Saturday, May 23, 2009

Robot Scientist; Ross King, Adam, & "Automation of Science"

Robot achieves scientific first
"A laboratory robot called Adam has been hailed as the first machine in history to have discovered new scientific knowledge independently of its human creators.

Adam formed a hypothesis on the genetics of bakers’ yeast and carried out experiments to test its predictions, without intervention from its makers at Aberystwyth University.

The result was a series of “simple but useful” discoveries, confirmed by human scientists, about the gene coding for yeast enzymes. The research is published in the journal Science.

Professor Ross King, the chief creator of Adam, said robots would not supplant human researchers but make their work more productive and interesting."

The Paper, "Automation of Science":
"The Automation of Science
Ross D. King,1* Jem Rowland,1 Stephen G. Oliver,2 Michael Young,3 Wayne Aubrey,1 Emma Byrne,1 Maria Liakata,1 Magdalena Markham,1 Pinar Pir,2 Larisa N. Soldatova,1 Andrew Sparkes,1 Kenneth E. Whelan,1 Amanda Clare1

The basis of science is the hypothetico-deductive method and the recording of experiments in sufficient detail to enable reproducibility. We report the development of Robot Scientist "Adam," which advances the automation of both. Adam has autonomously generated functional genomics hypotheses about the yeast Saccharomyces cerevisiae and experimentally tested these hypotheses by using laboratory automation. We have confirmed Adam's conclusions through manual experiments. To describe Adam's research, we have developed an ontology and logical language. The resulting formalization involves over 10,000 different research units in a nested treelike structure, 10 levels deep, that relates the 6.6 million biomass measurements to their logical description. This formalization describes how a machine contributed to scientific knowledge."

Robot Scientist Website:

"The Robot Scientist is perhaps the first physical implementation of the task of Scientific Discovery in a microbiology laboratory. It represents the merging of increasingly automated and remotely controllable laboratory equipment and knowledge discovery techniques from Artificial Intelligence.
The robot in our lab

Automation of laboratory equipment (the "Robot" of Robot Scientist) has revolutionised laboratory practice by removing the "drudgery" of constructing many wet lab experiments by hand, allowing an increase in both the scope and scale of potential experiments. Most lab robots only require a simple description of the various chemical/ biological entities to be used in the experiments, along with their required volumes and where these entities are stored. Automation has also given rise to significantly increased productivity and a concomitant increase in the production of results and data requiring interpretation, giving rise to an "interpretation bottleneck" where the process of understanding the results is lagging behind the production of results.

The research fields of Computational Scientific Discovery and Bioinformatics have emerged in part as a response to this bottleneck. Both disciplines use computational approaches from Statistics and Machine Learning to provide an "automated understanding" of the experimental results.

It has become typical practice in Bioinformatics to separate the data collection or experimentation process and the understanding process, where large numbers of experiments are conducted and then specially designed data mining tools are used to identify correlations in the data that might represent hitherto undiscovered scientific knowledge.

This knowledge will initially correspond to the goals of the scientific task, but increasingly the internet repositories that are often constructed to store the data have become the focus of less directed scientific study, where "hidden" knowledge not originally anticipated by the goals of the scientific task may be found. However, this "scrapyard" approach is partly a result of overexperimentation where many unnecessary experiments were conducted along with the potentially informative ones.
PC and Sciclone

The Robot Scientist makes use of an iterative approach to experimentation, where knowledge aquired from a previous iteration is used to guide the next experimentation step. This is a process known as Active Learning, where the learner can plan its own agenda, i.e. decide how best to improve its knowledge base and how to go about acquiring this information. The Robot Scientist uses the laboratory robot to execute the experiment(s) selected as most informative; has a plate reader to analyse the experiments, generating data corresponding to the scientific observations; uses abductive logic programming to generate valid hypotheses that explain the observations; and uses these hypotheses to determine the next most informative experiment. At the beginning of any investigation, the Robot Scientist has not discovered any information, therefore all possible hypotheses are equally valid. As the directed discovery process continues, each new observation (or experiment/interpretation cycle) will invalidate some of the hypotheses, thereby excluding incorrect discoveries. The experiment selection process aims to choose the experiment most likely to refute the most hypotheses. This iterative process allows irrelevant experiments to be avoided, potentially saving both laboratory time and the cost of using unnecessary reagents and biological materials."

Ross King CV:

No comments:

"Not by strength, by guile."