Skip to main content
. 2021 Jul 30;19:4593–4602. doi: 10.1016/j.csbj.2021.07.032

Fig. 4.

Fig. 4

Comparison of neural network architecture. a. The Baseline model is trained to only predict whether a compound is a natural product. b. NP_AUX predicts additional surface descriptors. c. autoencoder-like NP_AE aims to encode the ECFP into a neural fingerprint from which the ECFP should be reconstructed. Additionally, the fingerprint is used to predict whether a compound is a natural product.