Robot biologist solves complex problem from scratch

Oct 13, 2011
This is a cartoon of robot biologist. Credit: Michael Smeltzer, Vanderbilt University

First it was chess. Then it was Jeopardy. Now computers are at it again, but this time they are trying to automate the scientific process itself.

An interdisciplinary team of scientists at Vanderbilt University, Cornell University and CFD Research Corporation, Inc., has taken a major step toward this goal by demonstrating that a computer can analyze raw experimental data from a biological system and derive the basic mathematical equations that describe the way the system operates. According to the researchers, it is one of the most complex scientific modeling problems that a computer has solved completely from scratch.

The paper that describes this accomplishment is published in the October issue of the journal and is currently available online.

The work was a collaboration between John P. Wikswo, the Gordon A. Cain University Professor at Vanderbilt, Michael Schmidt and Hod Lipson at the Creative Machines Lab at Cornell University and Jerry Jenkins and Ravishankar Vallabhajosyula at CFDRC in Huntsville, Ala.

The "brains" of the system, which Wikswo has christened the Automated Biology Explorer (ABE), is a unique piece of software called Eureqa developed at Cornell and released in 2009. Schmidt and Lipson originally created Eureqa to design robots without going through the normal trial and error stage that is both slow and expensive. After it succeeded, they realized it could also be applied to solving science problems.

One of Eureqa's initial achievements was identifying the basic by analyzing the motion of a double pendulum. What took Sir Isaac Newton years to discover, Eureqa did in a few hours when running on a personal computer.

In 2006, Wikswo heard Lipson lecture about his research. "I had a 'eureka moment' of my own when I realized the system Hod had developed could be used to solve biological problems and even control them," Wikswo said. So he started talking to Lipson immediately after the lecture and they began a collaboration to adapt Eureqa to analyze biological problems.

"Biology is the area where the gap between theory and data is growing the most rapidly," said Lipson. "So it is the area in greatest need of automation."

Software passes test

The biological system that the researchers used to test ABE is glycolysis, the primary process that produces energy in a living cell. Specifically, they focused on the manner in which yeast cells control fluctuations in the chemical compounds produced by the process.

The researchers chose this specific system, called glycolytic oscillations, to perform a virtual test of the software because it is one of the most extensively studied biological control systems. Jenkins and Vallabhajosyula used one of the process' detailed mathematical models to generate a data set corresponding to the measurements a scientist would make under various conditions. To increase the realism of the test, the researchers salted the data with a 10 percent random error. When they fed the data into Eureqa, it derived a series of equations that were nearly identical to the known equations.

"What's really amazing is that it produced these equations a priori," said Vallabhajosyula. "The only thing the software knew in advance was addition, subtraction, multiplication and division."

Beyond Adam

The ability to generate mathematical equations from scratch is what sets ABE apart from Adam, the robot scientist developed by Ross King and his colleagues at the University of Wales at Aberystwyth. Adam runs yeast genetics experiments and made international headlines two years ago by making a novel scientific discovery without direct human input. King fed Adam with a model of yeast metabolism and a database of genes and proteins involved in metabolism in other species. He also linked the computer to a remote-controlled genetics laboratory. This allowed the computer to generate hypotheses, then design and conduct actual experiments to test them.

"It's a classic paper," Wikswo said.

In order to give ABE the ability to run experiments like Adam, Wikswo's group is currently developing "laboratory-on-a-chip" technology that can be controlled by Eureqa. This will allow ABE to design and perform a wide variety of basic biology experiments. Their initial effort is focused on developing a microfluidics device that can test cell metabolism.

"Generally, the way that scientists design experiments is to vary one factor at a time while keeping the other factors constant, but, in many cases, the most effective way to test a may be to tweak a large number of different factors at the same time and see what happens. ABE will let us do that," Wikswo said.

Explore further: MU researchers develop more accurate Twitter analysis tools

Related Stories

Eureqa, the robot scientist (w/ Video)

Dec 07, 2009

(PhysOrg.com) -- A new program, Eureqa, takes raw data and formulates scientific laws to suit, and it is available by free download to all scientists.

New file format will help 3-D printing progress

Jul 22, 2011

(PhysOrg.com) -- A newly approved standard for 3-D printing file interchange will greatly enhance 3-D printing capabilities, says Cornell's Hod Lipson, who led the development of the standard.

