The identification of brain tumours can be improved by studying DNA methylation data, reports a study published online this week in Nature.
Correctly diagnosing tumours is vital for the treatment of cancer. With around 100 known types, however, tumours of the central nervous system are especially hard to identify accurately. To address this, Stefan Pfister and colleagues developed a machine-learning program that classifies methylation data. Methylation is the process by which methyl group molecules are attached to DNA, thereby changing the availability of the information within the DNA. This process not only occurs naturally during normal cell function - giving each cell a characteristic methylation fingerprint - but also in diseases such as cancer. Methylation can therefore reveal information on the type of tumour and the cell type from which these tumours were formed.
Trained with reference data taken from around 2,800 patients with cancer, the authors’ program can use the methylation fingerprints to identify 91 tumour types. They tested it on 1,104 central nervous system tumours that had already been examined manually, and identified misdiagnoses in 12% of cases. Alongside improving diagnostic accuracy, the program’s objectivity allows for new rare tumours to be identified as such - unlike with manual classification, in which there is a pressure to assign tumours to known categories, even in atypical cases.
To make the new approach accessible, the authors have generated a free online tool that analyses uploaded data in mere minutes. The tool has already been used over 4,500 times since December 2016 and gives users the option to share their data to further refine the algorithm. The integration of methylation identities into an automatic classifier of brain tumours, the authors conclude, also provides a blueprint to create similar tumour-classifying algorithms for other cancer types.
Ecology: Fast-growing trees die young and could affect carbon storageNature Communications
Epidemiology: US COVID-19 cases may be substantially underestimatedNature Communications