Government shouldn't pay for something that the private sector is already doing. Full stop.
Perhaps we should think about what PubChem is doing relative to science, not the private sector. For peer reviewed publications, especially for research funded by the public, the only way to validate the results is to have access to them in the context of the science being studied.
PubChem is not just storing molecules, but the raw data of the interaction of these molecules with biological assays. The molecules, by themselves, are probably not that useful. This is something the private sector is not really doing.
Full disclosure here, my company writes the software that NIH is using to write PubChem. They do get it for no cost though.
Responses like this are bound to be ignored out of hand. Metakit is used widely in certain areas such as bioinformatics and chemistry. I have found that it scales far better than sqlite for certain circumstances and is very, very fast when used correctly. I have made several metakit applications as fast as berkeley db and with full relational algebra support! I liked it enough that I wrote a Dr. Dobb's article on metakit.
The fact that views can automatically hold subviews is a huge bonus. I.e tables in tables so there is no need for joins.
I have a 1.5 gig metakit database the millions of rows per table that dies under sqllite. Conversely, I have some complicated sql that works very well under sqllite that metakit doesn't handle very well.
Now, I have to mention the new metakit virtual machine thrive, while currently only implemented for integer datafields, is incredibly amazing without really any optimization, but, alas still a while out.
Now, if you are stuck with sql, by all means use sqllite, your transition will be much, much faster. But if you want to use a full relational algebra approach, try metakit. If you want to do financial transactions, try ksql (built on kay) holy crap is that fast for financial data.
Here is a financial benefit for you, I (as an Open Source author), can take my code with me. Sure someone else can use it as well if they happen to have a copy or ask me for one. A lot of coding these days is in a service-based economy. Having IP I produce for one client/business translatable to the next earns me more money quicker.
The nice thing about this is once you are in the door there is a snow ball effect. "Hey, I've can solve your problem and I can do it in X days at the cost of Y using this GPL'd piece of code (that you will be given at the end) or I can make a new implementation which you will own that will take 10X days and cost 15Y." Which one would you choose as a business?
The problem is that if the original source contains contributions from anyone but the original author, without a lot of leg work they cannot change the license on the code and release a non-free version.
Now, this might not be the case here but it does show you to be careful if you accept contributions and bug-patches and want to pull a similar deal in the future.
Yes - there are formulae for computing forces. But no, they don't bear much relation to reality Any simulation running anywhere near realtime is likely to be a purely classical simulation - balls and springs.
You are preaching to the choir and any computation chemist worth his salt will tell you how far we have to go. The Merck force field actually performs better at generating bio-active conformations when electrostatics is turned off, mainly because the electrostatics of the protein completely masks the electrostatics of the ligand.
I sit next to the guy who wrote Fred and he will gladly admit one of the best predictive scoring function is the order of the entry of the compound in the pharma companies database. I.e. old compounds are not active - at all.
If you care to check you'll find very few cases of a drug discovery, say, resulting from a theoretical prediction about receptor binding.
I disagree with this, but it does depend on your starting point. If you have a known molecule bound to a protein, we are very good at finding new binders, even from different chemical classes. That being said, de-novo design leaves a bad taste in my mouth.
Finally, looking at the Scripps' website, they seem to be looking at very macro-structures like viruses. These were most likely imaged in solution using an nmr to generate the computer models. There is some pretty good software for doing these large scale fits (the homepage appears to be down). Essentially they are not doing drug-protein interactions but larger scale models. The question they are asking is if these molecules interact, how do they fit? They aren't predicting interaction, they already know that.
This might be a cart-before the horse because if they generate a computer model that packs the structures together, it isn't surprising if the 3d printed ones do. I have more faith in this than predicting drug-receptor bindings.
I love NPR and am quite the news junky, but when the NPR reporter said under Fiorina, HP became a "computer company" I realized that the HP I knew died a long time ago:)
bzzzt, nice try. You do know that you can run bittorrent without the GUI don't you? Check out btlaunchmany.py. This has multiple benefits including that you can download several torrents in the same process plus the memory footprint is much, much, smaller. There is even a command to download all torrents in a directory, very useful stuff.
The size of the application comes from one thing, wxWidgets which, you guessed it, is written in C++ and not python. The MacOS version runs directly on ObjectC and is much smaller. There might be a direct windows port, python can interface with the win32 api, check out venster. So now you have two options for reducing the size of bittorrent, get coding!
