reduction approaches, building upon existing single-view methods. It thoroughly covers the theoretical background and offers a clear explanation of the methodologies. The proposed methods are tested on both synthetic and real datasets. Additionally, the paper introduces variations like a weight-updating scheme and a pre-training approach, enhancing applicability to real-world data. The paper is well-crafted and articulated in professional English.
However, I have a few comments for further improvement.
Comment 1. Code Documentation and User Accessibility: The GitHub repository linked to the paper needs improvement in terms of documentation and user-friendliness. Please include code that is directly linked to the cases used in the study, along with comprehensive tutorials to facilitate reproducibility and ease of use. This would aid users in better understanding and applying the proposed methods to their datasets.
Comment 2. The paper could benefit from comparison with alternative models in multi-view representation learning like deep generative methods. This comparison should explore both the advantages and disadvantages of the proposed methods relative to these models, providing insights into their applicability in various scenarios.
Comment 3. The paper should include information on the time and computational costs associated with the proposed methods. Understanding the scalability, especially in relation to the number of samples, is crucial for practical applications.
This paper presents extensions of manifold learning techniques (SNE, LLE, and ISOMAP) for dimensionality reduction and visualization in multi-view data contexts. These multi-view approaches aim to yield more comprehensible sample projections than when visualizing each data-view individually. The authors compare their multi-view methods with existing approaches using both real and synthetic datasets. Among these, multi-SNE demonstrates superior performance.
Strengths:
Overall, the language appears professional and clear, adhering to academic standards.
The article contributes significantly to data visualization by introducing innovative adaptations of well-established manifold learning techniques (multi-SNE, multi-LLE, and multi-ISOMAP) for multi-view data. The methods are elaborated upon with clarity and precision. Furthermore, the real-world applicability of these methods, especially in complex datasets like multi-omics single-cell data, is commendably highlighted, underscoring their practical relevance and utility.
1. Expanding the testing of these methods across diverse datasets would enhance the paper by demonstrating their effectiveness and robustness in varied scenarios. Have the authors explored the application of these methods on additional datasets?
2. Investigating the sensitivity of these methods to hyperparameters and providing detailed guidelines for their selection would greatly benefit the paper. This would ensure the methods' broader applicability and ease of use. Could the authors elaborate on the hyperparameter selection process for each proposed method?
1. The potential increase in computational complexity due to the multi-view approaches warrants a detailed discussion of computational demands and potential optimization strategies. How do the proposed multi-view methods (multi-SNE, multi-LLE, and multi-ISOMAP) compare with existing single-view methods in terms of accuracy and computational efficiency?
2. A more comprehensive discussion on the limitations of the proposed methods and the identification of potential avenues for future research would balance the paper and guide forthcoming studies. How could the proposed methods be adapted or extended to accommodate larger datasets or real-time data processing?
In the manuscript, the authors proposed extensions of t-distributed SNE (t-SNE), LLE and ISOMAP, for dimensionality reduction and visualisation of multi-view data, which demonstrated superiority against existing canonical methods and other variations in both qualitative and quantitative fashion. I want to commend the authors for the following points:
1). Very thorough and detailed introduction of existing t-SNE, LLE and ISOMAP methods: the authors provided a very detailed introductions in the Section 1 – Materials and Methods to the existing t-SNE, LLE and ISOMAP methods as baseline and start-point for their extension work. Such thorough introduction also laid foundations for future discussion, for example, fine parameter tuning.
2). Very comprehensive data and results sections: in both sections, the authors demonstrated how various synthetic and real-world data-sets were utilized to test the performance of introduced extensions against benchmarks (canonical methods, other variations to handle multi-view data). In the end, the authors tested their methods on the real-world multi-omics single cell data to further showcase the applicability of their proposed methods.
