A method for automated de novo optimisation of functional transition-metal complexes

Transition metal complexes are vital components in functional compounds and materials ranging from lasers to industrial and biological catalysts. The search for and development of such complexes is still largely manual and with an almost complete lack of modern methods for in silico virtual screening and design. Here, we propose a method for automated de novo optimisation of transition metal complexes which is expected to facilitate the development of, for example, homogenous catalysts. Our overall meth od is based on genetic algorithms (GAs), which, in principle, are able to find global optima. However, GAs require large populations and many iterations to converge. This implies that, for the method to be practical, the computational cost of obtaining th e fitness measure of each individual molecule during the optimisation must be kept to a minimum. Thus, a hierarchy of cost-efficient computational methods, ranging from methods based on classical force-fields, via semi-empirical methods, to first-princi ples DFT methods, will be integrated with the GA optimizer. However, our main approach to speed up the GA optimisation involves the replacement of most of the expensive molecular-level calculations with a QSAR/QSPR model based on computationally inexpensi ve structure descriptors. The GA optimizer can thus be allowed to work with larger populations and over more generations than with a procedure involving the calculation of the "true" fitness of each individual molecule. The QSAR model is applied repeatedl y until a large portion of the structures being tested fall outside its confidence limits. Then, the fitness (response) is explicitly calculated by a molecular-level method, and the obtained "true" responses are used to refine and update the QSAR model. Finally, the proposed method also involves approaches to automation of the molecular descriptor and response calculations and to integration of the different software components of the overall method.