Event details from: A seminar series on interesting developments in molecular simulation

Date
2 September 2022
Time
11:00

Lecture 1

MIL: A problem formulation for computational chemistry data

by Enrico Riccardi (University of Oslo)

Abstract

Machine learning is a field growing at a remarkable rate, up to the point at which researchers and developers have to specialize in one of the sub-branches and are generally unaware of the development of the whole field. To support sustainable development, self standing approaches and packages are released, allowing direct access to the latest ML methods. Yet, any ML application has specific constraints and, depending upon the consequent dataset characteristics, models can be trained with very, very different outcomes. This commonly translates in the availability of very good solutions to rather uncertain questions.

This motivates the interest in problem formulations that, by design, handle uncertain questions: multiple instance learning (MIL). In the talk I will introduce the weakly supervised method, discuss how data properties can dramatically affect (any) model accuracy, the need for metadata representation and why such an approach can be valuable in computational chemistry and, in particular, in rare events.

Lecture 2

Simulating the permeability of phospholipid membranes

by An Ghysels (Ghent University)

Abstract

Permeability is a key property of biological membranes. Several strategies have emerged to predict membrane permeability from molecular dynamics (MD) simulations, which suffers from slow convergence of the statistics. The counting method, the Bayesian analysis to an inhomogeneous diffusion model, and a new methodology based on path sampling are presented. These methodologies are challenged in various applications. As a first application, oxygen transport is discussed, where oxygen should cross the membrane to reach its final destination at the cytochrome c oxidase protein where it is consumed. Next, the transport in so-called rafts is investigated, which are ordered regions floating in the generally disordered two-dimensional sea of phospholipid molecules. For some applications, the methodologies needed to be made compatible with curved membranes instead of flat membranes. An example of curved membranes is the residence time of drug molecules captured inside liposomes, which are small spherical membrane particles used as nanocarriers for e.g. vaccines. Another example of curvature is transport through caveolae, which are highly curved membrane invaginations that are characteristic for young healthy heart cells, but which gradually flatten with age for unknown reasons. Finally, the oxygen transport in grey matter is discussed, since axons in the brain are surrounded by myelin sheets consisting of stacked membranes.

Venue

Science Park 904, lecture room D1.113

This lecture is sponsored by the John van Geuns Foundation.