Derisking\nFuture Agrochemicals before They Are Made:\nLarge-Scale In Vitro Screening for In Silico Modeling of Thyroid Peroxidase\nInhibition

Inhibition of thyroid peroxidase (TPO) is a known molecular\ninitiating\nevent for thyroid hormone dysregulation and thyroid toxicity. Consequently,\nTPO is a critical off-target for the design of safer agrochemicals.\nTo date, fewer than 500 structurally characterized TPO inhibitors\nare known, and the most comprehensive result set generated under identical\nconditions encompasses approximately 1000 compounds from a subset\nof the ToxCast compound collection. Here we describe a collaboration\nbetween wet lab and data scientists combining a large in vitro screen\nand the subsequent development of an in silico model for predicting\nTPO inhibition. The screen encompassed more than 100,000 diverse drug-like\nagrochemical compounds and yielded more than 6000 structurally novel\nTPO inhibitors. On this foundation, we applied different machine learning\ntechniques and compared their performance. We discuss use cases for\nin silico TPO models in agrochemical research and explain that model\nrecall is of particular importance when selecting compounds from large\nvirtual compound collections. Furthermore, we show that due to the\nhigher structural diversity of our training data, our final model\nallowed better generalization than models trained on the ToxCast data\nset. We now have a tool to predict TPO inhibition even for molecules\nthat are only available virtually, such as hits from virtual screenings,\nor compounds under consideration for inclusion in our screening collection.\nStructures and activity data for 34,524 compounds are provided. This\ndata set includes almost all inhibitors, including more than 3000\nproprietary structures, and a large proportion of the inactives.