Talk: The Scientific Python Tool Stack for Biomedical Research

Organizers

• Dr. Prabhu Ramachandran (IITB) https://www.aero.iitb.ac.in/~prabhu/

Overview
Whether it is for predicting the 3D structure of DNA, to develop new clustering algorithm for time course gene expression data, or to analyse terabytes of videos of cells exposed to toxins or drugs, the Scientific Python Tool Stack presents a unique set of tools, easy to use yet flexible enough for a wide range of applications.

I will describe in this talk three applications, and how our we leveraged some of the tools of Scientific Python Tool Stack (numpy, scipy, Matplotlib, scikit-learn, autograd, cython…) for better and faster research. First, I will talk about Pastis, a tool to infer 3D structures of DNA from contact map matrix, and how we moved from a C++ code base to a Python one without loss of speed. Second, I will present how using appropriate data structures for matrix balancing algorithms, used for preprocessing of DNA contact maps can be competitive with parallized C++ code. Third, I will show how autograd helped us solved complicated optimization problems without having to tediously derivate complicated functions by hand, allowing us to do stability analysis on clustering of time course gene expression data.

Speaker Bio

Dr. Nelle Varoquaux is a postdoctoral fellow at the Department of Statistics at UC Berkeley and a BIDS data science fellow.

She received a PhD in computational biology from École des Mines de Paris in 2015. Her research interests are in statistical machine learning and scientific computing applied to molecular biology problems, such as inferring
the 3D architecture of the genome or data-integration methods to better understand gene regulatory networks. Recently, as part of the interdisciplinary institute BIDS, she also studies open-source communities and contributors in collaboration with social scientists.

She is a core-developer of scikit-learn, MarkUs and Matplotlib, was Program Chair for Scipy in 2015 and 2016.