給Nature投稿：一個失敗的故事

60多年前（45，46年），浙江大學教師貝時璋，王淦昌，束星北，何增祿四位教授在Nature上共發表了6篇論文。自那以後浙江大學教師在就沒有在nature上以第一作者（或通信作者），且第一單位發表過論文（第二單位還是有的，例如童利民教授）。值得一提的是，上個月18日劉樹生教授的一篇論文，實現了第二個頂級期刊Science零的突破。

近20年來，浙大每年都有一些教授向Nature，Science投稿，主要集中在生科院，醫學院，農學院等。據我所知，其中汪元美，羅琛，吳平等教授的工作已經很接近發表在Nature，Science上的標準。劉樹生教授更是十年磨一箭，要不發都難。

2007年10月28日，我也斗膽敲了一下Nature的門。

07年初寫國家自然基金標書時，有一篇重要的參考文獻（發表在nature上），而且作者的實驗數據是公開的。然後我開發了一個算法進行計算，發現其中的最重要的結論是有問題的。於是我想應該可以發一篇好的期刊文章，就斗膽投了nature的子刊：nature medicine （影響因子：28.8）。投的欄目是brief communication。結果5天后退稿，沒有過編輯這一關，根本就沒能進入peer review（同行評議）階段。

他們說我是對nature的文章提出爭議，所以建議我投nature雜誌的Letter to the Editor欄目，當時我對nature的這個欄目發表的文章還不十分了解，以為在這個欄目發了，在SCI收錄時，文章的類型可能是Editorial Material，那麼這篇文章連一篇SCI都不算。這是因為中國科技信息研究所統計各單位的SCI論文時，只有文章類型為article，review，letter才算SCI論文。那樣我不是虧大了？另一方面，投nature我還是有些怯場，和世界級牛人在國際頂級期刊PK, 怕怕的。其實主要還是自己底氣不足，雖然我叫板他們的最主要結論，但不是全部結論，我甚至通過計算重複了他們的另一個結論 （後來我把他們這個結論也從另一個側面否定了)。結果，我沒有聽從Nature medicine編輯的忠告，我投了New England Journal of Medicine。

“新英格蘭醫學”SCI影響因子51.3，是世界上影響因子排第二的期刊，醫學期刊no.1，浙大在這個期刊也沒有實現零的突破。新英格蘭醫學也發有letter to the Editor欄目，對於挑戰其他刊物的文章它也發（一般期刊只刊登挑戰本期刊上發表的論文的Letter）。投稿的結果還是：退你沒商量！不過，14天后才退。這也算是一個進步，正好兩周。國際著名期刊不像中國期刊，他們的編輯都是專家級的，能使自己的文章在新英格蘭醫學編輯部待上兩周，雖然很失望，但心中還是有些快慰。

投到Nature去！這是因為，Nature偶爾還發一篇data reanalysis文章，就是對已經發表的一些數據進行再分析，但是投稿前要給編輯部發一個擬投論文的大綱。這樣的文章如果發了，SCI檢索時文章的類型是Article。因此，就斗膽發了一個大綱過去，3天后又退了。

事到如今，還是不甘心，就硬着頭皮投nature的“Letter to the Editor”。其實現在的nature根本就沒有“Letter to the Editor”這個欄目，這個欄目現在叫Brief Communications Arising，這個欄目把nature原來的brief communication也合併進去了，SCI收錄時把這個欄目的文章歸類為Letter類型。還沒有把manuxxxx上傳上去，再研究Author inxxxxation 發現有這樣的說法：

Brief Communications Arising are exceptionally interesting or important scientific comments and clarifications on original research papers or other peer-reviewed material published in Nature.

Submissions should challenge the main conclusions of the Nature paper and contain new, unpublished data to support the arguments.

Submissions that contradict only part of the Nature paper are not considered unless they concern a matter of exceptional interest.

而且：

* Comments should be sent to the authors of the paper under discussion before submission to Nature, so that disputes can be resolved directly whenever possible and points where both parties agree removed from the xxxxted contribution. Allow 2 weeks for the original authors to respond.
* When a contribution is xxxxted to Nature, copies of correspondence with the original authors should be enclosed for the editor's inxxxxation, even if the original author has failed to respond. The correspondence should accompany the submission as an attachment clearly labelled as 'Correspondence with the Nature authors'.

