May 7, 2006
[ return to the news list ]
Nature (03/23/06) Vol. 440, No. 7083, P. 409
Muggleton, Stephen H.
Inexpensive data storage and increasingly efficient technologies have led to the use of automated data collection and processing techniques throughout the sciences as researchers deal with exponentially growing bodies of data. As climate-modeling and astronomical experiments continue to fill massive databases, scientists are becoming increasingly dependent on computational power to identify and analyze connections between different datasets, bringing capabilities previously thought only to be possible in theory into real experiments. Scientists are using machine-learning methods from computer science, such as neural nets and genetic algorithms, to mine data and produce new hypotheses automatically as they look to develop new drugs or mitigate the effects of climate change. Researchers are often hamstrung by incompatibilities between different models, though computer scientists are developing new formalisms that coalesce mathematical logic and probability calculus to form a kind of probabilistic logic. With experiments having already proven the ability of robotic scientists to conduct tests to distinguish between opposing hypotheses, a microfluidic robot scientist with active learning and autonomous experimentation capabilities could appear within 10 years. Laboratories on chips already exist, thanks to computer-directed microfluidics, and a similar scaling process could be applied to robot-scientist technology. One such application, a sort of programmable, chemical Turing machine, could perform a wide range of chemical operations, preparing and testing compounds automatically. The microfluidic Turing machine could also serve as a model for cellular metabolism simulations, or as the basis for an artificial cell that could be used in drug testing.
For the full article see http://www.nature.com/news/2006/060320/full/440409a.html
Posted by rab5 at 08:02 PM
[ return to the news list ]
