In our paper out today in TrAC Trends in Analytical Chemistry we look back on a decade of research into the application of machine learning to the diagnosis of cancer from spectral data. This combination of rapid analytical technique and automated data processing offers much promise as a fast, and in some cases non-invasive, diagnostic tool. So why isn’t it used yet in clinical practice? Our conclusion was that limited sample sizes and validation undermine the outcomes of most published studies in this area, leading us to recommend larger studies and public sharing of data to advance the field. For all the details, see the paper.