Quantifying the probability of clinical trial success from scientific articles

We sought to analyze how the number and quality of publications predict clinical trial success for a set of gene-disease associations. Limiting the scope of our analysis to genes in the protein kinase family and to oncology indications, we extracted gene-disease relationships from more than 12 million article titles and abstracts published between 1992 and 2012. We integrated these data with clinical trial information for FDA-approved kinase inhibitors and kinase inhibitors that failed owing to lack of efficacy. We found that, up until the year when a compound enters clinical trials, the cumulative number of publications about a gene-disease relationship corresponding to the compound's mechanism of action is, at the median, 30 for approved compounds but only four for failed compounds.