Will Graph Neural Networks Revolutionise Computational Olfaction?

This week on Journal Club session Michael Schmuker will talk about a paper "Will Graph Neural Networks Revolutionise Computational Olfaction?".

How to predict the smell of a molecule given it's chemical structure? A group of researchers around Google's Alex Wiltschko have used Graph Neural Networks (GNN) to develop apparently outperform previous approaches to predict the smell of an odorant [1]. They proposed the "Principal Odor Map" (POM): a latent space which GNN learn from a data set of several thousand odorants. They show how the POM improves prediction of scent [2], how it aligns with metabolic pathways producing odors [3], and how it gives rise to the design of new mosquito repellents [4]. The group has now launched a computational olfaction startup [5]. In my talk I will introduce the GNN method and their approach to produce the POM, summarise their results, and discuss how the performance of their model compares to other established methods.

In my talk I will introduce the GNN method and their approach to produce the POM, summarise their results, and discuss how the performance of their model compares to other established methods.


Papers:

Date: 2023/02/24
Time: 14:00
Location: online

Share this post on: Twitter| Facebook| Google+| Email

Categories

Tags

olfaction synaptic plasticity robotics machine learning Associative memory Purkinje cell Computer Science Neuromorphic hardware neural networks neuroscience Convolutional Neural Networks PCA Neuronal Morphology optimization Evolutionary algorithms Dendritic computation Persistent homology Deep learning computational modelling Open position cerebellum Neurons computational neuroscience Software development COVID-19 Network connectivity Studentship Homoeostasis Structural plasticity Artificial intelligence