如何正确回复审稿意见

步骤：

1.仔细阅读审稿人的每一条意见，并对审稿意见进行分类；
2.结合自己的论文，仔细揣摩审稿人的意图，尝试记录下每一个意见的解决方案；
3.依据审稿意见修改论文；
4.撰写Response。

回复原则：
态度要诚恳，措辞要礼貌：答复审稿人的时候语言的基调一定是尊重的。逐条回复，一一对应：作者应该非常严肃对待审稿人的意见，并认真考虑审稿人的每一项建议。尽量满足审稿人的需求：不管你觉得审稿人提出来的问题有多么的苛刻或者无法理喻！尽量解答他们的疑问和满足他们的要求。如果不能满足，请有理有据的说服审稿人；逻辑清晰，不写废话：要让编辑和审稿人能快速抓住你文章的亮点和修改的意义。

1. Response开头

清楚的展示标题“Respose to the Review Comments”，然后用一段话来感谢审稿人和编辑：
We would like to thank you for your careful reading, helpful comments, and constructive suggestions, which has significantly improved the presentation of our manuscript.

接下来简述自己对文章的改动：
We have carefully considered all comments from the reviewers and revised our manuscript accordingly. The manuscript has also been double-checked, and the typos and grammar errors we found have been corrected. In the following section, we summarize our responses to each comment from the reviewers. We believe that our responses have well addressed all concerns from the reviewers. We hope our revised manuscript can be accepted for publication.

自己现在来看觉得写的简单了些，为了获得审稿人的好感，应该写的充实一些，建议大家对自己的主要修改进行提炼，写于最后：
In our revisions, we paid specific attention to 1)… , 2)…, 3)…, and 4)…

2. 针对审稿人的具体意见进行回复

先分享一些常用的开头：

1. We gratefully appreciate for your valuable suggestion/ comment.
2. Thank you for your rigorous consideration/ comment/ nice suggestion/ advice.
3. We gratefully thanks for the precious time the reviewer spent making constructive remarks.
4. We feel sorry for the inconvenience brought to the reviewer.
5. Thank you so much for your careful check.
6. We totally understand the reviewer's concern.
7. Thank you for pointing out this problem in manuscript.
8. Thank for your comments.
9. Thank you for the above suggestions.
10. It is really true as Reviewer suggested that……
11. We are very sorry for our negligence of ……...
12. We have re-written this part according to the Reviewer’s suggestion
13. We are very sorry for our incorrect writing ……...
14. We have made correction according to the Reviewer’s comments.
15. As Reviewer suggested that……
16. Considering the Reviewer’s suggestion, we have ……

这个时候一般会出现审稿人对于整篇论文的一个评价，只要你的论文有一定的价值，都会收到审稿人的一些肯定的描述：
this paper provides a very comprehensive analysis and review on ****. It includes different levels of reviews on ***, ***, and ***. The logic is very clear and whole article is well written. In addition, to prove more insight about how to integrate the knowledges on current ****, authors run different levels of experiments to make further analysis. Finally, future direction and challenges are demonstrated for other researchers to make improvement.

那么这个时候我们要感谢审稿人对你的肯定：
Response:
We thank the reviewer for reading our paper carefully and giving the above positive comments.

当审稿人觉得你的论文没有太多的价值，他想reject你的论文的时候：
In this manuscript, the authors proposed a survey study, reviewing the applications of ***, in the framework of ****. This is a critical topic and the number of research studies that came out in the last 2 years is immense; however, the number of recent papers included in this survey are quite low, which negatively effects the timeliness/currentness of the study. There are many critical issues related to the current state of the manuscript. Below, I list my specific concerns/issues:

大家也不用慌，只需要感谢他的意见就好：
Response:
Thank for your comments.

好的，到了这一步，我们就要开始直面恐惧，那就是逐条逐点的回复审稿人提出的具体的意见。我们首先要做的，就是搞清楚审稿人的意图，他是存在疑惑；指出问题；给出建议。

面对不同的意图，我们回复也是有区别的。
2.1 存在疑惑：

这种意见一般出现两种状况：

(1) 在你肯定论文表述没有出错的情况下，审稿人认为你的论文中存在疑点，没有看懂，需要你来解释，那么你可以在response中进行解释，最好是引经据典，让审稿人明白你的意图。

(2) 如果是因为论文中的错误或者表述不清而造成的疑惑，那我们需要对论文进行修改，然后依据修改的部分进行回复。
Case1：
The authors may want to explain the reason why they have chosen to mention some of the methods at the beginning of Section 3.2 rather than mentioning them in the subsections 3.2.1 and 3.2.2?
Response:
Thank you for pointing out this problem in manuscript. We have restructured Section 3 in the revised manuscript. Some methods mentioned at the beginning of subsection 3.2 have been moved to another subsection. In the revised manuscript, subsection 3.1 is dedicated to describing methods of using unsupervised learning for descriptors. Consequently, the descriptions of methods [63], [65], [69], [70], [71] have been moved to subsection 3.1. Subsection 3.2 focuses on the extraction of sequence features. The descriptions of methods [74], [75] remain in subsection 3.2.

