Eventbrite Link: https://www.eventbrite.co.uk/e/molecules-graphs-ai-workshop-tickets-55016184890

Description:
The representation of molecules as connected graphs and the application of graph theory has been very useful in defining aspects of molecular structure, molecular energy levels and identifying unique topology features. In the workshop we seek to explore the ways in which molecular graphs can be used to drive property and other predictions using Machine Learning and other AI techniques. All ideas welcome – come and discuss and debate and come up with new plans! There will be three keynote talks, and a chance for participants to present their areas of research and ideas. Lunch will be provided, and the day will end with networking drinks.

Keynote Speakers
There will be three keynote talks from experts in this field.

• Professor Mahesan Niranjan – Professor of Electronics and Computer Science at the University of Southampton & AI3SD Co-Investigator. Nirajnan works in the area of machine learning, and his research interests are in the algorithmic and applied aspects of the subject. He has worked on a range of applications of machine learning and neural networks including speech and language processing, computer vision and computational finance. Currently, the major focus of his research is in computational biology. Some of his work (e.g. the SARSA algorithm in Reinforcement Learning) have been fairly influential in the field. He has held several research grants from the Research Councils in the UK, and the European Union. Currently, his main focus is on architectures and algorithms for Deep Learning and inference problems that arise in computational biology.
• Professor Sophia Yaliraki – Sophia is a Professor of Theorerical Chemistry at Imperial College London. Sophia is a Professor of Theorerical Chemistry at Imperial College London. Her research interests are in the theory of self-assembly in biology and molecular electronic devices and coarse-graining and model reduction techniques in dynamics.
• Professor Patrick Fowler – Patrick has been a Professor of Theoretical Chemistry at the University of Sheffield since 2005. Prior to that he had worked at both the University of Durham as a Senior Demonstrator, the University of Cambridge as a Postdoctoral Research Fellow and the University of Exeter where he became a professor. His research focuses on molecular properties, ring currents, aromaticity, fullerenes, molecular electronic devices, symmetry and discrete mathematics in chemistry, and he has published extensively in these areas. He was also elected a Fellow of the Royal Society in 2012.

Keynote Abstracts

• Inference from Outliers – Professor Mahesan Niranjan: Classic machine learning is largely about classification and regression problems. However, many practical problems of interest in genomics, condition monitoring, medical diagnostics and security are better posed as problems of detecting novelty. In this talk, I will describe two applications of extracting useful information from novel data, in problems relating to modelling cellular protein concentrations and the solubility of synthetic chemical molecules. The algorithmic framework poses a robust support vector regression problem and the resulting non-convex optimisation problem is solved using a difference-of-convex formalism. (Part of this work is supported by grant EP/N014189/1, “Joining the Dots: From Data to Insight” from the EPSRC).
• Unsupervised, multiscale learning through atomistic graphs: From molecules to systems – Professor Sophia Yaliraki: We have derived an all-scale graph partitioning approach that preserves atomistic physico-chemical detail and by using diffusive processes on the graph (both on the node and the edge space), we have shown that we can obtain the behaviour of biomolecules and biomolecular assemblies at different timescales without the need of any reparametrisation or a priori selection of relevant timescales. The approach is computationally efficient and general and can be applied to molecules, molecular assemblies as well as data. We will showcase the theory with examples from predictions and experimental verification of mutations that control protein dynamics at different scales (AdK), prediction of allosteric sites for drug design and communication and signalling in multimers and assemblies (ATCase, Rubisco). Finally, the application of this unsupervised learning approach to trajectories and free text will be briefly discussed.
• Source-and-sink models for molecular conduction – Professor Patrick Fowler: This talk describes recent progress in Sheffield in describing ballistic molecular conduction with the Ernzerhof source-and-sink-potential (SSP) model. SSP gives a broad classification of conduction behaviour at the graph theoretical level. We have been able to derive selection rules, classifications and intuitive descriptions that remain useful at higher levels of theory. This talk is based on joint work with Barry Pickup (Sheffield), Irene Sciriha (Malta) and Martha Borg (Sheffield).

Programme:

• 10:00-10:30: Coffee & Registration
• 10:30-10:45: Welcome Introduction – Professor Jeremy Frey
• 10:45-11:15: Unsupervised, multiscale learning through atomistic graphs: From molecules to systems – Professor Sophia Yaliraki
• 11:15-12:00: Inference from Outliers – Professor Mahesan Niranjan Presentations from Participants to initiate discussions
• 12:00-12:45: Presentations from Participants to initiate discussions
• 12:45-13:45: Lunch
• 13:45-14:15: Source-and-sink models for molecular conduction – Professor Patrick Fowler
• 14:15-14:45: Initial Discussions to form Working Group Topics
• 14:45-15:15: Coffee
• 15:15-16:30: Working Group Discussions
• 16:30-17:00: Report back and form Action Plan
• 17:00-18:00: Drinks reception

FAQs:

1. Who should attend?
Anyone with an interest in Molecules, Graphs, Artificial Intelligence, Machine Learning, Deep Learning. We welcome members from academia, industry and government. We are always looking to grow our Network+ and bring in people with a wealth of experience in the many different subject areas that are needed so that we can form interdisciplinary partnerships and work together to further the field of Scientific Discovery.

2. What will I get out of it?
You will be able to network with likeminded people who have research interests that complement yours. There will be two keynotes around the topics of Molecules, Graphs and AI to spark discussion and ideas. There will be an opportunity to present your own research interests/areas of expertise briefly, with plenty of opportunity to have general discussions and some specific topic based discussions in smaller groups. Members of the Network Executive Group will also be in attendance so you will be able to find out more about our Network and the opportunities we have available including funding opportunities and the types of events we will be running (e.g. workshops, conferences and hackathons).

3. What are the aims of the workshop?
This workshop is aiming to help the Network+ to drive progress in this area and facilitate collaboration by introducing people to make new interdisciplinary teams, and to produce new grant applications. To achieve this we may commission literature reviews, papers, or small scale investigations to test out new ideas. We welcome ideas and suggestions about how to go forward in this area and how best to achieve our aims.

4. What are the main themes of the workshop?
Molecules are often represented by connected graphs showing in some sense the bonding between the atoms. These graphs can serve as the input in the quantum chemistry packages to determine the 3D molecular geometry and with high level and time-consuming calculations obtain the electron density and electric fields surrounding the molecules and with even greater difficulty the interactions between molecules obtained and used as the basis for simulation for collections of molecules. In this workshop we wish to explore how the graphs themselves can be used to generate molecular properties, predict drug activity, suggest crystal structures, without going through the QM route. Can the graphs be the input to Machine Learning models? What fundamental properties of the graphs directly relate to molecular behaviour? How do these relate to the topology of the systems? Can we predict drug solubility, drug binding or the nature of molecular assemblies and crystals structures? For example graph theory predictions of Huckel Molecular orbital energies give very fundamental predictions about the nature of some molecules. The graph approach is very useful when considering similarity between molecules and the transformations between molecules and we seek to explore how AAI can play a role in these areas too.