[AIRG] Neural networks for simulating biological cells, 03/06, CS 3310

[MATTHEW NATHAN BERNSTEIN <mnbernstein@xxxxxxxx>]

Hi AIRG,

Next week I will talk about a recent tool for simulating a biological cell using neural networks.

Title: Using deep learning to model the hierarchical structure and function of a cell
Authors: Jianzhu Ma and others â
Paper: https://www.nature.com/articles/nmeth.4627
Presenter: Matt Bernstein
âTime: 4pm - March 6, 2019â
Place: CS 3310

Summary:â

We tend not to think of neural networks as being easy to interpret; however, when the structure of the neural network matches the specifications of the problems, neural networks can be powerful tools for understanding the structure of the data. This has been most evident with convolutional neural networks trained on image data, which learn a hierarchy of features for making predictions on images. This week, I will discuss a project that builds neural networks for which their architecture is based on the structure of known biological pathways and subsystems in a cell.  The authors show that when they train these neural networks to predict the behavior of the cell, they are able to examine the activation of neurons in the network to gain insight into the machinery of the cell. I think this work is interesting because it provides a compelling case of how combining prior knowledge with neural networks can produce powerful tools for not only making predictions, but also gaining novel insights.