Recommended for you

Chameleon: Cloud computing for computer science

Aug 26, 2014

Cloud computing has changed the way we work, the way we communicate online, even the way we relax at night with a movie. But even as "the cloud" starts to cross over into popular parlance, the full potential ...

User comments : 25

Adjust slider to filter visible comments by rank

Display comments: newest first

georgert
5 / 5 (10) Oct 13, 2011
Hod Lipson is an amazing scientist. His studies and demonstrations on heuristic robotics and AI are mind blowing. This YouTube video that outlines his work is an hour well worth it:
http://www.youtub...6sLl6dVg
Nanobanano
2 / 5 (7) Oct 13, 2011
Interesting.

Basicly, what he is doing with all of these nested models of models is a generalized brute force "guess and check" algorithm.

This is actually all humans really do in many learning environments, as far as "self taught" methods.

Formal education and being taught formulae by a teacher has masked much of that process of "guess and check".

This algorithm should be very good at Chess automatically, a few minor tweaks, and it would be probably almost as good as an ideal chess engine.
antonima
not rated yet Oct 13, 2011
Thats cool, guess and check has long been overlooked by high-brow science. But, its by far the best approach in many disciplines, especially relatively unstudied ones such as biology.

I wonder how far this kind of computerized scientist will take research, and whether it is more or less likely to think outside of the box than a real person.
Cube
5 / 5 (1) Oct 13, 2011
just downloaded Lipson's eureqa software, its so similar to something i was thinking about for a while now, but actions speak louder than words and he made the program so hats off to his work!

as an engineer i can only drool at the possibilities...
PhotonX
5 / 5 (1) Oct 13, 2011

This algorithm should be very good at Chess automatically, a few minor tweaks, and it would be probably almost as good as an ideal chess engine.


@Nanobonano: Interesting thought. Wouldn't it be interesting to enter this guy in some bonafide chess tourmanemts, to see how long it would take to move up the ratings and play up to snuff. Far more comprable to the way us meatbags have to do it.
Nanobanano
1.2 / 5 (6) Oct 13, 2011
Well, existing chess engines have rules which attempt to define how strong a position is, and rate each position. this allows the engine to select the best possible move to obtain the best possible position, by looking n moves ahead and producing all possible trees of moves.

In theory, if your ranking system of the algorithm is not flawed, then this should allow a chess engine to play an ideal game, that is, it would always make the best possible move every turn.

But the problem with an ideal game is that a human could memorize the ideal game, and then play white and beat the black computer.

So a computer cannot play an ideal game, because if it does play an ideal game, the human will memorize the entire game and always win when they play white.

I'll explain how I really noticed this below, as this doesn't allow enough characters.
Nanobanano
1.7 / 5 (6) Oct 13, 2011
Let's say it's a hypothetical white computer.

If the computer always made the best possible move, and let's say the "real" best possible move, not just what somebody or some algorithm computes...

If it always made the best possible move, then the first move as white should always be the same, UNLESS there is a tie for best possible move. If there is a tie, it should randomly pick a move.

Well, I noticed that chess engines do not always make the exact same first move as white, even on max difficulty setting.

This implies that:
1, The engine is not ideal, and at some point is making an approximation involving a random seed.
2, the engine legitimately thinks two or more moves are equally powerful and randomly picks one.
3, the engine inserts a random error.

Since 3 would not make sense on maximum difficulty setting (if you are designing something to beat Kasparov, you'd make what you believe to be the best possible engine.)

It would seem 1 or 2 must be the case.
CHollman82
1 / 5 (2) Oct 13, 2011
I think the point you're trying to make, in 10,000 words or less, is that there is no "best" move, because that would imply knowledge of the probability of the opponent to play specific moves... unless the AI can read the opponents mind or see the future the true "best" move is always going to be unknown.

I could imagine an algorithm designed to beat master chess players losing to an amateur, because the amateur is so unpredictably irrational.
Nanobanano
3.3 / 5 (6) Oct 13, 2011
I think the point you're trying to make, in 10,000 words or less, is that there is no "best" move, because that would imply knowledge of the probability of the opponent to play specific moves... unless the AI can read the opponents mind or see the future the true "best" move is always going to be unknown.

