Loading…
NF

Norma Forero

Student Speaker | Étudiant.e

Infectious diseases cause devastating illnesses in humans, crops, and livestock, among others. And despite our knowledge and the multiple strategies to reduce transmission, we still have outbreaks from known pathogens and the emergence of new diseases, without mention how these transmission cycles can be modified by biodiversity loss, land transformation, and the worldwide temperature rise.

Differential equations have been widely used in the study of outbreaks, like SI models and its variations, which have been useful for understanding disease transmission for many years. But even with the utility of the model, one of the main problems lies in finding parameters with reliable values that give us accurate predictions about the dynamic of the tranmission across time. The data that are available are often limited to the infected population (out of many components), and maybe the rate at which the pathogen is spreading by the use of the R0 estimation methods.            

Recently, Rackauckas et al 2020 published Universal differential equations (UDEs), a novel methodology that could bridge the gap between differential equations and machine learning by merging the best of both. UDEs train the unknown interactions by incorporating physical constraints, delayed interactions, implicitly-defined events, and intrinsic stochasticity in the model. This is achieved by replacing parameters that are unknown by an artificial neural network, which when trained through differentiation, gives parameter values and interpretable mechanisms.            

Accordingly, our main goal is to use UDEs with available data with different transmission cycles and different pathogens, followed by analyzes of the predictory capacity of the model to find unknown parameters and their utility in the field of disease ecology.