Chemical Informatics and Graphs

Chemical Informatics and Graphs

Problem being addressed

Chemical compounds can usually be represented as a graph (network) of covalent bonds. How can this graph data structure be modelled using machine learning for molecular and chemical discovery?


The availability of chemical compound databases that have complete maps of the covalent bond structure of chemicals presents a unique resource for chemical engineers and materials scientists. Essentially, these databases can be used to do chemistry without needing to get into a lab. The question is how can machine learning be used to assist this process? Using graph kernels the network representations of covalent bonds can be mapped onto neural network architectures and these architectures can be used to study the interactive properties of different chemicals in the database. This paper introduces three new graph kernels for studying chemical compounds, and applies the methods to public datasets relating to toxicology and cancer.

Advantages of this solution

These types of methods have potential to reduce costs and development time when it comes to chemistry.

Solution originally applied in these industries

chemical and materials

Chemical & Materials Industry

Possible New Application of the Work


Agriculture Industry

Pesticides tend to remain in the environment and accumulate in the food chain. Perhaps this research could be used to analyse pesticides and make their long term negative impact clear to regulators, so that pesticide manufacture can be abolished outright.


Environment Sector

A direct application of this research could be to understand the relationship between environmental pollutants and health. Given a set of chemicals that are present in the air, for example, what is the impact on the proteins created in the lungs?


Mining Industry

Chemical processing is used extensively in the mining industry. It would be worth investigating what the long term effect of these chemicals are on the environment, and this could be done using chemical databases and machine learning methods by, for example, considering the effect of soil chemistry and atmospheric chemistry on breaking down the compounds used in mining.

Source DOI: #############check-icon

search-iconBrowse all blitzcards