A Case Study in Machine Learning for Combinatorial Chemistry

Combinatorial chemistry is an advance of this decade that makes it possible to synthesize over a million compounds in a day. These compounds can then be tested for biological activities of various kinds via high-throughput screening, in the search for new drug leads. Unfortunately, the active compounds may have undesirable properties such as metabolic instability or toxicity. Nevertheless, machine learning can be used to determine what the active compounds have in common structurally, in contrast to the inactive compounds; this information can then be used to guide the design of improved structures. This paper describes the challenges raised by this application of machine learning, and it presents a case study of the approach. The paper focuses on the computational issues involved, as opposed to the chemical and biological issues.