Principle

In pursuit of the task of predicting potential endogenous ligands for nuclear receptors, we have developed a deep clustering model, as outlined in prior works . This model is founded upon the architecture of an autoencoder, comprising both an encoder and a decoder. What sets this model apart is its fusion of unsupervised reconstruction loss, clustering loss, and supervised classification loss within a deep learning framework. Consequently, this model not only excels in encoding and reconstructing input data but also possesses the capacity for classification and feature clustering, thereby enabling effective predictive functionality.