The electronic structure of molecular systems containing transition metal atoms is traditionally studied using methods based on density functional theory (DFT). Although such an approach has been quite successful, the treatment of large systems, be they transition metal complexes, bioinorganic molecules or the solid state, is still extremely computationally demanding at this level, and may not be practical for many systems of interest. In this paper we describe how semi-empirical MO methods can be used to overcome these computational bottlenecks, yet still provide predictions of the necessary accuracy. We describe strategies to achieve this by focussing on: (i) obtaining appropriate parameters for transition metal atoms via a genetic algorithm, to be used within a parallelised implementation of neglect of differential diatomic overlap (NDDO) methods, and (ii) the use of multilevel treatments which involve DFT and semi-empirical methods to describe different regions of the molecule. Here we show by reference to histidine and porphyrin complexes, the importance of a correct partitioning of the organic substrate. We illustrate the potential of such a dual level approach by reporting preliminary results showing the catalytic role of the enzyme, dimethyl sulfoxide reductase.