I could imagine an algorithm designed to beat master chess players losing to an amateur, because the amateur is so unpredictably irrational.


Well, the old algorithms struggled if you spotted the computer a pawn or bishop, it would actually play poorly, because it didn't understand the consequences, so you are half right.

But the new engines are essentially a optimized tree that attempts to find all possible moves, except if they rate a certain position as "inferior" they kill that "branch" in the tree and quit searching there.

But players play the game much like the computer. If you're only thinking one move ahead, you're really not even playing yet...
Nanobanano
2 / 5 (4) Oct 13, 2011
But my concentration has come down a lot, but I used to think all my moves ahead, and have my first 10 or 11 moves planned out.

I can't do that any more, and I tend to miss stuff even just 2 or 3 moves out now.

What you are supposed to do when you play chess is in your head, you play BOTH sides. Consider your ideal move, and then consider what the opponent's ideal move would be in response, and then what would your ideal move be in response, etc.

Like I said, I used to run that out 10 or 11 moves in my head, can't any more, and not saying I was perfect then, because I wasn't.

Have you ever seen "searching for Bobby Fischer"? the movie with the boy who plays chess, and you see him playing against himself as he plays both white and black? Well, the irony is every game should be like that anyway.

In the movie, he got to where he played without the pieces on the board, and just memorized where they are.

Well, that's what the computer does every game anyway...
Deesky
5 / 5 (6) Oct 13, 2011
Thats cool, guess and check has long been overlooked by high-brow science.

What's 'high brow' science? It's folly to suggest science has ignored this approach. Genetic or evolutionary algorithms use similar methods to cover large search spaces by trial and error.

But, its by far the best approach in many disciplines, especially relatively unstudied ones such as biology.

You think biology is relatively unstudied???

whether it is more or less likely to think outside of the box than a real person.

It's unlikely to do any kind of thinking, just to try out many variations and try to draw relations between them. It doesn't require any hunches, ideas or insights that a human might bring to the table. Its power lies in automation and relentless data fitting/pattern matching that a real person would find tedious.
Deesky
4.6 / 5 (5) Oct 13, 2011
This algorithm should be very good at Chess automatically, a few minor tweaks, and it would be probably almost as good as an ideal chess engine

Why would this make it good at chess? At best, it would be able to deduce the laws (rules) of chess, but there are no 'laws' of winning.
georgert
5 / 5 (1) Oct 13, 2011
...I wonder how far this kind of computerized scientist will take research, and whether it is more or less likely to think outside of the box than a real person.


In the YouTube video cited in the first post, Lipson is asked about AI and the "Singularity." His most germane point was that while programs such as Eurequa can extrapolate heretofore unknown mathematical equations based on large data sets, the problem is in some cases the equations accurately represent the data, but their meaning is unknown. He said he is working on a deeper computer system that tries to explain through analogy what the math represents. But, Lipson said, perhaps it could be similar to someone trying to explain Mozart to a dog. How's that for out of the box? Again, strong recommendation to watch the video.
Nanobanano
1 / 5 (1) Oct 13, 2011
anyway, Josh Waitzkin.

I'm just terrible with names.

Point is, that's the way everything works (until someone discovers a general rule,) we just don't realize it, I think...
Nanobanano
1 / 5 (1) Oct 13, 2011
Why would this make it good at chess? At best, it would be able to deduce the laws (rules) of chess, but there are no 'laws' of winning.


Yes, yes there are laws of winning.

Else how do you think a computer chess engine calculates the "best" move?

It uses a positional algorithm to rate each move in a family of moves on trees, discarding moves that are rated weak, and keeping moves that are rated strong.

It continues to "predict" the next best move by recursively thinking "n" moves ahead, and then re-rating each position over and over, until it finds the best possible position "n" moves out. Then it makes the best possible move to obtain that position.

What many fail to realize is Chess is highly deterministic.

Hopefully, you're not just pushing pieces around on a board.

Hopefully, you are trying to predict what your opponents next "n" moves will be, and trying to make the right move based on that prediction.
Nanobanano
1 / 5 (1) Oct 13, 2011
For example, you can rate a position based on how many squares are under attack by your pieces, and how many enemy pieces are under attack by your pieces, and vice versa.