Obligatory rant: If you are tired of the BitTorrent community using python, then write it in another language and stop complaining, nothing in life is really for free.
and that is pretty much it. Most of the standard library is supported out of the box and you can easily include special case code. It even handles templates pretty well. Using their new fangled directors, we can even subclass C++ classes in python itself! It also supports java, guile, C#, ruby...
Maybe I overstated the issue, but I know that principle components of Pratt and Whitney engines, especially some air-intakes are completely designed in Russia and it is not an insignificant piece of the development process.
I am in agreement with you, but these companies are in fact companies and some of their work does get outsourced.
A large chunk of our military aircraft production, especially engines, is out sourced to russian engineers. This is beginning to happen in a lot of fields where design work is completed in Russia and production happens in China/India. So if you want to study aircraft engines, where would YOU go? I would go to A) where they are being designed and B) where they are being built.
I've begun to notice that entrepenuers are following the design and production. This business model is generating hot beds of innovation out side the country. If you follow the history of technological innovation, the production and design areas are critical. Sarnoff, Menlo Park, Xerox Park, Silicon valley, etc. You can't design in a vacuum and being near the technicians and engineers that actually make what you are working on is essential.
The way our military is currently structured, I can't believe that anyone would consider loosing scientific dominance would not be a matter of national security.
Our business here have this wierd notion that China and India are second class academically so it is okay to outsource engineering and techinal work because American's will always be the innovators. I have always thought that this was stupid and I'm seeing now that this is simply isn't true.
I actually thought that the XUL documentation was pretty good. It was complete enough that (as of 2001) I had written an automated display layer that could read xul xml files in python and display them with Tk/Qt or wxWidgets.
The effort of doing that pretty much made me realize that your other comment is quite true, XUL is way too embedded in mozilla. My python+wxWidgets display layer was smaller than the mozilla installation by far.
However, XUL + mozilla + xpcom + python is still incredibly powerful, Active States' Komodo is written in this fashion and is quite popular. David Ascher gives quite good talks on the development process. I would love to see him write a book on the subject. A good (if dated) interview is located here.
In this case there is some wonderful synergy between popular conceits (i.e. the wonder of social networks) and science (i.e. the wonder of protein networks). One interesting aspect of both such networks is the concept of hubs.
Let's take the yeast protein interaction network as an example. This network describes the physical interaction of various proteins in that protein A binds to protein B and does something useful, like telling your mitochondria to make more energy. It turns out that some proteins interact with WAY more proteins than on average. In yeast some interact with >260 proteins while the average is around 3. For social networks, think of Kevin Bacon here.
Now, well add some complexity. Suppose you want to initially model a human disease in yeast. Now you have to find the networks that the human has in common with the yeastie. If this happens to include the proteins associated with the disease you would like to model, bang, you're in luck. Chances are that this small network will include a hub, but in my experience they are just hangers-on and not really related to the disease model.
Now back to social networks. Let's suppose you identify a collection of the population with some desireable relationship, let's say for example that you notice that a cluster of people tend to buy low-end luxury cars. You can annotate their small network with as much data as possible, where they went to school, what year, the cost of the houses they own, where they live, etc. This becomes a kernel from which you can find other social networks in other places. Now that you have this matching piece, who would you choose to give the wicked low discount on a BMW? You guessed it, the hub. More people would see her/him driving along beaming and glowing in a spanking new beamer. I can almost guarantee that this will happen at some point in the future. Currently the hub's are celeberities because they are 'seen' by more people, but it is kind of impersonal.
In any case, the main problem is that social networks are HUGE so much more complicated to search in this way. If you are interested, you can check out my paper in PNAS
Now if I could just find some investors for the social network side...
Well, I find that open-sourcing software I write is a form of job security. In a nutshell I can take software from one company that folds or fires me to another.
The question you should ask yourself is are you likely to make money on the software itself or whether you are more likely to make money by using the software during your job.
Open-source actually allows you to do both, by the way.
This really seems like you trying to get something for nothing. Just so we are on the same page, I also develop for an academic institution. There are plenty of good widget systems that are free to use, wxWindows comes to mind for instance. It just boils down to the fact that you don't want to release your source code.
