Predicting high-throughput screening results with scalable literature-based discovery methods

The identification of new therapeutic uses for existing agents has been proposed as a means to mitigate the escalating cost of drug development. A common approach to such repurposing involves screening libraries of agents for activities against cell lines. In silico methods using knowledge from the biomedical literature have been proposed to constrain the costs of screening by identifying agents that are likely to be effective a priori. However, results obtained with these methods are seldom evaluated empirically. Conversely, screening experiments have been criticized for their inability to reveal the biological basis of their results. In this paper, we evaluate the ability of a scalable literature-based approach, discovery-by-analogy, to identify a small number of active agents within a large library screened for activity against prostate cancer cells. The methods used permit retrieval of the knowledge used to infer their predictions, providing a plausible biological basis for predicted activity.