文摘
Rational drug design is a process of finding new medications based on the knowledge of the biological target. Candidate molecules can fail to become drugs for various reasons such as poor absorption or toxicity. A successful drug must therefore satisfy multiple,sometimes competing objectives. In this dissertation we developed a computational tool for automated design of novel small organic molecules with potential to exhibit high activity against a desired drug target. A parallel genetic multi-objective algorithm acts on molecules to mutate and recombine molecular fragments. A novel neural network based scoring function for estimating binding affinity to a drug target guides the search in the molecular space towards molecules optimized to bind well to the target. A second objective function calculates similarity of the prospective ligand to known drugs to guide the search towards molecules exhibiting other drug-like properties. The method was tested on several drug targets and a comparative study between our method and another common technique used in computer-aided drug design (virtual screening of an existing chemical database) was conducted. Our findings demonstrate that generally,the set of molecules generated by our method is better optimized for the imposed objectives than the set obtained by virtual screening.