Chemical Data in Self Organizing Maps as Input for Artificial Neural Networks

What is it about?

We used the output of self organizing maps (SOM) reducing high dimensional data as an input for artificial neural networks (ANN) to predict novel antimicrobial peptides. With these so-called hybrid networks we were able to increase correct predictions of active peptides by 6-15% compared to standard feed-forward ANNs.

Featured Image

Why is it important?

Preprocessing molecular descriptor data into a two-dimensional map for training neural networks could establish a link to “deep” image recognition and analysis methods like the convolutional network approach for advanced machine learning and quantitative structure activity relationships (QSAR).