【Postponed】High-dimensional Statistical Modeling Team Seminar (Lotfi Slim, NVIDIA)

Title: Genomics in the era of Machine Learning

Abstract:
Genomics is key to understanding the biology of complex diseases and the diagnosis of critically-ill patients. Yet, several computational and algorithmic bottlenecks still limit the deployment of genomics on a large scale. Machine Learning has quickly proven to be one of the most promising approaches to tackle these challenges. For instance, Deep Learning (DL) has been successfully used to cope with the accuracy and throughput requirements of Next-Generation Sequencing (NGS) instruments. The output of such an instrument is a large set of sequences (reads) which jointly determine the mutation profile of an individual. Computer vision-based models such as DeepVariant are regularly used for this task. Additionally, new sequencing technologies that complement DNA sequencing are being developed, most notably single-cell and spatial transcriptomics where Variational Autoencoders (VAE) have become the de facto approach for encoding cell features. This talk will cover the aforementioned applications among others, in addition to the current technological limitations which call for further machine learning development.

Bio:
Lotfi Slim is a Genomics Solution Architect at Nvidia. He helps both academia and industry achieve the best possible performance for their genomics workloads. Genomic data offers unique computational challenges for which Lotfi leverages GPUs through Deep Learning and HPC CUDA code. Prior to joining Nvidia, he developed during his PhD new machine learning algorithms for the detection of statistical interaction in Genome-Wide Association Studies. His PhD was under the supervision of Clement Chatelain (Sanofi), Jean-Philippe Vert (Google Brain) and Chloe-Agathe Azencott (Institut Curie/Mines ParisTech). He holds a master’s degree from Ecole Normale Superieure in Statistical Learning and an engineering degree from Mines ParisTech in applied statistics.