Trolltech has a free as in freedom liscense, take it or leave it. I have a lot of respect for them for doing this. Their source code is released and open to the public. They seem to have made a compelling product since you would like to use it. This is the dream for most programmers:)
You mention shareware as a way for developers getting payed but in the same breath don't want to pay TrollTech for their work.
Okay, this is a bit harsh, I think, based on your post. What do you think should be a per-seat price for TrollTech? Here is a potential alternative: BlackAdder, which is an integrated Qt + Python gui designer and runtime, runs at $400 a seat or $300 a seat for a developer networks. Python + Qt royalty free = $400 = yummy.
By the way, having no real relevance to your post, as someone who works in academia and insists that every government funded program that I work on releases open source software, I'm assuming your proposals would involve seeking government grants which, by the way, I pay for in taxes. Why isn't grant funded source code GPL'd in the first place since I'm paying taxes for it?
X11 is not the desktop. X11 provides the layer on which the desktop can run. I can prototype my own desktop on my linux box, reconfiguring FVWM or windowmaker to match the PDA metaphor and then run it on the PDA.
Hell, having a remote PDA desktop is a killer app. Forgot your contact information? No problem, just page your pda to get it out of sleep mode, and then ssh a new xclient to it.
I'm answering this a bit late in the game myself, but if you want a document format than includes the entire document including the schema, binary data, query capability, and, in fact, the source to query itself, it already exists.
In case your wondering, it's called LISP.
To anyone skilled in the art (as patents demand) doing the same thing in XML is a trivial extension.
It's these cases where philosophy can bear fruit or really confuse the issue.
On one side if the coin there are the "platonists" who consider math as the uncovering of ideal, eternally existing, abstract objects. On the other there are the "formalists" who consider mathematics as more of a game where theorems are developed logically from axioms chosen arbitrarily.
Then there is Rueben Hersh's connotation that mathematics is what mathematicians do. In lay-man's terms, without mathematicians there would be no math. He further implies that the math we know is decided by the mathematicians. Consider fluxions versus calculus, two different solutions to the same problem and one was superior. They came about in two different ways because the principle mathematicians were different people.
If we follow the platonists approach then math should not be patentable because it just is. A platonist would consider patenting math like patenting a mountain.
Hersh would contend that math is the creation of the mathematician and hence, as a product of personal endeavor, should be patentable.
In terms of the courts view on software, we are stuck in a Hershian situation, so what is the solution? If we had known that this would be the case, the EFF (electronic frontier foundation) or the GNU project could have started patenting software twenty or thirty years ago, thereby capturing the axioms on which software was founded. THIS is what should piss people off. It pisses me off in computer science and biology. Companies are standing on the shoulders of giants who didn't have the opportunities available to them, especially when most of the underlying infrastructure was built with public funds.
Patents are expensive, but there goal is to offer short-term incentives for development in order to release knowledge into the public domain. Sometimes we forget that patents are designed to release knowledge. However, I would favor a progressive patent law where software patents last only five years and drugs targeted at monogenic diseases that don't affect many patients (like huntington's disease and spinal muscular atrophy) last for forty.
Unfortunately she's made enough compensation that she will never have to work again, ever. Imagine how the nature of some corporations would change if the top executives would also need to have jobs next year.
Not that I'm too flaming a liberal, I want to amass my own pile of swag but the current terrible statistic is that America has the one if the highest disparities between high and average salaries in the world.
Slashdot Mirror
Perhaps we should think about what PubChem is doing relative to science, not the private sector. For peer reviewed publications, especially for research funded by the public, the only way to validate the results is to have access to them in the context of the science being studied.
PubChem is not just storing molecules, but the raw data of the interaction of these molecules with biological assays. The molecules, by themselves, are probably not that useful. This is something the private sector is not really doing.
Full disclosure here, my company writes the software that NIH is using to write PubChem. They do get it for no cost though.
The fact that views can automatically hold subviews is a huge bonus. I.e tables in tables so there is no need for joins.
I have a 1.5 gig metakit database the millions of rows per table that dies under sqllite. Conversely, I have some complicated sql that works very well under sqllite that metakit doesn't handle very well.
Now, I have to mention the new metakit virtual machine thrive, while currently only implemented for integer datafields, is incredibly amazing without really any optimization, but, alas still a while out.
