Abstract—Machine learning is widely used for phenotype prediction from gene expression data. However, deep learning, that is currently one of the most performant methods, have been very few studied for this problem. In this paper we construct a deep multilayer perceptron using different regularization methods to deal with the problem of small training samples. A large set of unlabeled data is used in an unsupervised pre-training procedure in order to improve the learning of the neural network. The results on several public microarray datasets show that the deep learning improves significantly the performance of the state-of-the-art.

Index Terms—Deep learning, phenotype prediction, gene expression data.

The authors are with IBISC, Univ. Evry, University Paris-Saclay, IBGBI, Bd de France, 91000 Evry, France (email: blaise.hanczar@ibisc.univ-evry.fr).

Cite: Blaise Hanczar, Mathieu Henriette, Toky Ratovomanana, Farida Zehraoui, "Phenotypes Prediction from Gene Expression Data with Deep Multilayer Perceptron and Unsupervised Pre-training," International Journal of Bioscience, Biochemistry and Bioinformatics vol. 8, no. 2, pp. 125-131, 2018.