Machine Learning Helps Identify Primate Species Likely to Spread Zika
Investigators used computer algorithms to predict primate species likely to be positive for Zika virus in Central and South America.
Machine learning may be an important tool in controlling and eradicating the Zika virus, according to a recent study that used machine learning to predict the virus among primates in Central and South America.
The study by investigators at the Cary Institute of Ecosystem Studies and IBM was published in the journal Epidemics. The machine learning model identified known flavivirus carriers with 82% accuracy and predicted the risk of Zika among primate species.
"We were surprised to find that very common primate species were predicted to have high risk of carrying mosquito-borne flaviviruses, including Zika virus," lead author Barbara A. Han, PhD, disease ecologist at the Cary Institute of Ecosystem Studies, told Contagion®. "In Central and South America, the possibility of spill-back infection (from humans to wild primates) is alarming. If Zika virus establishes a sylvatic cycle it could be exceedingly difficult to control."
Those species with more than 90% risk scores for the virus included species common in developed areas: tufted capuchin, the Venezuelan red howler, and the white-faced capuchin.
"Two of the species we found to be high-risk for carrying ZIKV was white-fronted capuchin (Cebus albifrons), which is commonly kept as pets and captured for live trade in southern America, and spider monkey (Saimiri boliviensis) which is hunted for bushmeat," study coauthor Subho Majumdar, PhD, senior inventive scientist in the data science and AI group of AT&T Labs Research, told Contagion®. "The insight from such findings for clinicians/health care providers is that there can be many unconventional sources for infectious diseases that may increase their outbreak potential, and data-driven insights can often help narrow down suspect sources of information in such situations."
Investigators used multiple imputation and Bayesian multi-label machine learning to fill in data gaps and assign risk scores for potential Zika positivity. Along with Zika, they assessed the traits of yellow fever, dengue fever, Japanese encephalitis, St. Louis encephalitis, and West Nile virus along with the biological and ecological traits of 18 primate species that have tested positive for any of these mosquito-borne diseases.
"The extent of missing data was a surprise," Majumdar said. "Our feature set consisted of bodily and behavioral features of all primates, obtained from a publicly available database (Pantheria). In spite of being fairly well-studied, it was surprising to see that many well-known primate species had so many features missing."
Machine learning was used to fill in missing data about such traits as metabolic rate, gestation period, litter size, and behavior based on connections between organisms.
"The next steps are to validate model predictions with targeted field surveillance of particular species, and bench research to characterize how infectivity and competence of Zika virus may vary across these species," Han told Contagion®. "The wildlife link will be increasingly important for informing human epidemiology and infection prevention. Zika virus is the most recent of many future examples. Therapeutics (vaccines) tend to focus on protecting humans, but in addition to this perhaps it's time to consider animal therapeutics and their role in preventing future spillover from established sylvatic reservoirs."
The outbreak of Zika, which began in 2015 and spread to more than 30 countries, has diminished, but risk remains, particularly among laboratory and biomedical researcher workers.