Case 2：

Why are references 122 and 124 not indicated in Table 6?
Response：
We thank the reviewer for pointing out this issue. The reference [140] (reference [122] in the original manuscript) is a model based on multi-modality, and reference [93] (reference [124] in the original manuscript) uses CNN-RNN structure for descriptors. So, it is inappropriate to introduce them in subsection 4.1. We have moved the description of reference [140] to subsection 4.4 and the description of reference [93] to subsection 4.6 in the revised manuscript. They are not indicated in Table 8 (Table 6 in original manuscript) in the revised manuscript.

2.2 指出问题：

当审稿人指出你的问题时，你一定要冷静，仔细思考自己的不足，然后去揣摩审稿人的意图，然后进行修改和回复。
Case 1：

I am not sure if reference 124 relevant to be placed in Section 4.1.
Response：
We thank the reviewer for pointing out this issue. We have moved the description of reference [93] (reference [124] in the original manuscript) to subsection 4.6 in the revised manuscript.

Case 2：

The problem of fixed length input sequence (for protein sequences or for compound SMILES sequences) for the deep neural network models is not mentioned at all.

审稿人认为我们并没有提及”固定长度的输入“这一问题。因此，我在本文专门拿出一个段落来描述这个问题，首先分析这个问题产生的原因，然后分析这个问题造成的影响，最后查找文献，列举可行方案。

Response：
Thank you for the above suggestion. Following your suggestion, we have added a new subsection called “Fixed length size” in Section 6 to described the problem of fixed length input sequence for the deep neural network models.

Case 3：

审稿人认为我们的论文与另外一篇论文在结构和内容上十分相似，感觉我们的手稿是那一篇论文的补充版本，暗示我们的工作没有意义。

这个意见是一个十分中肯的意见，同时也是十分让人头疼的意见。鉴于这个审稿人是倾向于要拒绝我的论文，因此我们在处理这条意见的时候需要十分的小心。

首先我们要感谢审稿人的意见，然后我们表明我们对论文进行了大改，以此来提高我们论文的意义。
Thank you for your comment. In the revised manuscript, we have restructured our manuscript and focused on feature extraction and CPR prediction based on deep learning in content to highlight the significance of this manuscript. The main modifications are as follows:
然后我们开始列举我们做出的比较大的改动，同时说明改动的意义。
第一点，我们是针对审稿人的第一点，认为我们的手稿与其他论文中的表格过于雷同。（首先要声明的是，该部分是对数据集的一个统计分析，这在许多文献中都有出现，有些许雷同不可避免）我们这个时候就要着重突出我们的不同，比如说我们提供的统计信息是最新的；我们还添加其他别人没有的东西。
（1） We have updated the statistical information in all tables in our manuscript to ensure that it is the latest data. And we have collected some publicly available datasets and described them in Section 1 of the supplementary material. We have also provided statistical information, attributes, and download links of these datasets to guide the readers in choosing the dataset for developing and testing their methods.
第二点，也是针对审稿人的第一点，认为我们的手稿与其他论文中的内容有些许雷同。那么我们也是强调我们与别人不同的地方。
（2） In Section 3, various feature extractions in CPR prediction are comprehensively discussed. In addition to the commonly used neural networks such as CNN, RNN and GCN, this section contains the current relatively novel models, such as Transformer [97] and GAT [99].
第三点，是针对审稿人的第二点，认为我们的参考文献不够新。那么我们就说明，我们去点了很多老旧的东西，加入许多新的东西，以及说明新加进来的文献对我们的综述是有意义的，而不是随便添加的。
（3） In the revised manuscript, we have deleted 43 old references which had already been reported in detail in other reviews. And we have expanded 27 methods published in recent 2 years. The introduction of these methods will improve the comprehensiveness of our manuscript and provide readers with more perspectives. For example, we added DGraphDTA [102], GraphDTI [101], the model proposed by Wang et al. [83], and InterpretableCNN-CPR [120] to the convolution-based method and added GNN-CPI [88], MolTrans [131], MT-DTI [129], TransformerCPI [132] and DeepCDA [92] to the attention-based model. Furthermore, we have added three new subsections, “Multi-modal-based methods”, “Generative deep learning approaches” and “Other methods”, in Section 4. These methods take different perspectives to facilitate the discovery of deep learning in CPR prediction and present novel directions that deserve to be explored.
第四点是为了凸显出我们论文的意义，展现出我们的态度，说明我们的论文与别人的是不同的。
（4） To follow the latest research, we have added more state-of-the-art methods, which have been proposed in the past three years, for comparison in Section 5. In the revised manuscript, we compared 9 methods on CPI prediction task and 10 methods on CPA prediction task.

