Open Notebook Science Solubility Project

Live melting point feed rdf: http://lxsrv7.oru.edu/~alang... - Andrew Lang

Live solubility feed rdf: http://showme.physics.drexel.edu/onsc... - Andrew Lang

How can I get the feeds to only list the latest 10 or 20? - Egon Willighagen

because PCA can be viewed as a dimension reduction method, so if you don't know which N of M descriptors to choose (M being large) the hope is that the first N PC's will contain the informative descriptors. - Rajarshi Guha

depends on other things... in case of multi linear regression, it's a must to get (somewhat) stable models... it addresses co-linearity between original descriptors - Egon Willighagen

I wouldn't say PCA is a must. It's just a way to get fewer descriptors - Rajarshi Guha

What's the drawback? You can't measure the importance of individual descriptors? - Andrew Lang

your coefficients gets messed up, because any linear combination will work... coefficients for descriptors can easily swap from positive to negative, resulting in multiple interpretations. e.g. xlogp increases Foo, while in another model is decreases that same Foo - Egon Willighagen

As long as the version history is not public anyway, I think you can safely go for keeping the post "unpublished" at a place where only your collaborators have a access. An option compatible with a public version history would be to encrypt the entry and to post the key for decryption later. - Daniel Mietchen

Carl - in my opinion requiring a password makes that part of the notebook completely closed. Selectively sharing data with collaborators, editors, etc. is standard practice in science, isn't it? - Jean-Claude Bradley

I'd say that it depends on what is useful to you and the project. If you want to share with collaborators even while the data is closed then a password is probably a good way to do that. Also means you can enable access for referees if appropriate. Either way if you have to do a manual release to make it public it is essentially closed until you do that. From my perspective one of the appeals of getting it up there even if not public is that it is at least easy to flick that switch when the time comes. - Cameron Neylon

But readers of the notebook will be frustrated if they think that you're hiding key/juicy things. In terms of public participation, that would be a little frustrating. The other thing I was going to say was that our IT guys at the Uni are very nervous of the idea of "switch flicking" to make bits of private books public. The reason being human error - a fleeting lapse of concentration... more... - Matthew Todd

The problem with that approach is that the user ends up using either the public one or the private one but not both. I know it appeals to IP and IT types but from a usability perspective its a great way to persuade people to use paper notebooks. Any credible system will have to have a good way of keeping parts private and helping the user decide how and when to make them public....and... more... - Cameron Neylon

BTW this is currently a problem for OSDD - you can access the data with a password but you have to agree to keep it confidential presumably because they want to retain the option to patent. But that makes the data effectively unusable - and it may even inhibit research. Once researchers see what is there they may avoid a research track to avoid giving the appearance of "stealing IP". They can't even credit the work behind the password because it is neither published or public. - Jean-Claude Bradley

I would start a separate password protected WP blog notebook. That way there are no annoying restricted posts in among your public ones. A separate blog also keeps author confusion to a minimum (or is that just me?). On publication you can export and import to the open notebook site, which will keep all experimental publication dates etc correct. - Dave Lunt

Thanks all for the feedback. I've been keeping the notebook completely open and attempting to make sure it really contained all content without substantial delay for about a year and a half, with daily entries. Recently (http://www.carlboettiger.info/archive...) I started some work on data I was asked not to share before publication. I could shift to a some-content, no delay model, but separate notebooks would be rather cumbersome. As security is easy to manage I just marked these posts private. - Carl Boettiger

I could make these few posts password protected instead of private -- I was curious if that would seem more open (you'd know I was doing something, see the title, could check back and see if I released it a year later, etc) instead of private, where they would not appear to exist. Just to clarify, 94% (currently) of posts are open without delay, but I need a mechanism to ensure the 6%... more... - Carl Boettiger

Carl - I don't think there is a right or wrong way to do it - it just depends on what is most convenient for you. For a substantially different project it sometimes makes sense to have a separate notebook for ease of processing and directing others to follow (we did this to separate the measurement of solubility from organic synthesis reactions). For a closed project I would think it is easier to have a separate notebook. - Jean-Claude Bradley

question: in your opinion, Egon, who should have done what instead and who should be doing what right now? - Claudia Koltzenburg

would this in any way resonate with your ideas: "Markets can be Egalitarian Meshworks with P2P Dynamics" http://blog.p2pfoundation.net/the-pla... - Claudia Koltzenburg

http://ontogenesis.knowledgeblog.org/ and http://taverna.knowledgeblog.org/ are 'blogged books' using Wordpress as a commodity platform for rapid publication. There are a number of posts on knowledgeblog.org detailing the motivations for this and the advantages it can bring. Blogs can be particularly useful in situations where a subject is too niche for a publisher to be especially interested, or when writing about a fast moving field where traditional publishing is simply too slow to keep up. - Simon Cockell

These are excellent examples! Thank you very much! I especially like the Taverna collaboration. I found Engelbart Book http://engelbartbook.com/ and Seaside programming text http://book.seaside.st/book but still looking. - Patricia F. Anderson

