Physicist Says Sharing Data Could Speed Scientific Discovery

The World Health Organization announced last Friday that data from experiments that made deadly bird flu more contagious will be published, despite U.S. concerns that the information could be used by terrorists.

That decision comes after an expert panel determined that the danger posed by naturally occurring flu viruses is so great it outweighs a possible terrorist threat. Sharing information freely, the panel said, could help scientists identify a possible pandemic.

“Most labs make no systematic effort to share data with other scientists.”

Physicist Michael Nielsen

Could Sharing Speed Discovery?

The free exchange of scientific information has long been a cause for physicist Michael Nielsen. In his book  "Re-Inventing Discovery: The New Era of Networked Science," (excerpted below) he makes the argument that new online tools can make science a more collaborative process, and that a more open process would dramatically speed up the rate of scientific discovery in fields from mathematics to astrophysics, to genetics and medicine.

Nielsen says security concerns, like those about bird flu research falling in the wrong hands, are one of the few reasons we should think twice about sharing knowledge. But he says the majority of basic scientific information poses no such threats and should be treated as open science.

"Most labs make no systematic effort to share data with other scientists," Nielsen wrote in the Wall Street Journal.

As one biologist told Nielsen, he has been "sitting on the genome" of an entire species, a treasure trove of information, for over a year without putting it online.

No Incentive To Share

Nielsen told Here & Now's Sacha Pfeiffer that scientists don't want to invest the time and effort it takes to share their data because there is no reward for it — jobs and research dollars go to those who publish more, not those who share more.

That's why, Nielsen argues, the model for scientific publishing hasn't changed since the 17th century, even though new online tools could make a more open and collaborative process possible, dramatically speeding the rate of scientific discovery.

Nielsen sees this as one of the biggest challenges facing modern science — big enough for him to leave the world of academia and become a public advocate.

When Amateurs Rival Professionals

It’s not just in astronomy that citizen science is useful. One of the big open problems in biology is to understand how the genetic code gives rise to an organism’s form. Of course, we’ve all heard many times that DNA is the “blueprint for life.” But even though the slogan is familiar—it is, after all, the fate of great slogans to become cliches—that doesn’t mean anyone yet understands in detail how DNA gives rise to life. Suppose biologists had never seen an elephant’s trunk. Could they look into an elephant’s DNA and somehow see the trunk there—that is, predict the trunk’s existence based solely on the sequence of base pairs in an elephant’s genetic code? Today, the answer to this question is no: how DNA determines an organism’s form is one of the mysteries of biology.

To help solve this mystery, a citizen science project called Foldit is recruiting online volunteers to play a computer game that challenges them to figure out how DNA gives rise to the molecules called proteins. That challenge may sound a far cry from deducing the existence of the elephant’s trunk—it is a far cry—but it’s a crucial step along the way, because proteins carry out many of the most important processes in our bodies. Aside from its intrinsic scientific interest, Foldit is also interesting as a demonstration of the great complexity of work that can be done by volunteers. In Galaxy Zoo, participants mostly carry out simple tasks, such as classifying a galaxy as spiral or elliptical. In Foldit, players are asked to tackle tasks that would challenge a biochemistry PhD. And, as we’ll see, the top Foldit players are doing those tasks extraordinarily well.

Before we discuss Foldit in detail, let’s talk a bit about proteins in general. Biologists are obsessed by proteins, and with good reason: they’re molecules that do everything from digesting our food to contracting our muscles. A good example of a protein is the hemoglobin molecule. Hemoglobin is one of the main components in our blood: it’s the molecule our bodies use to move oxygen from our lungs to the rest of our body. Another important class of proteins are the antibodies in our immune system. Each antibody has its own special shape that lets it lock on to viruses and other intruders in our body, tagging them for attack by our immune system.

At present we only partially understand how DNA gives rise to proteins such as hemoglobin. What we do know is that certain sections of our DNA are protein coding, meaning that they describe a specific protein. So, for example, there’s a protein-coding section for hemoglobin somewhere in your DNA. That region is a long string of DNA bases, which starts: CACTCTTCTGGT. . . . It turns out to be helpful to divide that string of bases into triplets, which are called codons: CAC TCT TCT GGT. . . .The way proteins are formed is that each codon in the protein-coding section of your DNA is transcribed into a corresponding molecule in the protein called an amino acid. So, for example, the first codon for hemoglobin, CAC, gets transcribed into an amino acid known as histidine. I won’t explain exactly what histidine is, or what it does—for us it doesn’t much matter. What matters is that everywhere the CAC codon appears in the DNA sequence for hemoglobin (or any other protein), it gets transcribed to histidine. In a similar way, the second codon, TCT, gets transcribed into the amino acid serine. And so on. The resulting protein is a chain containing all those amino acids—so hemoglobin is a chain containing histidine, serine, and so on.