3). Very organized results discussion: in the Section 3 -Results, the authors lay-out the results in a very organized fashion starting with 4 foundational questions and subsequent contexts are centered on such 4 questions. What’s more interesting is that authors also offered an ablation study to investigate if all data-views are necessary/beneficial to separate clusters, which leads to the discussion of noise level in the multi-view data. The manuscript could benefit more if authors can directly link the subsequent section to the foundational 4 questions in the Section 3. For example, in question 1 (line 369), directly put section number 3.1 after it.
In general, the manuscript is clearly written in professional, unambiguous language.There are still some minor issues on the basic reporting side worth author’s attention:
1). Line 22, detailed metrics needed to support the claim of “to have the best performance”;
2). Section 1.1 Notation, dimension p needs to be added;
3). Abbreviation MV-tSNE1 needs to be given at earlier position, perhaps at line 163;
4). Line 168-169, the subsequent section indexes need to be given to support the claim of “multi-SNE producing the best separation…”;
5). Line 280, “due to the nature of their data” is confusing, re-edit the language here;
6). Line 583, “The produced embedding…” is confusing, does it mean the produced embeddings from weigh adjustment method?
7). Line 637 and 645, wrong figure references, should be Figure 12 instead of Figure 7 in both locations.
The manuscript provided a well-defined extension of existing t-distributed SNE (t-SNE), LLE and ISOMAP, for dimensionality reduction and visualisation of multi-view data. Authors offered detailed mathematical explanation of such extensions from baseline methods and provided a comprehensive comparison of such extensions against canonical methods and other variations on various synthetic and real-world data. Such comparisons demonstrated the superiority and applicability of extensions proposed using both qualitative (figures) and quantitative (4 metrics, ACC, NMI, RI and ARI) fashions. The manuscripts could benefit more if authors could:
1). The proposed multi-SNE method is computed using iterative approach, and authors briefly mentioned stopping criteria from Line 179 to line 182, using either hard stopping rule with 1000 iterations or so-called “no significant changes”. Additionally, the proposed multi-SNE method is also demonstrated to have the best performance. Combining the above 2 points, it would be greatly beneficial if authors could add more details discussing the convergence nature of such multi-SNE method: if convergence is guaranteed or not, if not, why 1000 iterations is a good-stopping criteria here.
2). In the parameter sections, Line 462, also line 261, the parameter S here is confusing. As authors already stated, SNE parameters depend on the Perlexity while LLE and ISOMAP depend on the nearest neighbours, does S here just mean the number of nearest neighbour threshold (as Perlexity also defined on the local neighbours)? If so, authors should directly to call S as nearest neighbours threshold parameter to avoid confusions.
Overall, the manuscript’s research question is well defined and authors’ research did fill an identified knowledge gap.
In the data and results sections, authors already did very well on providing comprehensive and robust details on:
1). How synthetic data were generated;
2). How proposed methods performed against other benchmark methods on both synthetic and real-world datasets;
3). Demonstrating the proposed method performance using both quantitative and qualitative fashion. (I want to commend authors’ extensive and well-organized figure and table demonstrations here.)
There are several minor issues still worth author’s attention:
1). More details needed on how clusters are defined in the synthetic data (Section 2.1). Authors did very well on how each view were generated in the synthetic data, but relatively weak on the how clusters were defined/generated, is it related to the distinct polynomial functions (line 307)? If so, this should be called-out explicitly.
2). Figure 4 and 5, legends are missing to indicate the color meaning. For example, which color stands for which digit, as shown already in the Figure 11. Additionally for the Figure 11, it would be beneficial if authors can add True clustering plot (Figure 4 right) again for reader’s convenience.
3). Line 485, more elaborations needed for the statement of “Overall, ISOMAP-based multi-view algorithms showed higher variability than the other multi-view methods”. What does variability mean here? Does it mean ISOMAP method is more sensitive to the number of nearest neighbours? (but LLE seems more sensitive)? Or does it mean its performance is not stable? Language re-edits are needed here.
All text and materials provided via this peer-review history page are made available under a Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.