This paper used computational methods based on chemical structural similarity to predict the off-target activity of 656 marketed drugs against 73 protein targets. About 600 of the ∼900 predictions were then tested experimentally, and around 50% were found to be genuine. For example, the methods predicted that the non-steroidal synthetic oestrogen chlorotrianisene (which is linked to abdominal pain) had potent affinity for cyclooxygenase 1 (an enzyme found in gastric mucosa); this prediction was confirmed using an ex vivo assay. So this model can detect previously unappreciated side effects, albeit with a high false-positive rate.