Acute lung infection (or pneumonia) is one of the leading causes of death in children worldwide—the WHO estimates that, in 2017, it killed more than 808,000 children under the age of five.

Clinically speaking, all pneumonias look very similar, but they can be caused by very different types of organisms, namely a virus or a bacterium, and sometimes even malaria (a parasite). Viral pneumonias in children are often milder and can resolve without specific treatment, but bacterial pneumonias require the rapid administration of appropriate antibiotics, otherwise they can quickly progress towards death.

The problem is that there are no diagnostic tests that can easily and rapidly distinguish between viral and bacterial pneumonias.

A machine-learning approach

Quique Bassat and his colleagues hypothesised that the body responds differently to a virus, a bacterium or a parasite, and that this different response could be reflected in the type of proteins circulating in the blood. To make a long story short, they were correct. They used a state-of-the-art technology to analyse over 1,000 proteins in a few drops of blood obtained from nearly 200 children who had pneumonia and in whom the causal agent had been correctly identified. They performed the study in Manhiça, a malaria-endemic region of Mozambique, where pneumonia is the leading cause of childhood deaths. The analysis revealed significant differences in protein expression between bacterial, viral and malaria-induced pneumonias. Thanks to a machine-learning model, they identified five protein markers that could correctly distinguish bacterial from viral infections with high sensitivity and specificity (i.e. capable of detecting 90% of positive cases and discriminating 95% of negative cases).

“With the appropriate technology, these markers could serve as a basis for future rapid diagnostic tests that could be performed in the field and help distinguish between those children who need antibiotic treatment and those who do not,” says Bassat.