Modularization Method to Reuse Medical Knowledge Graphs

Round 1

Reviewer 1 Report

The aim of this paper is the description of the generation and reuse of an ontology in order to manage EHRs. After a quick state of the art, the authors  describe the generation of a Medicament knowledge graph and a Disease knowledge graph with evaluation mainly founded on "modules".

It is an interesting study but the authors did not clearly showed the general contribution nor its theoretical soundness. Moreover they did not argued deeply on the choices made at each stage of the process. It is more an engineering paper than an article.

Major :

- use either "ontologies" or "knowledge graphs" with adequate definition

- Lack of technical contribution. Neither novel design or architecture, nor original methods or components of the system are proposed in this paper, which limits the overall contribution. The authors should add related work and more background. Moreover, the contributions should be clarified. From my point of view, the overall system described in this paper is a pure engineering work that aggregates standard methods in every step, without any novel design and technical innovation.

- No available resources are provided : contributions in the domain of ontologies generally implies references to resources generated. Besides, lacking open-sourced resources further diminishes the significance and reusability of this paper. As the authors could imagine, without any reusable resources or concrete technical contributions, there are limited takeaways for the reader to learn from the paper.

- There are no commonly-used concepts, and haven't been formally defined in this paper either. It seems like the authors randomly create the categories and their names without any explanation.

- No quality and correctness verification of the graphs. Since the medical data could be incorrect with poor quality, there should be some automatic quality verification or correction methods in the graph construction process. If graphs rely on standard ontologies and do not need to be audited please proceed and provide any evidence to show the quality and correctness of the constructed graphs. Moreover, even for standard ontologies, knowledge fusion/alignment is a complex and difficult task to conduct. Metrics ar not enough.

- the state of the art should be updated to include also recent publications on the domain (instead of Shimizu et al. published in 2021 all the others are quite out-to-date), only 11 papers is not enough

- the developed modules seem not to be used conjointly in a general framework but independently

- everything is described through protege. How does work the system ? Inputs ? Outputs ? How does it rely on the generated graphs ? ... all these are not clear in the paper.

Minor :

- desambiguate some acronyms (OWL, KGM, SPARQL, ...)

- define T-Box and A-Box, since all the readers are not familiar with those definitions

- explain the aim of the SPARQL queries, difference with other standard queries ?

- explain the general structure of the EHRs and how they are related to the new graphs

Author Response

Regarding the technical contribution, we have highlighted the principal characteristics of reported method in lines form 220 to 232.

Use either "ontologies" or "knowledge graphs" with adequate definition. Response: definitions have been added in lines 30 and 50.

Lack of technical contribution. Neither novel design or architecture, nor original methods or components of the system are proposed in this paper, which limits the overall contribution. Response: Sections as “Introduction”, “Proposed method” and “Conclusions” have been modified to highlight the technical contributions and the originality of the proposed method, since we describe the differences and advantages of our method.

The authors should add related work and more background. Response: Several related works were reviewed and added, such as references [1-5].

The contributions should be clarified. From my point of view, the overall system described in this paper is a pure engineering work that aggregates standard methods in every step, without any novel design and technical innovation. Response: The main contributions of our method are described, since it contributes to the generation of new and innovative approaches in the knowledge representation field.

No available resources are provided : contributions in the domain of ontologies generally implies references to resources generated. Response: Public access to knowledge graph modules are available at http://aisii.azc.uam.mx:8082/HealthKnowledgeGraphs/index.html

No quality and correctness verification of the graphs. Since the medical data could be incorrect with poor quality, there should be some automatic quality verification or correction methods in the graph construction process. Response: all knowledge graphs modules have been evaluated with reasoners, resulting as consistent.

The state of the art should be updated to include also recent publications on the domain (instead of Shimizu et al. published in 2021 all the others are quite out-to-date), only 11 papers is not enough. The related works have been augmented in Section 2.

The developed modules seem not to be used conjointly in a general framework but independently. Modules have been imported into Patient Health Record Knowledge Graph which is downloadable form http://aisii.azc.uam.mx:8082/HealthKnowledgeGraphs/index.html

Disambiguate some acronyms (OWL, KGM, SPARQL, ...) Response: All concepts have been disambiguated on first use.

Define T-Box and A-Box, since all the readers are not familiar with those definitions. These definitions were clarified in lines 200 t0 204.

Explain the aim of the SPARQL queries, difference with other standard queries. SPARQL is explained in lines 273 to 276.

Explain the general structure of the EHRs and how they are related to the new graphs. EHR is downloadable from http://aisii.azc.uam.mx:8082/HealthKnowledgeGraphs/Integrated/PatientElectronicRecord.owl

 

Reviewer 2 Report

The references are limited and out of date, please add more.

Author Response

References have been added.

Reviewer 3 Report

Research Design can be more clearly described perhaps

Author Response

The main design considerations of our method are described, since it contributes to the generation of new and innovative approaches in the knowledge representation field.

Reviewer 4 Report

The manuscript is written well and is correct as per scientific contribution. Although overall the manuscript has many positive aspects but yet I feel there are some minor issues that should be resolved before it's acceptance: I recommend the following minor revisions:

1. Moderate language corrections are needed.

2. Research methodology is not so convincing, it needs improvement.

3. Given tables and figures should not be splitted on two pages.

4. Write all references on same pattern.

Author Response

Comments have been addressed.

Round 2

Reviewer 1 Report

I have no other comments at this stage, all the requested adds-on being included in the final version.