很顯然，我們只是挑戰原作者的部分結論，而談不上main conclusions，因此只好暫停投稿過程，結果nature發email要我們繼續上傳manuxxxx。由於信心不足，只好完全放棄投稿。

沒有金剛鑽，別攬瓷器活。不過心裡還是希望投一個高影響因子的期刊，類型是一般研究文章。因此，我又回過頭來看原文，再進行計算。原來有些結論由於醫學知識沒有明白，所以，這次和co-author們見面，面談，後來又打了很多電話交流。終於徹底把文章研究明白了，結果又發現的nature原文作者的兩個破綻，這時，我們的底氣足了很多。完全可以談得上挑戰其main conclusions。所以，我們不打算投其他期刊，準備再殺它個回馬槍。因為我們覺得有理由理直氣壯地投nature了。我們的manuxxxx摘要如下：

Both mRNA and microRNA (miRNA) profiles can be used for the classification of human cancers. In the experiments done by Lu et al., the samples used to examine the miRNA and mRNA expression profiles were identical, making their two studies directly comparable. However, the main conclusions of their study are not correct, including (1) sufficient cancer diagnostic inxxxxation is encoded in a relatively small number of miRNAs; (2) among the epithelial samples, gut-derived samples all clustered together when using miRNA expression profiles, whereas this phenomenon was not observed in mRNA expression profiles; and (3) by using miRNA expression profiles, poorly differentiated tumors (PDT) could be classified more accurately, whereas mRNA expression profiles were almost completely inaccurate when applied to the same samples. Here we show that inxxxxation encoded in miRNAs is not sufficient to classify cancers; gut-derived samples cluster together more perfectively when using mRNA expression profiles compared with miRNA's; and PDT could be classified by mRNA expression profiles at the accuracy of 100% (versus 93.8% of miRNA's). Further more, we proved that mRNA expression profiles have higher ability of tissue classification than that of miRNA's.

編輯的初步決定（11月1日）：把我們的文章給nature原文章作者

Dear Dr Peng

Thank you for xxxxting your Communications Arising entitled "Cancer classification by expression profiles: microRNA versus mRNA" to Nature. This message is to tell you that we are sending your paper out to the Nature authors for their response according to our policy (see our guidelines to authors of Communications Arising www.nature.com/nature/authors/gta/commsarising.html). Once we have received their reply, we shall decide whether or not to send the exchange out for review.

This decision is reached after discussion with the appropriate specialist editors and depends on a number of factors, including the likely impact of the criticism, the topicality of the discussion, and its interest to the non-specialist readers of this section of the journal (many debates are referred to the specialist literature at this point or the issues are addressed in the xxxx of a published clarification from the criticised authors).

If we have several comments under consideration on one of our published papers, this may introduce delays while we coordinate these steps in the editorial process, so please allow four weeks before sending us any status enquiries.

Please note that consideration of your comment by Nature is contigent on there being no discussion of its content with the media in advance of publication (see www.nature.com/nature/authors/policy).

Yours sincerely

Angela

Angela K Eggleston, PhD

Senior Editor

Nature

――――――――――――――――――――――――――――――

Nature的guideline寫道：

If the Nature authors do not respond within 10 days of receipt of the comment, the editor will proceed without the response. Late Replies may not be considered for publication.

也就是說，nature按理只給10天給原作者response。那麼應該在11月11日前應該收到原作者的response。這一個Email同時提醒我們有可能要等4個星期，再問文章的結果。

等了六個星期後，我發了email詢問：

我們的Email全文如下：

Dear editor,

After the submission of our manuxxxx entitled "Cancer classification by expression profiles: microRNA versus mRNA" (tracking number: 2007-10-11332) to Nature, six weeks have passed. So I would like to inquire the status of our manuxxxx.

