Characterizing Phenotype Abnormality by Variational Auto Encoder
Variational auto-encoder (VAE) acquires not only a low dimensional representation of the data, but also the probability distribution of the data, in its latent space. Therefore, if an input data is not from the trained category, it fails to find an appropriate point in the latent space, which leads to a poor reconstruction. Thus, it can be used for data abnormality detection. We use VAE to detect a phenotype abnormality in a biological system. When each of lethal genes in early embryogenesis is knocked down by RNAi technique, an observed abnormality can be linked to the corresponding gene function. We train VAE by the wild type (WT) data without any gene manipulation, and use to characterize a phenotype of RNAi embryo in two cell stage. Abnormality is defined by a data reconstruction error, and several genes are found whose absence causes the abnormality, including some already known genes.
KeywordsVariational auto encoder Anomaly detection Latent space Biological dynamics data
- 1.Kingma, D.P., Welling, M.: Auto-encoding variational bayes. In: International Conference on Learning Representations (2014)Google Scholar
- 5.Zeiler, M.D., Krishnan, D., Taylor, G.W., Fergus, R.: Deconvolutional networks. In: 2010 IEEE Computer Society Conference on Computer Vision and Pattern Recognition, pp. 2528–2535 (2010)Google Scholar
- 6.Ioffe, S., Szegedy, C.: Batch normalization: accelerating deep network training by reducing internal covariate shift. In: International Conference on Machine Learning, pp. 448–456 (2015)Google Scholar