How can new mathematics and GPUs help understand Parkinson’s disease and other complex disorders? Clive Bowman

September 23, 2016 –
14:00 to 15:00
Access Grid Room (room 277)
Oxford e-Research Centre, 7 Keble Road, Oxford

Seminar No booking required Open to all Coffee and cakes
The Oxford e-Research Centre is pleased to welcome Professor Clive Bowman from the Mathematical Institute, University of Oxford and Olivier Delrieu from C4X Discovery Ltd. They will present a seminar entitled “How can new mathematics and GPUs help us understand Parkinson’s disease and other complex disorders?”.

This event is open to all and no booking is required. Coffee and cakes will be provided and there will be the opportunity to talk to the speakers after the event.

About our speakers

Clive Bowman, CSci FRSM FRSS FLS
Mathematical Institute, University of Oxford

Professor Clive Bowman is a Royal Society Industrial Fellow. His research at the University of Reading and Mathematical Institute, Oxford while at Daiichi-Sankyo, covered high dimensional visualisation of diverse clinically relevant data. The ‘individualised divergences‘ methodology allows simultaneous analysis of all data types filtered by reference to a base group. Passionate about success in medical ‘Big Data’, he is an EU expert, entrepreneur and has held a variety of novel and challenging senior pharmaceutical industry, government regulation, academic body and governance committee posts. He is willing and ideally placed to bring considerable analytical expertise to generate collaborative success.

Olivier Delrieu, MD, MSc, MBA
VP Clinical Development & Mathematics, C4X Discovery Ltd

Olivier Delrieu trained in Neurology at the Faculty of Medicine of Lille, France, received an MSc in Human Genetics at the Pasteur Institute of Paris and achieved an executive MBA at the EDHEC Business School. He has international academic and pharmaceutical experience in clinical development, target discovery and genetics. His personal interest is in technologies and business models supporting the development of personalized medicines and reduction of attrition.

Abstract

The objective of C4X Discovery is to deliver better and safer medicines that could reach patients more quickly. One of our technologies is a new mathematical method, Taxonomy3. Applied to large genetic datasets it allows us to better understand complex disorders, and to provide robust biological hypothesis to develop new medicines. These analyses are computer-intensive, and so we recently moved our software from CPUs to GPUs in collaboration with Mike Giles, Wes Armour and Nassim Ouannoughi at the Oxford e-Research Centre.

R&D in the pharmaceutical industry is a long, expensive and complex process. At the start, a new biological hypothesis drives the discovery of new drug candidates. When proven safe in animals and healthy volunteers these candidates are then tested in patients to confirm their efficacy as medicines. This process takes about 10 years. Unfortunately, the initial biological hypothesis is proven wrong 50% of the time. This figure is worse for neurodegenerative disorders.

Encouragingly, when the initial biological hypothesis is based on genetic clues, the probability of success in clinical development is significantly higher. However, classical statistical methods can explain only a fraction of the genetic load expected to be found in complex disorders. This gives a significant opportunity for new mathematical methods to add value by extracting more genetic insights.

The Taxonomy3 method is based on ‘individualised divergences’ (for an introduction to these negentropies see: – Geometry Driven Statistics (ed: Dryden, I L & Kent, J T). John Wiley 337-355). Decomposing the dimensionality of this nonlinear transform of data highlights the evidence particular features carry for the comparison of interest. Here whether a person has a characteristic disorder or not.

The method is implemented as a C++ / openMPI program running on an Amazon cloud Linux cluster. A typical resampling analysis requires several thousand CPUs for a couple of weeks. To allow a cost-effective analysis of large datasets, we recoded some modules in CUDA to make use of GPUs. Some modules now run about 100 times faster, with an overall cost reduction of 5 to 10 fold.

This work allowed us to apply this technology to a Parkinson’s disease genetic dataset. We discovered evidence for new genes of interest specific to patient subgroups. These findings had a significant business impact for our company.