(175d) An Adjacent Matrix Method for the Topological Description of a Molecular Structure in Drug Development | AIChE

(175d) An Adjacent Matrix Method for the Topological Description of a Molecular Structure in Drug Development

Authors 

Sun, W., Beijing University of Chemical Technology
Dai, J., Beijing Univ of Chem Tech
Ma, F., Beijing University of Chemical Technology
Wang, J., Beijing University of Chemical Technology
With the emergence of new diseases and the progress in medical science, efforts have never stopped to develop efficient medicine for specified diseases and clinical symptoms. Drug properties and clinical performance are essentially decided by its molecular structure. Amongst various methods, data analysis and feature extraction make it feasible to digitize a molecular structure and relate it to its clinical performance. As an effective digitized expression, molecular descriptor has been widely adopted in this field, most of them are obtained by corresponding adjacent matrix, which is used to describe the connection among atoms. However, adjacent matrix is not limited to describe the connection among atoms or functional groups, it can also be used to represent the relationship among self-defined molecular groups flexibly. If an adjacent matrix is established based on atoms, the number of elements could be significantly large, which makes hard to extract the feature structure for particular therapeutic effect and most computation is focused on connections among atoms. If the self-defined molecular groups can be served as basic units for adjacent matrix, computational time will be greatly reduced and more distinct influence of structure on drug properties can be obtained.

In this paper, different adjacent matrix is defined to explore the relationship between drug structure and its performance from multi-scale perspective. According to different partition rules, the structural relationship among atoms, functional groups and self-defined molecular groups can be represented by corresponding adjacent matrices respectively. Then neural network technology is adopted to analyze the connection between drug properties and the topological description, which is transformed from the adjacent matrix. A suitable definition of the adjacent matrix is confirmed to connect drug structure and its performance, on this basis, the contribution of different molecular structure on therapeutic effect, adverse effect and tolerance of the drug can be summarized, which is helpful for property prediction during drug development. Besides, antihemorrhagics drugs are taken as an example for validation purpose. The method proposed in this paper offers an alternative to study the influence of molecular structure on the performance of the drugs and to assist drug developers in designing drug structure.