Case 4: 语法错误

The use of language in the manuscript is generally okay but the way the sentences are structured are problematic from place to place. It feels like they are written in the style of the everyday use of the English language, different from the style of the scientific literature. Since this is a survey study, the text is naturally quite lengthy and the problematic sentences are scattered throughout the manuscript. The authors should go over the entire text carefully to change the style considering the customary style of scientific literature
Response：
Thank for your comments. We have thoroughly checked and corrected the grammatical errors and typos we found in our revised manuscript.
or:
We are very sorry for the mistakes in this manuscript and inconvenience they caused in your reading. The manuscript has been thoroughly revised and edited by a native speaker, so we hope it can meet the journal’s standard. Thanks so much for your useful comments.

2.3 给出建议：
Case 1：

Figure. 2 is a little bit confusing for the readers. It is supposed to demonstrate ***. however, based on figure 2, it is more like to pipeline graph which may need to be modified.

这条意见很明显，是给出建议，审稿人觉得我的图2会给读者造成疑惑，建议我修改。那我自然是听从他的建议，修改了图2，然后回复：

Response：Thank you for pointing out this problem in our manuscript. According to the revised content, we have redrawn Figure 2 to clearly show ***.

还可以在文中添加一些细节，对图进行补充说明，方便审稿人看懂该图的意义。

Case 2：

In terms of results in section 5, it would be good to demonstrate the details of each model, like graph or others. Then readers would know more details about how to compare those results.

审稿人认为，我们应该为第五章节中的方法提供一些补充信息，方便读者来进行比较，至于使用什么方法，我们可以视情况而定。

Response：Thank you for the above suggestion. In Section 2 of supplementary material, we have provided detailed descriptions of the models, which are compared in Section 5. Descriptions of the models for CPI prediction are shown in Figures S.1 to S. 9 in the supplementary materials. And overviews of the models for CPA prediction are shown in Figures S.10 to S. 19 in the supplementary material.

Case 3：

Can the authors guide the readers about the datasets they can use for their methods?

此处虽然是个问句，但是审稿人的意图还是很明显的，就是建议我们添加一些细节来方便读者使用挑选和使用数据集

Response：
Thank you for the above suggestion. In the revised manuscript, we have collected some publicly available datasets and described them in Section 1 of the supplementary material. We have also provided statistical information, attributes and download links of these datasets to guide the readers in choosing the dataset for their methods.
The statistical information and attributes of these datasets are presented in Tables S.1 and S.2. and the download links are presented in the Tables S.3 and S.4 of the supplementary material.

Case 4:

There is something about the structure and organization of the manuscript that could be improved.

(1) Some of the methods are explained in too much detail; for example, Graph Neural Network while the other methods are coarsely given; for example, “Xie et al. [117] developed a DNN model using only the z-scores of 978 marker genes”?

(2)For example, information regarding CNN is given at the very end of Section 3.2.1. This could be moved to a previous subsection.“CNNs are considered as one of the best models for learning image feature. The topology of CNN is divided into multiple learning stages, which are composed of convolutional layer, non-linear activation and sub-sampling layer. Each layer uses a set of convolution kernels (filters) to perform multiple transformation. The main advantage of CNNs is that the feature extraction is performed in a fully automatic and data-driven manner, so there is no need to design feature selection or convolution kernels in advance. At present, there are many excellent image feature extraction networks that can be used by researchers, such as AlexNet [103], GoogleNet [104], DenseNet-201 [105], ResNet152 [106] and VGG-19 [107].”

首先我们分析审稿人的这个意见，属于一个总分的结构。首先，总的来说是建议我们修改论文中一些结构和组织（逻辑）上的问题，然后给出了2个具体的细节。这个时候我们就要核查，审稿人的两个具体的细节提示，是不是真的需要修改（一般来说是需要修改的），然后我们再浏览全文，找到类似的不合理之处，进行修改。由于我的第一稿论文确实再结构和组织上存在缺陷，我对很多地方进行了重写。于是，针对审稿人的意见，先是总的回复，我们改进了论文的结构，重写了一些章节来补充说明…问题（目的）。然后针对两处细节，进行补充说明，做到一一对应。

Thank you for the above suggestions. As suggested by the reviewer, we have improved the structure and organization of the manuscript. We have revised the Sections 3 and 4. We have deleted some redundant descriptions, and made supplementary explanations for some confusing descriptions.
(1) The overly detailed descriptions of graph neural network in Section 3 have been deleted. And the necessary information of the literature [114] in the second paragraph of subsection 4.1 has been added in the revised manuscript.
(2) The descriptions of CNN at the end of subsection 3.2.1 overlap with the first paragraph of Section 3 and the second paragraph of subsection 3.2. Therefore, the descriptions of CNN have been deleted in the revised manuscript.