We think that our comments on the original paper of Nature will cause widespread interest to both experts and non-experts. This is because microRNA is a hot spot in life science; the author of the original paper is a VIP in cancer research based on gene expression profiles, with a SCI H index of 45, total citations more than 15,000 times, and this paper citied 365 times; and the algorithm we proposed outperxxxxs the recent state-of-the-art algorithm. Moreover, by using other state-of-the-art algorithms, we can also prove that the original author's main conclusions are wrong.

I am looking forward to your reply. Thanks a lot.

Sincerely yours,

Sihua Peng, PhD.

2007-12-10

最後的結局：經過44天的漫長等待，最後還是退稿！

――――――――――――――――――――――――――――

退稿信：

――――――――――――――――――――――――――――

11th December 2007

Dear Dr Peng

Thank you for your submission to Brief Communications Arising, which I am afraid we must decline to publish. As is our policy on these occasions, we showed your comment to the Nature authors, and their response is enclosed below.

In light of this reply and of the competition in this section of the journal, we have regretfully decided that publication of this debate is not justified: this is because in our view it would not add sufficiently to our understanding or otherwise clarify the issues for our readers.

Thank you again, however, for writing to us.

With kind regards

Angela

Angela K Eggleston, PhD

Senior Editor

Nature

――――――――――――――――――――――――――

以下是原nature作者的response

――――――――――――――――――――――――――

Nature Author:

Peng and colleagues describe their efforts to reanalyze the miRNA results reported by Lu et al.

One of the great triumphs of the genomic era is that datasets are now publicly available, and can be analyzed by the entire scientific community, often leading to new insights into old data, or refining methodologies.

The present manuxxxx is somewhat difficult to xxxxuate because of the paucity of methodological detail. However, it appears that the stunning classification accuracy described by Peng et al (that was lacking in the Lu et al paper), is a result of the two-step feature selection approach. Indeed, it is not at all surprising that such an approach would yield the appearance of improved classification. The problem, well-known to those experienced in genomic analyses, is that there is a serious danger of over-fitting the model to a given dataset -- particularly in this high dimensional space. Peng et al use a leave-one-out classification method to address, this, but it is applied AFTER feature selection, at which point the damage is already done. It is imperitive that the feature selection be done outside the cross-validation loop, and ideally, the model would be applied to an independent set of samples. Likewise, it is not surprising in the least that if features are first

selected (i.e. for gut vs. non-gut derived tumors), that the samples will appear to cluster together. But clustering in the space of features selected in a supervised manner is meaningless, and the result is without question a result of serious over-fitting.

While it certainly remains possible that, as the authors suggest, the mRNA profiles contain diagnostically useful signatures, the proposed approach is not convincing, and indeed is fundamentally flawed. If this interpretation is a misunderstanding, then this issue could be clarified in the xxxx of a full-length report, rather than a correspondence, where such detail obviously cannot be included.

―――――――――――――――――――――――――――――――――

總結:

1．Nature把我們拒了，還不給我們的申辯的機會。所以，心裡很不服氣，到這裡也算發泄一下。他們在author guideline里寫道：

The editors will not consider appeals against decisions not to publish Communications Arising from Nature Articles and Letters unless the grounds for appeal consist of a previously overlooked and important scientific point and are clearly explained in these terms.

2．我們在稿件狀態詢問信中寫道：Moreover, by using other state-of-the-art algorithms, we can also prove that the original author's main conclusions are wrong. 因此希望搞生物信息學，數據挖掘，機器學習的高手繼續挑戰這篇文章的主要結論的正確性。原文是：Nature 435, 834-838, 2005.

3．這是我第一次以第一作者（兼通信作者）的身份投稿，我的文章寫法，與編輯及nature原文作者的溝通Email的措辭有否不妥的地方，請有經驗的老師指教一下，以便下次不犯同類錯誤。

4．這個失敗的故事也許能給各位投nature的老師一些有益的啟示。浙大在走向世界一流的過程中，需要更多像劉樹生教授那樣的在Science，Nature上發表文章的人。我覺得，浙江大學如果不在這兩個期刊每年發表5篇以上的論文，不能稱為世界一流大學（我們浙大建校110年總共只發了7篇）。像哈佛大學那樣的世界頂尖一流高校，在這兩個期刊每年共發表100篇以上（只算通信作者）。