Applications of Deep Learning for Drug Discovery Systems with BigData

Round 1

Reviewer 1 Report

The manuscript did not provide even a single example on the application of deep learning that has led to drug discovery. Mentioning examples of model-based ongoing (successful/failed) drug discovery efforts will provide solid context to the topic.

It is highly recommended to summarise deep-learning-based major drug discovery programs to date, incorporate corresponding biological or clinical databases, etc.

Author Response

Comments and Suggestions for Authors
The manuscript did not provide even a single example on the application of deep learning that has led to drug discovery. Mentioning examples of model-based ongoing (successful/failed) drug discovery efforts will provide solid context to the topic.

According to the reviewer’s comments, we added sentence in page 10 and references.

It is highly recommended to summarise deep-learning-based major drug discovery programs to date, incorporate corresponding biological orclinical databases, etc.

According to the reviewer’s comments, we added Table 1.

Reviewer 2 Report

Paper deals with important topics biomedical.
The authors have researched a prototype of a Drug Discovery System based on the BigData approach. 

However, I have a number of suggestions:    
1. Abstract should be extended by the obtained results in part of the performance evaluation of the proposed approach. Moreover, add limitations of the proposed method as well.  
2. In the intro section I would suggest adding a more detailed description of the existing solution and point the core differences and advantages of your proposed solution in comparison with the existing one. Moreover, please fix images, and place them vertically, it's hard to read them in such a format, for example, in Figure 2.
3. Authors should clearly point-by-point describe the main contributions of this paper. It should somehow resonate with the title of the work. 
4. Conclusion section should be extended by the obtained authors' results and limitations, with reinform with the vision of future research and utilization. 
5. A lot of references are outdated and unlinked, please revise all of them. Please fix it by using 3-5 years old papers in high-impact journals.

Author Response

Paper deals with important topics biomedical.
The authors have researched a prototype of a Drug Discovery System based on the BigData approach. 

However, I have a number of suggestions:    
1. Abstract should be extended by the obtained results in part of the performance evaluation of the proposed approach. Moreover, add limitations of the proposed method as well. 

According to the reviewer’s comments, we added some sentences about performance evaluations and limitations in Abstract section indicated by yellows on 23-25, in page 1.

2. 
   In the intro section I would suggest adding a more detailed description of the existing solution and point the core differences and advantages of your proposed solution in comparison with the existing one. 

According to the reviewer’s comments, we added some descriptions about differences, comparison, and advantages of your proposed solution in introduction sections indicated by yellows on 172-176, in page 5.

Moreover, please fix images, and place them vertically, it's hard to read them in such a format, for example, in Figure 2.

According to the reviewer’s comments, we corrected Fig2.

3. 
   Authors should clearly point-by-point describe the main contributions of this paper. It should somehow resonate with the title of the work. 
   4. Conclusion section should be extended by the obtained authors' results and limitations, with reinform with the vision of future research and utilization. 

According to the reviewer’s comments, we added some descriptions in conclusion sections indicated by yellows in page11.

5. A lot of references are outdated and unlinked, please revise all of them. Please fix it by using 3-5 years old papers in high-impact journals.

Reviewer 3 Report

It’s a review article of deep learning in drug discovery. It summarizes and reviews the application of deep learning into drug discovery with BigData. It also illustrates the various technical problems from the perspective of real-world problem; and introduces some approaches to these technical challenges. However, some parts in this article do not link deep learning to the drug discovery area. I’ll explain by section.

Section 1. Introduction

Introduction demonstrates why and how AI technology is used in the drug discovery system. However, it's better have a roadmap at the end of the introduction section to provide a description of the rest of the paper so that the readers know where you are going and what to expect. For example: “The paper is organized as follows. In Section 2, we discuss…”

Section 2. Types of deep learning network models 

This section demonstrates the main types used in drug discovery. The references are related to bioinformatics/ drug discovery. However, the text is not well illustrated to show how these deep learning network models are involved in the process of drug discovery. For example, the authors only talk about CNN and FAN are used in image processing, and RNN can be used in time series data. But how are they used in drug discovery specifically? 

Section 3, Technological Application in BigData and Deep Learning

The first paragraph in section 3 talks about the deep learning history, which is too long, and not related to drug discovery. The text does not include/cite Figures 1&2. References [131],[133] are not related to drug discovery. “That same year, a group of researchers at Google published a paper on image dis- 216 crimination of cats using neural networks, …”—missing a reference.

Section 4. Deep Learning and Technical problem

This section is well written. Technical problems from the real-world drug discovery problem have been demonstrated. This section links deep learning with drug discovery. 

Section 5. Approach to Technical Challenges for Technical Problems in Deep Learning

Approach to technical problems addressed in Section 4. Same issue here. Too many references are not related to drug discovery.

A table is suggested to show methods with advantages and disadvantages, what kinds of problems can be solved by the methods. A table is more readable to the readers.

Author Response

It’s a review article of deep learning in drug discovery. It summarizes and reviews the application of deep learning into drug discovery with BigData. It also illustrates the various technical problems from the perspective of real-world problem; and introduces some approaches to these technical challenges. However, some parts in this article do not link deep learning to the drug discovery area. I’ll explain by section.

Section 1. Introduction

Introduction demonstrates why and how AI technology is used in the drug discovery system. However, it's better have a roadmap at the end of the introduction section to provide a description of the rest of the paper so that the readers know where you are going and what to expect. For example: “The paper is organized as follows. In Section 2, we discuss…”

According to the reviewer’s comments, we corrected last sentences in Introduction section.