"I want to make an announcement that my book entitled The P=NP Question and Gödel’s Lost Letter is now published. It is available here at Amazon.com, for example. I started writing a book that became a blog. Now the blog has become a book." http://rjlipton.wordpress.com/2010... - JoeCamel

These are awesome! I am still working on that collection of books born as blogs / born digital for a blogpost. Need to get my Yahoo Pipes alternatives post out first ... - Patricia F. Anderson

Great work! Could you provide more details about the data you used for the nodes, edges and coloring? I am excited to look at this more deeply. - Don Pellegrino

I have uploaded all the data to the wiki: http://onschallenge.wikispaces.com/Experim... Gephi's quickstart guide is really good for showing how to do colours, etc: http://gephi.org/users... - Andrew Lang

But I don't think I can make either date unfortunately... - Cameron Neylon

Wave with details summarized: http://egonw.github.com/wave1... - Egon Willighagen

I thought Heiko had said it was open access, but today I am having the same problem you are. How bizarre! Friday, I had no trouble getting to it. Buzzard. Graham, I'mm DM you my email address. - Patricia F. Anderson

Delicious stream from the paper: http://delicious.com/dioc... . - Daniel Mietchen

The irony of this article being behind a pay wall... the upside is that more biologists will probably read it. - Greg Tyrelle

Folks, an update. The authors had been told that this would be free access. The editor believed that this would be free access. It WAS free access for a day, but then the publisher changed it. The editor agrees it needs to be free access and is negotiating with the publisher to restore this, but still in negotiations. Disappointing so far, but we are working on it. - Patricia F. Anderson

Great, keep us updated, I would like to read the article. - Greg Tyrelle

but defaults are pretty good - since RF's don't overfit, you can just use a large value of ntree, you will just loose out on time and storage - Rajarshi Guha

you must have generated that poicture froim the model, right? just use it for prediction - Rajarshi Guha

Thanks Rajarshi, I used ntree = 10,000? Make mtry big too? I was looking for a way to get the coefficients, e.g. LogS = c1 VCCLogP + c2 ATSm3 + ... - Andrew Lang

woah! how many obs are there? Probably don't need to go beyond 800! As for coefficients - there aren't any as such. You could look at individual trees (see Klimt for a way to vsualize ensembles of trees) - Rajarshi Guha

awesome Andy! This should help us out a lot - Jean-Claude Bradley

I've been trying to build a model - using mathematica - forward stepwise selection of descriptors (I just learned what this is called - I thought I came up with it myself): http://onschallenge.wikispaces.com/Solubil... but since I want to use it to make predictions, I'm worried about overfitting. I guess random forest was the wrong choice because I need an explicit formula. So I guess I want to check cross-validation at each step in my forward stepwise selection? - Andrew Lang

Yes, that's one way. But in general stepwise selection procedures can lead to biased models since you are throwing a certain descriptors with testing their combinations with others. Why do you need an explicit formula? - Rajarshi Guha

I'd like to make a webservice: SMILES -> Solubility in Methanol. - Andrew Lang

you don't need a formula for that? Just eval descriptors and get the prediction from the pre-exisitiung model. That's what my latest web deployed model service did (on toposome) - Rajarshi Guha

The way I do it is to get the descriptor values from the CDK-REST services and then plug them in to the formula and out pops the predicted value (all done with PHP). - Andrew Lang

Oh ok. But then you're stuck to OLS and OLS related models, right? - Rajarshi Guha

Yes - so I guess random forest is out. I wanted to try it because I am worried about overfitting and I guess introduced bias from stepwise selection now too. - Andrew Lang

How about using random forests to filter 'top' N descriptors, then do backward stepwise regression? - Andrew Lang

no difference according to ChemSpider - well phenacetium is a salt name (which I don't understand because amides can't really form salts easily) - Jean-Claude Bradley

ok good - book lists them separately but I bet he just didn't check. - Andrew Lang

Mickey, it isn't my field, but I'm pretty sure he means it in the ethnographic sense. - Michael R. Bernstein

If he finds an answer, I'd love to know. I have also looked for an aid to qualitative research, but not sure that it has the kind of structured data collection that makes for good software. Once data is collected and some of the thematic analyses are begun, then something like Excel becomes useful, but I suspect that is not what he means. - Mickey Schafer

D0r0th34, he says yes. - Michael R. Bernstein

D0r0th34, I'd like a pointer to such a product too. - Meryn Stol

Livescribe has been working on a transcription program, but I don't think it is up yet. Dragon speaking naturally has an excellent reputation, especially the more recent and pricey versions, and they are supposed to handle vocal variety better (most voice transcription programs are trained to a single voice). - Mickey Schafer

I am the colleague - and likely going to give it a go with TAMS Analyzer. Seems well suited for coding text transcribed from interviews and not too many features, just enough to code text adequately. See what happens this summer.... Thanks Michael for posting this question. - Mark Pugsley

spss - Dan owns Comicsforge.com