A simulation-based deep learning framework for spatially explicit malaria modeling of CRISPR suppression gene drive mosquitoes

Abstract

Engineered CRISPR gene drives are a promising new strategy for fighting malaria and other vector-borne diseases, made possible by genome engineering with the CRISPR-Cas9 system. One useful approach to predict the outcome of a gene drive mosquito release is individual-based modeling, which can be spatially explicit and allows flexible parameters for drive efficiency, mosquito ecology, and malaria transmission. However, the computational demand of this type of model significantly increases when including a larger number of parameters, especially due to the chasing phenomenon, which can delay or prevent successful population elimination. Thus, we built a simulation-based deep-learning model to comprehensively understand the effects of different parameters on Anopheles gambiae mosquito suppression and malaria prevalence among the human population. The results suggest that reducing the embryo resistance cut rate, reducing the functional resistance forming rate and increasing the drive conversion rate plays the major role in mosquito suppression and related phenomena. We also observed that the parameter space for eliminating malaria was substantially larger than that for mosquito suppression, suggesting that even a considerably imperfect drive may still successfully accomplish its objective despite chasing or resistance allele formation. This study shows that suppression gene drives may be highly effective at locally eliminating malaria, even in challenging conditions.