Section 2. Types of deep learning network models 

This section demonstrates the main types used in drug discovery. The references are related to bioinformatics/ drug discovery. However, the text is not well illustrated to show how these deep learning network models are involved in the process of drug discovery. For example, the authors only talk about CNN and FAN are used in image processing, and RNN can be used in time series data. But how are they used in drug discovery specifically? 

According to the reviewer’s comments, we corrected last sentences in Introduction section.

Section 3, Technological Application in BigData and Deep Learning

The first paragraph in section 3 talks about the deep learning history, which is too long, and not related to drug discovery. The text does not include/cite Figures 1&2. References [131],[133] are not related to drug discovery. “That same year, a group of researchers at Google published a paper on image dis- 216 crimination of cats using neural networks, …”—missing a reference.

According to the reviewer’s comments, we corrected Fig 1 and 2.

Section 4. Deep Learning and Technical problem

This section is well written. Technical problems from the real-world drug discovery problem have been demonstrated. This section links deep learning with drug discovery. 

Section 5. Approach to Technical Challenges for Technical Problems in Deep Learning

Approach to technical problems addressed in Section 4. Same issue here. Too many references are not related to drug discovery.

A table is suggested to show methods with advantages and disadvantages, what kinds of problems can be solved by the methods. A table is more readable to the readers.

Round 2

Reviewer 1 Report

Page 6, Types of deep learning network models lack adequate information. The authors must briefly mention the popular 3 to 4 types of DNN models and their performance when applied to the biomedical tasks of choice and the error rates.

 

performance of popular CNNs applied for AI vision tasks gradually increased over the years, surpassing human vision (5% error rate in the chart below).
Read more at: https://viso.ai/deep-learning/deep-neural-network-three-popular-types/ performance of popular CNNs applied for AI vision tasks gradually increased over the years, surpassing human vision (5% error rate in the chart below).
Read more at: https://viso.ai/deep-learning/deep-neural-network-three-popular-types/ 3 Types of Deep Neural Networks
Read more at: https://viso.ai/deep-learning/deep-neural-network-three-popular-types/

Author Response

According to the reviewer’s comments, we added some sentences for some DNN models and their performances, and fig 3 indicated by blue color.

Reviewer 2 Report

All recommendations are taken into account by the authors. Work is recommended for acceptance.

Author Response

Thank you

Reviewer 3 Report

1. still not satisfy with the end of Introduction. A paragraph (instead of one sentence) is recommended to navigate the reader.

2. the new (last) sentence in section 2: These CNN -> These models. Or maybe you can have a sentence for each model that explains how it's related to drug discovery. The references in Section 2 are indeed biomedicine-related, it shouldn't be hard to summarized in text.

3. Section 2 & 3 are still not well organized. You might want to combine both sections and consider to re-organize as follow:

one section for general overview/introduction of DL techniques /history (for example, the first paragraph in your current Section 3, and other lit review for general DL techniques, not for biomedicine-related).--and control the length of this section. Be sure it's not too long.

and the other section illustrates how DL is used in and benefits the biomedical research , as well as how it develops in biomedical research. need a better title for this section to indicate this section is related to drug discovery

4. Section 4, 5 : suggest to have each technical problem in one separate paragraph . -- more readable . so is in Section 5. 

5. change the title of Section 5: Approaches to Technical Problems in Deep Learning

6. not sure about the purpose of the last sentence you added in Section 5 in the revised version. The original purpose of my suggestion is to have a summarized table for all the techniques /references  you mentioned in Section 5. I didn't mean to add more references as a new table. If the new references can be separated /added into the four issues respectively, that'll be better, however, If you feel uncomfortable, you can delete Table 1 as well as the newly added sentence. Table formatting: If you want to add a table, then it should be shown in Page 11. 
Sorry for the confusion.

 

 

 

 

Author Response

1. still not satisfy with the end of Introduction. A paragraph (instead of one sentence) is recommended to navigate the reader.

According to the reviewer’s comments, we created the paragraph in the end of Introduction indicated by blue color.

 

2. the new (last) sentence in section 2: These CNN -> These models. Or maybe you can have a sentence for each model that explains how it's related to drug discovery. The references in Section 2 are indeed biomedicine-related, it shouldn't be hard to summarized in text.

According to the reviewer’s comments, we corrected this sentence that is These CNN -> These models that indicated by blue color.

 

3. Section 2 & 3 are still not well organized. You might want to combine both sections and consider to re-organize as follow:

one section for general overview/introduction of DL techniques /history (for example, the first paragraph in your current Section 3, and other lit review for general DL techniques, not for biomedicine-related).--and control the length of this section. Be sure it's not too long.

and the other section illustrates how DL is used in and benefits the biomedical research , as well as how it develops in biomedical research. need a better title for this section to indicate this section is related to drug discovery

 

According to the reviewer’s comments, we combine Section 2 & 3 and create sub-sections, 2-1 and 2-2.

 

4.Section 4, 5 : suggest to have each technical problem in one separate paragraph . -- more readable . so is in Section 5. 

 

According to the reviewer’s comments, we create sub-sections, 3-1 to 3-4 for each technical problems.

 

5. change the title of Section 5: Approaches to Technical Problems in Deep Learning

According to the reviewer’s comments, we corrected the title of Section 5 as “Approaches to Technical Problems in Deep Learning”.

 

6. not sure about the purpose of the last sentence you added in Section 5 in the revised version. The original purpose of my suggestion is to have a summarized table for all the techniques /references  you mentioned in Section 5. I didn't mean to add more references as a new table. If the new references can be separated /added into the four issues respectively, that'll be better, however, If you feel uncomfortable, you can delete Table 1 as well as the newly added sentence. Table formatting: If you want to add a table, then it should be shown in Page 11. 
   Sorry for the confusion.