Case 5：

I would suggest discussing more the details of the DNN models and their impact/relation on feature extraction and prediction. For example, in a CNN model, what are the impacts of using a convolutional layer+max pooling layer on the input protein or compound sequence? Similar discussions can be carried out on the layers of other convolution-based models, FFMLP-based models, and attention-based models.

在回复这个意见的时候要注意，审稿人列出了许多需要修改之处，我们一定要每一个都点到，不能遗漏任何一个，并且指明每一处修改在哪个地方。

Response:
Thank you for the above suggestion. We have added more details of some DNN models to illustrate their impact/relation on feature extraction and prediction. These DNN models include CNN+Max-pooling (in the second paragraph of subsection 3.2), Transformer (in the second paragraph of subsection 3.2), Graph attention network (in the first paragraph of subsection 3.3), stacked CNNs (in the third paragraph of subsection 4.2), FCDNN-based models (in the last paragraph of subsection 4.1), convolution-based models (in the last paragraph of subsection 4.2), attention-based models (in the first and last paragraph of subsection 4.3), and multi-modal-based models (in the last paragraph of subsection 4.4).

Case 6:

I would suggest using “Feed-forward Multi-layer Neural Networks” for the title of Section 4.1 which is currently DNN-based models.

当你不准备采纳审稿人的建议的时候，这个时候一定要注意措辞，要对审稿人进行婉拒，并且拿出十足的理由来说服审稿人。

Response：
Thank you for the above suggestion. In most cases, Feed-forward Multi-layer Neural Networks and Deep Neural Networks include convolutional neural networks and graph convolutional networks. Their input form is changeable, including one-dimensional tensor, two-dimensional tensor and three-dimensional tensor. While subsection 4.1 focuses on the methods processing one-dimensional feature descriptors. Fully connected deep neural network is a type of artificial neural network to process one-dimensional inputs, in which all the nodes in one layer are connected to the neurons in the next layer while there is no connection between nodes on the same layer. So, “Fully connected deep neural network-based models” may be more suitable as the title of subsection 4.1, which is “DNN-based models” in the original manuscript.

首先感谢审稿人的意见，然后委婉的指出审稿人意见中的不足指出，并给出充足的理由，最后提出自己的观点。注意文中的用词: “In most cases”," may be more suitable"等.
Case 7：当你完成不了审稿人的建议时，你应该详细说明原有，证明你努力过，只是没成功。

Response1：
Indeed, it will be more convincing if we get a comparative assessment on normal keratinocytes. However, the normal skin cell line TE 353.SK that we recently obtained could not be cultured stably under the condition of our lab at this time. We are also collecting human tissue samples from patients with primary melanoma and benign nevi. Based on our data, XXX expression will be analyzed in clinical samples by immunohistochemistry, and the correlation between XXX expression and the prognosis of melanoma will be determined in our subsequent investigations. Therefore, the referee’s concern is of importance for our further study, and we will show the results in our next paper for XXX.
Response2：
We are appreciative of the reviewer’s suggestion. Indeed, it will be more profound if we get the relevant results in vivo. We have purchased 10 thymus-null BALB/c nude mice (female, age 4-6 weeks) from the Animal Center of Chinese Academy of Medical Sciences. Osteosarcoma xenografts were established in nude mice according to a previous report (Gao et al., 2009) （我们去买了若干只小鼠，按照文献的方法操作）. Briefly, control lentivirus or XXX gene shRNA-expressing lentivirus infected XXX cells (1×105) were suspended in 0.1 ml PBS and then injected subcutaneously into the proximal tibia of each anesthetized nude mice (n=5 in each group). Unfortunately, the xenografts could not grow after inoculation for three weeks （满怀欣喜的接种到动物身上，然而实验这东西，哪能一遍就成呢，我们做了一遍，结果没能成瘤，实在是悲剧了）. The limitation of technical condition may be the main reason. Also, we doubt that the cell viability could be insufficient to initiate osteosarcoma tumorigenesis in vivo due to the long-distance transport. Therefore, we seek for the editor’s tolerance and understanding. Many thanks for your kind help!

总结

如果大家想系统的分析一下如何正确的进行Response，可以从Nature Communications观看大牛们的Response，在文章的末尾找到”Supplementary information“中，有些文章会”Peer Review File“。此外，EMBO旗下的杂志都附带详细的peer review过程可以直接下载，如
EMBO JOURNAL，建议大家在有空的时候可以多看看。