Safer, faster, better – high-powered forecasts

"In this project, we have switched out the lab for data," says Ruben Buendia Lopez, researcher at the University of Borås.

He has been working with pharmaceutical company AstraZeneca in Mölndal since 2017 to apply new models. The models, developed at the university, are meant to result in more certain forecasts for the company's research and development activities.

"It is a research-oriented company that researches and develops pharmaceuticals; it's a long process but it requires a lot of resources," says Ruben Buendia Lopez.

Provides a better picture of probability

Integrated in a research group, he is working to identify molecules that can combat specific diseases, something that can ultimately result in new medications.

This work has previously been done in laboratories but this is now being complemented by data analysis and forecasts, which enable a better picture of likelihood when it comes to new solutions.

"There are great hopes that machine learning and AI can streamline the process. I am very pleased to be here at the company and work with the material directly; the data infrastructure is impressive," says Ruben Buendia Lopez.

Data is everywhere today. Well-used data can drive the development of a company and anticipate future solutions and innovations. However, forecasts can be uncertain, while at the same time underpinning risk analyses and economic strategies for industry.

Based on algorithms

The project "DASTARD" at the University of Borås was initiated to improve forecasts, mainly by determining how likely they are. Through so-called Venn Predictors, based on algorithms, researchers can make mathematically correct assessments of probabilities. Whereas forecasts were previously seen as qualified guesses – without tools to quantify uncertainty – the new model gives a statement of guaranteed probability.

The scientific framework had been prepared by researchers Vladimir Vovk, Alex Gammerman, and Glenn Shafer in the book "Algorithmic learning in a random world" (2005). At the university, the ideas have been adapted to the purpose of the project. The method has been tested at AstraZeneca, where researchers use computer simulation to predict the properties of pharmaceutical substances. And it looks promising, according to Ruben Buendia Lopez.

Shortens research time

"At the moment, it may take 10 years to develop a new medicine; if you can cut half a year, that means more resources that can benefit everyone. But research is always slow, now it's about getting the results published and accepted within the research community," he says.

Interest in this type of predictive data analysis is growing rapidly. When the first major conference in the field, COPA, was held in 2012, a total of six scientific articles were received. Last year, there were over 50.

The project "DASTARD,” and Ruben Buendia Lopez's time at AstraZeneca will conclude at year end 2018/2019. However, he sees this work as just the beginning of exploration of an area with huge potential, where more certain forecasts can make a big difference.

"Especially in fields where you cannot afford to make reckless mistakes – as medicine," he says.

DASTARD

(Data Analytics for Research and Development)

  • A three-year research project at the University of Borås, in cooperation with AstraZeneca and Scania, which expires at year end 2018/2019. A total of ten researchers from the university and the companies are involved.
  • The overall purpose of the project is to support research and development processes with machine learning and data analysis.
  • A large part of the work has been devoted to "prediction with confidence" – to supplement forecasts with probability measurements.
  • At the pharmaceutical manufacturer AstraZeneca, better data forecasts can support faster and better drug development through silico modelling (computer simulation).
  • With truck manufacturer Scania, better use of collected data can reduce fuel consumption, and provide positive environmental impacts, such as data driven driver coaching.

Text: Christian Naumanen

Photo: Anna Sigge

Translation: Eva Medin