ANC Workshop - Domas Linkevicius

Tuesday, 4th October 2022

Scientific machine learning and partially observable chemical reaction networks

Abstract: A large part of biological systems containing chemical reaction networks can be investigated via the currently available experimental methodologies. Specifically, variables like the concentrations of species and reaction structure can be found. However, this is not the case for all systems, for example real neuronal synapses, which are my main interest. A synapse contains on the order of a thousand unique protein species, but current computational models capture only a small part of that. Due to the size of synapses, it is currently not feasible to track concentrations of species over time and difficult to derive reaction structures. In trying to tackle this problem I am attempting to construct a model that would combine a machine learning approach - neural ordinary differential equations - that would also benefit from the large amounts of already published work in computational neuroscience dealing with synaptic biochemical networks. In this work-in-progress talk I will present concepts and ideas which I am trying to use to arrive at a hybrid model that combines machine learning and chemical reaction network theory.

Event type: Workshop

Date: Tuesday, 4th October 2022

Time: 11:00

Location: G.03

Speaker(s): Domas Linkevicius

Chair/Host: Arno Onken