Preparing for the Next Pandemic: Predicting UV Inactivation of Coronaviruses with Machine Learning

The epidemic of coronaviruses has posed significant public
health
concerns in the last two decades. An effective disinfection scheme
is critical to preventing ambient virus infections and controlling
the spread of further outbreaks. Ultraviolet (UV) irradiation has
been a widely used approach to inactivating pathogenic viruses. However,
no viable framework or model can accurately predict the UV inactivation
of coronaviruses in aqueous solutions or on environmental surfaces,
where viruses are commonly found and spread in public places. By
conducting a systematic literature review to collect data covering
a wide range of UV wavelengths and various subtypes of coronaviruses,
including severe acute respiratory syndrome 2 (SARS-CoV-2), we developed
machine learning models for predicting the UV inactivation effects
of coronaviruses in aqueous solutions and on environmental surfaces,
for which the optimal test performance was obtained with R2 = 0.927, RMSE = 0.565 and R2 = 0.888, RMSE = 0.439, respectively.
Besides, the required UV doses at different wavelengths to inactivate
the SARS-CoV-2 to 1 Log TCID50/mL titer from different
initial titers were predicted for inactivation in protein-free water,
saliva on the environmental surface, or the N95 respirator. Our models
are instructive for eliminating the ongoing pandemic and controlling
the spread of an emerging and unknown coronavirus outbreak.