Intuitively, if your pieces are able to attack more squares than your opponent's pieces, then you are probably in stronger position, but not always.

If your pieces are able to attack more opponent pieces, then you are also probably in a stronger position, but not always.

If you control more pieces than your opponent, you are probably in a stronger position, but not always.

If your opponent is in check, you are probably in a strong positon, but not always.

If a move leads to a guaranteed mate in 3 more moves, you should take it, obviously...

And so on.

The idealized computer chess algorithms use a system of rating each move based on all these factors and more...
Deesky
5 / 5 (2) Oct 13, 2011
Yes, yes there are laws of winning.

Else how do you think a computer chess engine calculates the "best" move?

Purpose build chess engines use a heuristic approach which assigns weights to various moves at any point during a game. Then, based on the opponent's actual move, they're recomputed, etc.

These interim steps do not constitute laws, they are just incremental rules of thumb.
Nanobanano
1 / 5 (1) Oct 14, 2011
Do you realize I made a distinction above, between a hypothetical "ideal" move, and the best move a human or known computer algorithm would make? I clearly pointed out that the best known algorithm might not find the "ideal" move.

More precisely, it actually will find it, it just might not recognize it as such, and therefore rate it weaker than it should be rated, due to biases in the heuristics.

But in theory, it should be possible to construct an algorithm which always finds the "true" best move, recognizes it as such, and then makes it. When found, this algorithm will always play equal or better than any human or any other algorithm.
Deesky
5 / 5 (2) Oct 14, 2011
But in theory, it should be possible to construct an algorithm which always finds the "true" best move, recognizes it as such, and then makes it.

There is no 'true' best move in the totality of the game, only strong moves based on some criteria at any one time.
antialias_physorg
5 / 5 (5) Oct 14, 2011
Thats cool, guess and check has long been overlooked by high-brow science.

I'm pretty certain you have never worked in any scientific capacity. How do you think science works? Do we miraculously know the answer before we start checking? No. Hypothesis first, then test, then revision (or start from scratch if it fails miserably). Do this over many, many times and when then you (may) get something that works well enough to publish.

And what exactly is 'high brow' science? (and what would be 'low brow' science?)

As for the article. Impressive. This should be incredibly useful in the future.

One drawback I can see, though, is that it will find different unrelated formulae for related problems. What we are actually looking for are the overarching/underlying mechanisms. We may end up with a bag full of equations which work, but which aren't integrateable into an overarching framework. Good enough for industrial applications, but unsatisfying from a knowledge POV.
knikiy
not rated yet Oct 14, 2011
T.O.T.E.
TheGhostofOtto1923
2.3 / 5 (3) Oct 14, 2011
I suppose computers will get to a point where they can analyze the form and function of an animal and recreate the genetic code needed to produce it. This could come in handy for recreating extinct animals?

With a large enough database on genes and their expressions, rebuilding a T Rex might just be an extrapolation.
MRBlizzard
3 / 5 (2) Oct 15, 2011
"Biology is the area where the gap between theory and data ..."
Psychology is also in need of good theory. This might offer a chance of objective, non-human evaluation of psychological data. We might need ABE to translate the theory into human terms, or create newly defined words.
Come to think of it, some of the neural network analysis systems work fine, but don't lend themselves to human insight. ABE might be able to help us in that.
Finally, this should be adaptable to programming, code generation, even efficient code.
CHollman82
3 / 5 (6) Oct 17, 2011
Psychology will be antiquated by neurology as alchemy has been antiquated by chemistry and physics.

Psychology seeks to examine something that isn't real, the human "mind"... the human BRAIN is real, and neurology studies it, based on chemistry. That is where all knowledge about the human condition will be found.
Skultch
not rated yet Oct 19, 2011
This kaasinees character is starting to annoy me lately. He downvotes perfectly good points like the one CH just made, yet doesn't even comment. wtf, dude? I just want to know where you are coming from. Do you have some kind of superhuman perspective we can't understand? Why downvote at all if you aren't going to give your opinion on the subject?

btw, I know people rarely use it like this (unless it's spam), but it's supposed reflect RELEVANCY, not whether or not you agree with it.