Now, if you are stuck with sql, by all means use sqllite, your transition will be much, much faster. But if you want to use a full relational algebra approach, try metakit. If you want to do financial transactions, try ksql (built on kay) holy crap is that fast for financial data.
Ahh grasshopper, you are learning the zen of microsoft.
If an O/S works in the forest but no one is around to see, will it still crash for Bill?
The nice thing about this is once you are in the door there is a snow ball effect. "Hey, I've can solve your problem and I can do it in X days at the cost of Y using this GPL'd piece of code (that you will be given at the end) or I can make a new implementation which you will own that will take 10X days and cost 15Y." Which one would you choose as a business?
Now, this might not be the case here but it does show you to be careful if you accept contributions and bug-patches and want to pull a similar deal in the future.
Just like microsoft in general, it tries to make things monotone.
I sit next to the guy who wrote Fred and he will gladly admit one of the best predictive scoring function is the order of the entry of the compound in the pharma companies database. I.e. old compounds are not active - at all.
I disagree with this, but it does depend on your starting point. If you have a known molecule bound to a protein, we are very good at finding new binders, even from different chemical classes. That being said, de-novo design leaves a bad taste in my mouth.
Finally, looking at the Scripps' website, they seem to be looking at very macro-structures like viruses. These were most likely imaged in solution using an nmr to generate the computer models. There is some pretty good software for doing these large scale fits (the homepage appears to be down). Essentially they are not doing drug-protein interactions but larger scale models. The question they are asking is if these molecules interact, how do they fit? They aren't predicting interaction, they already know that.
This might be a cart-before the horse because if they generate a computer model that packs the structures together, it isn't surprising if the 3d printed ones do. I have more faith in this than predicting drug-receptor bindings.
Have you actually ever tried to compile with pascal? It's too damn fast, I'd never get a chance to read slashdot again. Long live templated C++!
I love NPR and am quite the news junky, but when the NPR reporter said under Fiorina, HP became a "computer company" I realized that the HP I knew died a long time ago :)
The size of the application comes from one thing, wxWidgets which, you guessed it, is written in C++ and not python. The MacOS version runs directly on ObjectC and is much smaller. There might be a direct windows port, python can interface with the win32 api, check out venster. So now you have two options for reducing the size of bittorrent, get coding!
Obligatory rant: If you are tired of the BitTorrent community using python, then write it in another language and stop complaining, nothing in life is really for free.
and that is pretty much it. Most of the standard library is supported out of the box and you can easily include special case code. It even handles templates pretty well. Using their new fangled directors, we can even subclass C++ classes in python itself! It also supports java, guile, C#, ruby...
It is really an amazing piece of work.
I am in agreement with you, but these companies are in fact companies and some of their work does get outsourced.
I've begun to notice that entrepenuers are following the design and production. This business model is generating hot beds of innovation out side the country. If you follow the history of technological innovation, the production and design areas are critical. Sarnoff, Menlo Park, Xerox Park, Silicon valley, etc. You can't design in a vacuum and being near the technicians and engineers that actually make what you are working on is essential.
The way our military is currently structured, I can't believe that anyone would consider loosing scientific dominance would not be a matter of national security.
Our business here have this wierd notion that China and India are second class academically so it is okay to outsource engineering and techinal work because American's will always be the innovators. I have always thought that this was stupid and I'm seeing now that this is simply isn't true.
By definition, just by receiving a patent you have "leaked" into public domain. The patent just gives you a temporary stay.
Komodo is written in XUL with xpcom+python.
The effort of doing that pretty much made me realize that your other comment is quite true, XUL is way too embedded in mozilla. My python+wxWidgets display layer was smaller than the mozilla installation by far.
However, XUL + mozilla + xpcom + python is still incredibly powerful, Active States' Komodo is written in this fashion and is quite popular. David Ascher gives quite good talks on the development process. I would love to see him write a book on the subject. A good (if dated) interview is located here.
Let's take the yeast protein interaction network as an example. This network describes the physical interaction of various proteins in that protein A binds to protein B and does something useful, like telling your mitochondria to make more energy. It turns out that some proteins interact with WAY more proteins than on average. In yeast some interact with >260 proteins while the average is around 3. For social networks, think of Kevin Bacon here.
