Fabian Theis (Institute of Computational Biology, Helmholtz Center Munich) will speak about
"Computational challenges in single cell genomics."
Single-cell technologies are on the verge of revolutionizing molecular biology. While single-cell analyses have previously been focused on proof of concept studies, the technology is now sufficiently robust to enable a broad range of applications, from basic biology and to questions in health and disease. Prime examples include data being generated for different tissues in the context of the Human Cell Atlas, an international effort that aims to profile hundreds of millions of cells from human organs. The scale at which single-cell data is being generated raises a series of computational challenges, from robust preprocessing to high-dimensional visualization and efficient comparison across subgroups and conditions in large-sample settings.
In this talk, I will show how machine learning can help solve some of these challenges at each step, from preprocessing data sets via deep learning to visualization based on diffusion processes and lineage estimation by graph-based methods. At each step I will aim to discuss computational challenges in upscaling those to "big data" scRNAseq. As example I will illustrate lineage estimation and population modeling from scRNAseq time-series of T-cell maturation. I will finish by discussing how to determine perturbation effects using variational autoencoders.