Fig. 2

Our proposed deep architecture. The architecture consisted of seven subnets to separately handle one-hot and six physical-chemical properties, and then their detective high-level representation into an ensemble layer to predict final results