Re: Dear Taiwan Fellows, pls judge yourself

看板Biotech (生命科學)作者mitbbsuser (mitbbs user)時間19年前 (2006/12/17 04:05)推噓6(6推 0噓 4→)

留言10則, 4人參與討論串3/8 (看更多)

※ 引述《nightcatman (夜貓)》之銘言： : ※ 引述《mitbbsuser (mitbbs user)》之銘言： : : Beware: some images are huge and IE or other explorers might zoom out them : : automatically. For the best result, pls view at 100% size. Thank you. : : The original Figure2C published on Cell with illustrations : : http://capa.zoto.com/img/original/c910fb371829d67b09568e6cb247292d-.jpg

: : Then here comes the photoshop difference blending results: : : http://capa.zoto.com/img/original/50d0d142ac7730621cfb559c2080c14b-.jpg

: : http://capa.zoto.com/img/original/95ca4d2a9ca9012fc24a91e00df2fe2c-.jpg

: : http://capa.zoto.com/img/original/f53e3344f877c27a8fab385b6b1a96e8-.jpg

: Not really. All of the "identical pattern" regions marked in the : published figure are actually not "identical" but just "similar" : if they are enlarged in a very high resolution and are compared : pixel by pixel. : In my opinion, only the edge between the -30 and -40 regions of : Fig2c is questionable. Well, you even start comparing every pixel. So why not bother yourself starting explaining those black dots and air bubbles which emerge at the same positions? Have you tried open a image in the same software, and save as jpeg or other format with compression to some extent? Even all the save-as images are from the same image with the exactly same parameters, they won't be exactly same pixel by pixel. Surprised,huh? Try yourself. Don't make any harsh conclusion if you haven't tried. Photoshop is only one of the approaches we used. Matlab work will tell you more specifically about how low the odd to get such identical pattern at background parts. Besides, we always mentioned there IS the odd to get such data without fabrication, though the odd is way to low. I think Cell thinks similarly to us, so they give Dr. Chang a Chance to prove such tremendous fortune which led them to such wonderful data. Well, it seems their "luck" just disapeared when it was in real need. : : Then here comes the so called repeat figure2C-1, which you could download : : yourself through the link lnalna posted. : : http://capa.zoto.com/img/original/f13297b8b79dc41a1a42b9c57330fa3a-.jpg

: : then a little photoshop adjustment like this: : : http://capa.zoto.com/img/original/a531df68a7ec7a0a46091555b5cbda00-.jpg

: : here is the final result: : : http://capa.zoto.com/img/original/1aebae3ec7ba8b874ae6aaf935ba596b-.jpg

: It is questionable. I agree. : : Some people also use Matlab Corr(2) calculation and ImageJ Frourier Transform : : to analyse the original figure2C published on Cell and got the same conclusion. : : I won't make any conclusion here. Judge yourself with your eyes. : : btw, the methodology shown here is exactly the same method which the top : : journals use, like JCB. : In this case, statistical correlation coefficient can't tell : much because these regions are expected to be natually similar, : i.e., the correlation coefficient should be natually very high. : So, unless you can get 100% correlation coefficient, you can never : confidently say the high correlation coefficient is due to the : nature similarity of these regions, or the human manipulation. : BTW, appealing to authority is a kind of logical fallacy. Yes, I : mean your last sentence. are expected to be naturally similar? you mean the backgrond? man, have you ever run gel yourself? the matlab data has controls directly from figure2C. Again, I already told you variations caused by image manipulations could account for the 1% incorrelation. Actually, according to the calculation doen by another student, it's not 1%, it's 0.1%. 10 folds less then those controls. : : If you want to argue about "false positive" due to the image compression. : : You'd better not. The original images are 500dpi, 8bit, grayscale TIFF. : : and JPEG compression could only give you some 8*8 squares. Here is a control : : from another Gel done by the other lab. : : http://capa.zoto.com/img/original/98feb861409c0c300478321085f28812-.jpg

: : although the bands are quite similar between different lanes, the background : : patterns are obviously different. : Strictly, this "control" can't prove anything. You can just say : usually there may be some differences between similar lanes, but : you can't say it's 100% a fake if there are only a few differences : can be found in pixel level between similar lanes. : At last, I would like to give some suggestions to people on mitbbs: : Question the points which are really questionable with precise words. : Don't just indiscreetly attact every points you think you are able : to attact with overstated and emotional words. Well, Well, why did you just skip the figure2C-1? huh? How did you explain that? Facing such obvious fabrication, you merely think it is questionable? Or just because you don't know how to explain it? We certainly are happy to take your advice on the attitude of some mitbbs guys, BUT, your explanation is weak and lacks actual test. Try convincing us with your own test or calculation, please. But in my point of view, it's meaningless, Cell already gave them what they deserve. I am not the judge, neither are you. Besides, the reason I spend my time here is I don't like to see a science affair being involved with politics or relationship between people from mainland and Taiwan. We really respect Taiwan scientists and a lot of researchers from the 2 sides of the strait work together either in China or in other countries. -- ※ 發信站: 批踢踢實業坊(ptt.cc) ◆ From: 69.227.164.126 ※ 編輯: mitbbsuser 來自: 69.227.164.126 (12/17 04:21) ※ 編輯: mitbbsuser 來自: 69.227.164.126 (12/17 04:23) ※ 編輯: mitbbsuser 來自: 69.227.164.126 (12/17 04:25)