How often do unexpected scientific discoveries occur? More often that you might think

A study assessed 1.2 million biomedical publications and measured the ‘unexpectedness’ of their findings

Alexander Fleming famously observed that a stray mould from the genus *Penicillium* prevented bacteria in a Petri dish from growing.Credit: Chris Ware/Keystone Features/Getty

Science is littered with serendipitous findings — such as Alexander Fleming’s chance observation that a mould killed the bacteria he was culturing, which led to the discovery of antibiotics. Now, a study has put a figure on just how often chance findings happen.

Around 70% of biomedical papers include outcomes that would not be expected from what the scientists proposed in their funding applications, according to the study in Research Policy¹. Larger grants produced more chance findings.

“The bottom line is that ‘unexpectedness’ is not rare — this came through loud and clear,” says Ohid Yaqub, a biochemist and social scientist at the University of Sussex in Brighton, UK. Yaqub led the work as part of a wider project to understand the role in research of serendipity, of which unexpectedness is just one aspect.

The paper takes researchers “beyond the anecdotal view of serendipity in science” and “for the first time verifies it on a quantitative and statistical level”, says Telmo Pievani, a philosopher of biological sciences at the University of Padua in Italy.

Grant analysis

Yaqub and his colleagues studied more than 1.2 million publications stemming from over 90,000 grants in biomedical science awarded by the US National Institutes of Health (NIH) between 2008 and 2016.

The team trained a machine-learning algorithm to analyse the text of the papers and classify the research according to scientific categories laid out by the NIH, such as ‘chronic pain’ or ‘climate change’. The researchers compared the results with the categories listed in the aims and expectations of the grant proposal.

Around 70% of biomedical publications included text in at least one category that wasn’t mentioned in their grant — the authors’ measure of ‘unexpectedness’. After discounting closely related terms — for example, liver disease and liver cancer — the team found that 58% of papers had unexpected findings. And, on average, around one-third of all the categories assigned to a given paper were deemed unexpected.

Publications from larger grants were more likely to contain unexpectedness, as were studies published longer after the grant was issued, says Yaqub.

Two kinds of applied research grant — those involving categories related to clinical research or disease — yielded fewer unexpected results than did streams directed towards basic science.

But one kind of applied grant broke that trend. Papers stemming from NIH requests for applications (RFAs), in which funders outline specific objectives, were more likely to yield unexpected findings than were those initiated by investigators, although the effect was small.

The study offers a major suggestion for research policies, says Pievani: “It is okay to fund both basic research and applied research, as long as both are open to unexpected results and do not eliminate anomalies too hastily.” One might assume that goal-oriented funding could treat serendipitous outcomes “as disturbing noise, but this research shows that this is not the case”, he says.