Knowledge-primed neural networks
Interpretable deep learning
Knowledge-primed neural networks (KPNN) are neural networks that are trained using a knowledge-based network structure, which enables interpretability after training. Networks used in KPNNs should consist of nodes with labels (for example proteins in biological networks) that are connected based on prior knowledge. After training, KPNNs enable extraction of node weights (importance scores) that represent the importance of individual nodes for the prediction.
Knowledge-primed neural networks enable biologically interpretable deep learning on single-cell sequencing data
Fortelny N, Bock C
Genome Biology  21, 190 (2020). DOI: 10.1186/s13059-020-02100-5
Read online (without subscription)
* shared first or shared senior authorship