Now, well add some complexity. Suppose you want to initially model a human disease in yeast. Now you have to find the networks that the human has in common with the yeastie. If this happens to include the proteins associated with the disease you would like to model, bang, you're in luck. Chances are that this small network will include a hub, but in my experience they are just hangers-on and not really related to the disease model.
Now back to social networks. Let's suppose you identify a collection of the population with some desireable relationship, let's say for example that you notice that a cluster of people tend to buy low-end luxury cars. You can annotate their small network with as much data as possible, where they went to school, what year, the cost of the houses they own, where they live, etc. This becomes a kernel from which you can find other social networks in other places. Now that you have this matching piece, who would you choose to give the wicked low discount on a BMW? You guessed it, the hub. More people would see her/him driving along beaming and glowing in a spanking new beamer. I can almost guarantee that this will happen at some point in the future. Currently the hub's are celeberities because they are 'seen' by more people, but it is kind of impersonal.
In any case, the main problem is that social networks are HUGE so much more complicated to search in this way. If you are interested, you can check out my paper in PNAS
Now if I could just find some investors for the social network side...
Well, I find that open-sourcing software I write is a form of job security. In a nutshell I can take software from one company that folds or fires me to another. The question you should ask yourself is are you likely to make money on the software itself or whether you are more likely to make money by using the software during your job. Open-source actually allows you to do both, by the way.
Yeah, but they can get eight Nighy's for one pu so it all works out in the end.
Trolltech has a free as in freedom liscense, take it or leave it. I have a lot of respect for them for doing this. Their source code is released and open to the public. They seem to have made a compelling product since you would like to use it. This is the dream for most programmers :)
You mention shareware as a way for developers getting payed but in the same breath don't want to pay TrollTech for their work.
Okay, this is a bit harsh, I think, based on your post. What do you think should be a per-seat price for TrollTech? Here is a potential alternative: BlackAdder, which is an integrated Qt + Python gui designer and runtime, runs at $400 a seat or $300 a seat for a developer networks. Python + Qt royalty free = $400 = yummy.
By the way, having no real relevance to your post, as someone who works in academia and insists that every government funded program that I work on releases open source software, I'm assuming your proposals would involve seeking government grants which, by the way, I pay for in taxes. Why isn't grant funded source code GPL'd in the first place since I'm paying taxes for it?
Hell, having a remote PDA desktop is a killer app. Forgot your contact information? No problem, just page your pda to get it out of sleep mode, and then ssh a new xclient to it.
In case your wondering, it's called LISP.
To anyone skilled in the art (as patents demand) doing the same thing in XML is a trivial extension.
On one side if the coin there are the "platonists" who consider math as the uncovering of ideal, eternally existing, abstract objects. On the other there are the "formalists" who consider mathematics as more of a game where theorems are developed logically from axioms chosen arbitrarily.
Then there is Rueben Hersh's connotation that mathematics is what mathematicians do. In lay-man's terms, without mathematicians there would be no math. He further implies that the math we know is decided by the mathematicians. Consider fluxions versus calculus, two different solutions to the same problem and one was superior. They came about in two different ways because the principle mathematicians were different people.
If we follow the platonists approach then math should not be patentable because it just is. A platonist would consider patenting math like patenting a mountain.
Hersh would contend that math is the creation of the mathematician and hence, as a product of personal endeavor, should be patentable.
In terms of the courts view on software, we are stuck in a Hershian situation, so what is the solution? If we had known that this would be the case, the EFF (electronic frontier foundation) or the GNU project could have started patenting software twenty or thirty years ago, thereby capturing the axioms on which software was founded. THIS is what should piss people off. It pisses me off in computer science and biology. Companies are standing on the shoulders of giants who didn't have the opportunities available to them, especially when most of the underlying infrastructure was built with public funds.
Patents are expensive, but there goal is to offer short-term incentives for development in order to release knowledge into the public domain. Sometimes we forget that patents are designed to release knowledge. However, I would favor a progressive patent law where software patents last only five years and drugs targeted at monogenic diseases that don't affect many patients (like huntington's disease and spinal muscular atrophy) last for forty.
That's just me though.
Unfortunately she's made enough compensation that she will never have to work again, ever. Imagine how the nature of some corporations would change if the top executives would also need to have jobs next year. Not that I'm too flaming a liberal, I want to amass my own pile of swag but the current terrible statistic is that America has the one if the highest disparities between high and average salaries in the world.