National Institutes of Health
Title: Inferring a Gene Network in Drosophila Blastoderm
Date: Monday, January 27, 2020
Place and Time: Room 101, Love Building, 3:35-4:25 pm
Refreshments: Room 204, Love Building, 3:00 pm
Abstract. In this talk, I will talk about imputing missing data with neural networks and inferring a gene network using least absolute deviation (LAD) regression. Fowlkes et al.  published a set of gene expressions measured from 6078 Drosophila blastoderm during 6 different time cohorts that spanned the 50 minutes prior to the onset of gastrulation. Out of 95 genes and 4 proteins, only 27 of them had complete temporal information in all the cells. To impute the missing data, we trained and tested neural networks on the genes with complete profile as predictors and the genes with missing profile as targets. We then used LAD regression to infer a gene network that describes the dynamics of genes in the cells. Compared to least squares (LS) regression, LAD regression is less sensitive to outliers in a data set. Thus, the inferred network is a better predictor of the gene dynamics in a majority of cells, than a network inferred using LS regression. Using the gene network, we predicted the effects of a gene knockout on the dynamics of gene evolution.  C. C. Fowlkes et al., A Quantitative Spatiotemporal Atlas of Gene Expression in the Drosophila Blastoderm. Cell. 133:364-374, 2008.