The probability of successful drug discovery is low, and there is a need to improve productivity.In recent years, attempts have been made to improve the efficiency of compound search by utilizing AI.Many methodologies such as deep learning have been proposed, but few studies have focused on AI learning data.In addition, it is said that the hit rate remains at 0.1% or less in the "High Throughput Screening (HTS) *" used in general drug discovery methods.In drug discovery so far, we have focused only on the few Hit compounds, and the many "failed data" groups that can be obtained have not been fully utilized and are simply discarded.

　The research team focused on the characteristics of the data in this field of drug discovery, that is, the characteristic that most of the compounds are inactive. In AI learning, if the ratio of compound data that does not show activity (negative example) is increased 1,000 times to that of compound that shows activity (positive example), the misclassification of AI will be less than 100/1, and the discrimination ability will be reduced. Has improved dramatically.

　By utilizing this knowledge, it will be possible to perform screening with a smaller number of verifications and drug resources, and it is expected to have a wide range of applications that will lead to significant cost reductions in drug discovery research.The research paper was published in the academic journal "molecular informatics" (impact factor: 3) on Thursday, March 18th.

* High Throughput Screening is a technology that identifies useful compounds from a large number of compounds with high efficiency.

Paper information:[Molecular informatics] Effect of Learning Dataset for Identification of Active Molecules: A Case Study of Integrin αIIb β3 Inhibitors