AI method speeds outbreak modeling in a research study

Seventy-two Ebola genomes from Sierra Leone’s 2014 outbreak became the test case for an INRAE team trying to make infectious-disease modeling faster. The researchers used neural posterior estimation, or NPE—a deep-learning method previously applied in neuroscience and astrophysics—to analyze the genetic data and estimate how the outbreak unfolded.

The underlying method starts with a phylogenetic tree, a map of genetic relationships between pathogens. Its branches reflect mutation and transmission history, while its branching points help identify which strain came from which parent strain. By combining those trees with dates, prevalence and locations, researchers can estimate parameters such as the speed of spread, the duration of infection and a pathogen’s geographic origin.

That combination is also the bottleneck. Building robust estimates requires many genetic sequences and comparisons with conventional epidemiological data, while the models must handle the dates, locations and population measurements attached to those sequences. The study reports that NPE produced estimates similar to those obtained with traditional inference methods, while calibrating the models much more quickly. The researchers describe it as the first use of this method with genetic data in a phylodynamic application.

Concretely, the gain is time and flexibility for outbreak teams: a faster model could support live testing of possible epidemic scenarios, add new genetic sequences as they arrive and potentially work with thousands of sequences. The same approach could also support the management of epizootics, or disease outbreaks among animals.

The study used a dataset of 72 Ebola genomes. The scientists have made detailed online tutorials available so others can replicate the analysis.