Being able to predict the amount of medicine that, once taken orally, effectively reaches the blood circulation is one of the most pressing issues in drug development. Now, researchers in Portugal and Italy have turned to the computer to help them develop an evolutionary new approach that could predict which drugs are able to achieve their healing purpose and which ones can be excluded from further tests.
Sara Silva of Instituto de Engenharia de Sistemas e Computadores – Inverstigação e Desenvolvimento em Lisboa (INESC-ID), Portugal, and Leonardo Vanneschi of the University of Milano-Bicocca, Milan, Italy (now developing his research at the Instituto Superior de Estatística e Gestão de Informação, ISEGI, of the Nova University of Lisbon, Portugal), explain that medications given by mouth must survive a tortuous and hazardous route to their site of action in the body. First, they must pass from the mouth, down the oesophagus and be absorbed by the gut wall. They are then carried to the liver by the hepatic portal vein. Along the way they might become bound and deactivated by plasma proteins and once in the liver enzymes could attack them breaking them down. Only a fraction of the drug initially swallowed might exit the liver, enter the blood stream and circulate to where it is needed and so have a therapeutic effect.
Identifying which drug molecules will survive the journey is difficult. Many drugs that show promise in early design and testing fail later as they prove not to be absorbed well by the gut or otherwise fail to make the grade. Additionally, break down products formed in the liver might be toxic and so side-effects become apparent.
Silva and Vanneschi have extended the well-known computational technique of genetic programming to help them devise a more efficient way of predicting which molecules will work when given by mouth. Genetic programming involves feeding the properties of known orally active drug molecules into the computer and selecting and combining those features of the computer programs that are successful in predicting oral bioavailability correctly. Ineffective programs are deleted and the successful features are then used to build a next generation of programs. With each successive generation the evolved programs gets fitter, in the Darwinian sense, hence the notion of genetic programming.
Once a fit program has been evolved, it can then be used to test candidate molecules the oral bioavailability of which is unknown. The team has now found a way to avoid “bloat” in the production of successive program generations, an issue that has plagued scientists hoping to develop an efficient algorithm for drug testing.
“Bloat free Genetic Programming: application to human oral bioavailability prediction” in Int. J. Data Mining and Bioinformatics, 2012, 6, 585-601