Chemo Synergy

Prediction of the lethality of combinations of chemo therapy drugs
Experience
Neural networks
Performance optimization
Cancer biology
Molecular biology
Cellular networks
Evolution of gene regulatory networks
Interdisciplinary science
Matrix based programming languages
Technologies
Lisa grid of Sara (Sara is the national supercomputer service)
Matlab
Summary
Cancer causes the most deaths in the Netherlands. Better therapies are needed, and any delay is payed in human lives. But it takes a long time to get approval for a new therapy. And for many cancers single drug therapies are not effective; when a drug targets some mechanism in a cancer cell, a backup mechanism often takes over. Well chosen combinations of existing drugs can hit both the target and the backup mechanism, and the drugs have already gotten approval.

ChemoSynergy attempts to predict the effectiveness of such combinations. It attempts to model the changes caused by one or more drugs, and how that leads to cell death. It is implemented as a recurrent neural network with a cleverly chosen kernel function, and a cleverly chosen topology. These kernel functions are capable of expressing the derivative of a concentration in a chemical reaction with stoichiometric constants of 1 (stoichiometric constants are the numbers you see in reaction formula, and most protein reactions in the cell have stoichiometric constants of 1). The topology is based on maps of protein pathways, and knowledge of common patterns in cellular chemical networks.

There are no results yet, as there are still some finishing touches needed. When it is complete it will be run on the Lisa cluster of Sara, the national super computer service.