Okay, so far, so good: DNA can be used as a recipe for generating proteins. Proteins, however, differ from DNA in that they each have their own special shape, unlike the completely regular structure of DNA. That shape is tremendously important. For example, as I mentioned before, the antibodies in our immune system are proteins, and the shape of an antibody determines which viruses it can lock onto. What’s going on is that as the information in the DNA is transcribed to form the amino acids in the protein, the protein “folds” into its shape. How this folding occurs is still only partially understood, but there are some basic rules of thumb that should give you the flavor of what’s going on. Some amino acids like to be near water—they’re called hydrophilic, from the Greek roots “hydro” and “philia,” for water and love, respectively. Since proteins inside a cell are surrounded by water, the protein will tend to fold so the hydrophilic amino acids sit on the outside, near the water. Histidine and serine are both examples of hydrophilic amino acids. By contrast, hydrophobic amino acids—amino acids that don’t like water—end up bundled up tight inside the protein. Sometimes these tendencies conflict: neighboring amino acids in the protein may be alternately hydrophobic and hydrophilic, with the result that the protein can end up folding into a very complex shape.

There’s an incredibly clever trick here that nature is using. The DNA is a completely regular arrangement of information, which makes it both easy to copy and relatively straightforward to transcribe into amino acids. But then competition between hydrophilia, hydrophobia, and other forces means that the protein can fold up to form complex shapes. By changing the DNA we can change the amino acids in the protein, which in turn causes the shape of the protein to change. What’s clever about this is that it takes us from the regularly arranged information in the DNA, which is easily copied, to the many possible shapes of the protein. A priori, shapes don’t seem so easy to copy. It’s as though you could trace over the blueprint for a house, and the traced version would then somehow spring into existence as a tiny model house. The DNA-protein connection is Nature’s way of making easy the seemingly impossible task of copying complex shapes.

But there’s a problem with this neat story. Just because we know the DNA sequence for a protein doesn’t mean we can easily predict what shape the protein has, or what the protein will do. In fact, today we have only a very incomplete understanding of how proteins fold. Complete structures—the exact shapes—are known for only 60,000 proteins, despite the fact that we know the DNA sequences for millions of proteins. Most of those complete structures have been found using a technique called X-ray diffraction—basically, shining X-rays at a protein and figuring out its shape by looking carefully at the X-ray shadow it casts. It’s slow, expensive, painstaking work, and the techniques are only gradually getting better. What we’d really like is a fast and reliable way to predict the shape from the genetic description. If we could do that, cutting out the slow and expensive X-ray diffraction step, we’d go from knowing the shape of 60,000 proteins to knowing the shape of millions. Even more significantly, such a method would be a tremendously powerful tool for helping us design proteins with desired shapes. This would, for instance, help us engineer new antibodies to fight disease.

To solve the protein folding problem, biochemists have turned to computers in an attempt to predict protein shape from the genetic description. To make their predictions they use the idea that a protein will eventually fold into its lowest energy shape, much as a ball will roll to the bottom of a valley between two hills. All that’s needed is good method for finding the lowest energy shape of a protein. This sounds promising, but in practice it’s hard to search through all the possible shapes, looking for the shape with the lowest energy. The difficulty is the number of different shapes a protein can potentially fold into. Proteins typically have hundreds or even thousands of amino acids. To determine the structure means knowing the exact position and orientation of every single one of those amino acids. With so many amino acids involved, the number of possible shapes is astronomical, far too many to search through even on a very powerful computer. Enormous effort has been put into finding clever algorithms that can be used to restrict the number of configurations that must be examined, and the algorithms are getting pretty good. But there’s still a long way to go before we can use computers to reliably predict protein shapes.

In 2007, a biochemist named David Baker and a computer graphics researcher named Zoran Popovic, both from the University of Washington, in Seattle, had an idea for a better way of solving the problem. Baker and Popovic’s idea was to create a computer game that shows a protein to the player, and gives them controls to change the shape, rotating the protein, moving amino acids around, and so on. Some of the controls built into the game are similar to the tools used by professional biochemists. The lower the energy of the shape the player comes up with, the higher their score, and so the highest scoring shapes are good candidates for the real shape of the protein. Baker and Popovic hoped that this might be a better approach to protein folding than the conventional approaches, combining state-of-the-art computational techniques with computer gamers’ persistence and abilities at pattern matching and 3-D problem solving.

