|
 Sven Hellberg, head of of computer-assisted drug development at the department of chemistry, and Ulf Norinder, senior researcher within computational chemistry at AstraZeneca’s research facilities in Södertälje, Sweden.
Professor Ulf Norinder, AstraZeneca: ”With Compumine’s RDS we raise the quality of our research.” An advanced data mining system is helping the research divisions at AstraZeneca in Södertälje save time, capacity, and money, while raising the quality of their research findings at the same time.
Download this article as PDF
As Professor Ulf Norinder, senior researcher at the AstraZeneca Department of Chemistry, says, "RDS allows us to focus, at a very early stage, on the important characteristics of the substances we are studying." But RDS, or Rule Discovery System, is in no way strictly tailored to the pharmaceutical industry. It can be used to find hidden connections in all kinds of databases.
Data mining can be defined as a system of using automatic methods to obtain hidden information from large amounts of data. RDS, Rule Discovery System, is one such, very advanced, method. The software prototype was developed by Henrik Boström, Lars Asker, and Per Lidén at the information technology company Compumine, which they founded in 2002. RDS has since been further enhanced in collaboration with AstraZeneca, primarily with the help of Professor Ulf Norinder. "Our business concept,” explains Compumine CEO Per Lidén, ”is based on intelligent use of the capacity of computers to process large quantities of information, thereby creating increased value for our customers.”
Through so-called multivariate data analysis, RDS identifies rules that explain complicated connections — ones that are often difficult or impossible to detect using other methods. At AstraZeneca, RDS is used to find models that explain the findings from a large quantity of previously performed and documented trials. The models can then be used, for instance, as a platform for making decisions about the next step of the research, sometimes even eliminating the need for additional laboratory experiments.
Many application areas To be sure, RDS has been further enhanced in collaboration with AstraZeneca, but the software can be used to detect hidden relations and rules in all kinds of large quantities of information — including customer databases. The background behind RDS is to be found in Henrik Boström’s research at the Royal Institute of Technology and at Stockholm University.
CEO Per Lidén explains: "My interest was in finding alternative ways of constructing mathematical computation models, which is how I came to work with Henrik. In August, 2002, we started Compumine and began developing Rule Discovery System." Before co-founding Compumine, Henrik Boström had been in contact with Ulf Norinder at AstraZeneca and now the contact was renewed. "From our point of view,” Prof. Norinder says, ”RDS was extremely interesting.”
Thousands of substances tested “Over the years of working to develop new medical products, we’ve tested tens of thousands of substances. Any system that could help us find hidden patterns and relations in this material and that could organize them into rules—for example, for assessing such a basic parameter as solubility in water—without having to test them in practice would facilitate our research considerably. Similar systems have been available in the past, but nothing really worked to our satisfaction.”
The unique feature of RDS is that it makes it easy to create models of high precision through consensus or ensemble modeling, which means that the program can generate many different models that work with and reinforce each other. "What we need,” says Ulf Norinder, ”are results that are maximally stabile and consistent, since you lose confidence in the system otherwise. The system has to be a reliable instrument in our work.” It was decided to make a test run of RDS at the Department of Chemistry at AstraZeneca, and RDS stood the test. The models of expert rules that were constructed with the help of RDS to assess substance solubility and permeability were so stabile and yielded such good results that the program and models were entered into the division’s prediction server. This means they are now available to all researchers at the Department of Chemistry.
The program was a success and brought with it a clear break in trend of the department’s activities. "The first rule models we built,” says Associate Professor Sven Hellberg, head of computer-assisted drug development at the AstraZeneca Department of Chemistry, ”dealt with water solubility, since that’s the first thing you test for in developing a new drug and since solubility, or rather, lack of it, is a major problem for us. A substance’s solubility is an essential condition for uptake in the body.”
Improved test results From 2002 to 2003, no particular improvements were seen in tests of the relationship between readily-soluble/low-soluble compounds. In mid-2004, when RDS had begun to be used more generally at the Department of Chemistry, there was a notable reduction in the number of low-soluble compounds that went so far as to testing.
"Now we focus on the more promising substances,” Prof. Norinder says. ”We reach conclusive findings faster — in short, we save time and capacity and raise the quality of our findings at the same time. This is enormously significant. You could say that RDS helps us concentrate on the essential characteristics of a new drug. We can identify the dead ends and side tracks at a much earlier stage.” The Department of Chemistry at AstraZeneca has gone on to create working models for solubility with the help of RDS.
Takes RDS one step further ”We’ve also begun to take this one step further,” says Ulf Norinder, ”doing the same thing for other important characteristics, like permeability. This will save us both time and money." According to Norinder, the pharmaceutical industry has been slow in designing simulation models, partly because they are so complex.
"But this kind of data modeling or in silico work will have significant impact on the development of medical products,” he says, ”since it aids researchers in predicting the characteristics of substances much faster and at a lower cost. Rule Discovery System is a tool that is easy to use and has the potential of solving many classification problems in the pharmaceutical industry. Besides generating a model with a higher capacity for prognosis accuracy than previously tested methods,” Prof. Norinder concludes, ”RDS makes it possible to study the models and understand the underlying reason for the classification.”
|
Rule Discovery System™
