Skip to main content
. 2024 Jul 1;64(14):5492–5499. doi: 10.1021/acs.jcim.4c00485

Figure 1.

Figure 1

Overview of the FS-CAP architecture. The context encoder (left) receives the Morgan fingerprint of each context compound multiplied by its associated activity value. A final context encoding is produced by aggregating the individual encodings of each context compound. The query encoder (right), which has different weights, receives the Morgan fingerprint of the query compound. A predictor network receives the concatenated outputs of each encoder and produces a final scalar activity prediction of the query compound.