Machine learning of pharmacodynamic effects

Beskrivelse

Abstract

MACHINE LEARNING OF PHARMACODYNAMIC EFFECTS
HARRIE C.M. BOONEN, JIAOWEI TANG
Department of Drug Design and Pharmacology, University of Copenhagen, Universitetsparken 2, 2100 Copenhagen

Monitoring of physiological surrogate end-points, for example in drug development, generates dynamic time domain data reflecting the effect of a certain intervention with a drug on the state of the biological system. Conventional pharmacological data analysis often reduces the information in this data by extracting specific data points, such as steady state or maximum response, thereby discarding potentially useful information. We developed a genetic fuzzy system algorithm that is capable of learning all information in time domain physiological data, and on the basis of this, can predict these responses without any mechanistic knowledge or assumptions about the biological system. Data of isometric force development of isolated small arteries were used as a framework for developing and optimizing a genetic fuzzy system (GFS). Briefly, genetic fuzzy systems are a computer search algorithm that can learn and predict dynamic patterns in data. We improved the performance of the genetic fuzzy system by implementing several computational strategies. The results showed that optimized fuzzy systems are able to predict contractile reactivity of arteries extremely accurate. In addition, optimized fuzzy systems identified significant differences in responses of arteries between groups that were undetectable using conventional analysis. We conclude that optimized fuzzy systems may be used in clustering or classification tasks and aid in objective prediction of drug effects and identification of safety issues.

Periode	15 jan. 2014
Begivenhedstitel	6th annual meeting of the : Protein Therapeutics
Begivenhedstype	Konference
Konferencenummer	6
Placering	Odense, DanmarkVis på kort