I was skeptical when I first heard about Foldit. It sounded like the dull educational computer games I saw in school when I was growing up in the 1980s. But I downloaded the game, and spent hours playing it over several days. At that point, the excuse “I’m doing research for my book” was rapidly becoming a euphemism for “this is a great way to procrastinate on writing my book,” and I forced myself to stop. So far, more than 75,000 people have signed up. People play the game because it’s good. It has the compelling, addictive quality all good computer games have: a task that’s challenging but not impossible, instant feedback on how well you’re doing, and the sense that you’re always just one step away from improvement. It’s the same addictive quality we saw earlier in the MathWorks competition, and which is also felt by many participants in Galaxy Zoo. Furthermore, like Galaxy Zoo, Foldit is deeply meaningful to many of the players. Einstein once explained why he was more interested in science than politics by saying, “Equations are more important to me, because politics is for the present, but an equation is something for eternity.” Each time you classify a galaxy or find a better way to fold a protein, you’re making a small but real contribution to human knowledge. For many participants, Foldit and Galaxy Zoo aren’t guilty pleasures, like playing World of Warcraft or other online games. Instead, they’re a way of contributing something important to society. One of the top Foldit players, Aotearoa, describes it as “the most challenging, exciting, stimulating, intense, addictive game I have ever played,” and comments that it provides a way for people to “offer something proactive to solving some of the worlds/societies most complicated puzzles, rather than waste time playing a ‘game’ that does not provide the same ‘rewards’ as folding protein does, this way!”

In addition to the individual motivation to play, Foldit also encourages collective problem solving by the players. There is an online discussion forum and a wiki, where players share news and discuss their strategies for protein folding. The game incorporates a simple programming language that players can use to create scripts—short programs—that automate game tasks. A typical script might implement a strategy for improving a fold, or identify which part of the protein’s current shape is in most need of improvement. Hundreds of such scripts have been publicly shared—an open source approach to protein folding. Many of the players work in groups, sharing their insights about the best ways of folding. All this work is greatly informed by the game score, which, as in the MathWorks competition, focuses participants’ attention where it will be most useful: when one of the high-scoring players shares a strategy tip or a script, other players pay attention. The players themselves are wildly varied, ranging from a self-described “educated redneck” from Dallas, Texas, to a theater historian from South Dakota, to a grandmother of three with a high-school education.

Just how good are the Foldit players at folding proteins? Every two years since 1994, there’s been a worldwide competition of biochemists using computers to predict protein structures. The competition, called CASP—Critical Assessment of Techniques for Protein Structure Prediction—is very important to the scientists who work on protein structure prediction. Before the competition starts, the CASP organizers approach some of the facilities that determine protein structure using the traditional approach of X-ray diffraction, and ask them what protein structures they expect to complete in the next couple of months. They then use those proteins as puzzles in CASP. Starting with the sequence of amino acids making up the protein, the CASP competitors are asked to predict the structure. At the end of the competition, teams are ranked on how close they come to the actual structure.

Foldit players competed in both the CASP 2008 and 2010 competitions. They performed extremely well, finishing near or at the top on many of the CASP challenges. Foldit developer Zoran Popovic summed up the results of the 2008 competition by saying that “foldit players are on a par, but not better than protein folding experts at trying to solve the same problem with all tools available to them. It also appears that foldit outperformed all fully automated server submissions.” Thus, a team of amateurs can be competitive with some of the world’s top biochemists, equipped with state-of-the-art computers. Popovic told me that his “ultimate goal is to show that experts are unequivocally inferior to the general population with this problem . . . a biochemistry PhD does not self-select for spatial reasoning. Structure prediction is all about 3d problem solving and very little about biochemistry.” Indeed, even specialists in protein-structure prediction usually spend only a small fraction of their time working directly on predicting protein structures. And while they have expertise that the amateurs don’t, much of that knowledge is incarnate in the mechanics of the game. That levels the playing field enough that the remaining disparity in expertise can be overcome by the greater time commitment of the Foldit players. It’s a symbiosis: the professionals develop the systematic understanding that underlies the mechanics of the game, and the amateurs then supply the dedicated artistry required to take best advantage of that systematic understanding.

Excerpted from REINVENTING DISCOVERY by Michael Nielsen.
Copyright (c) 2012 by Princeton University Press. No part of this text may be distributed, posted, or reproduced in any form by digital or mechanical means without prior written permission of the publisher.