1. ACLS International Internship Report
Norwich Mungkalaton
@ University of Edinburgh

2. University of Edinburgh
Internship term:
1 October 2017 to
11 March 2018
University was founded on th oldest.
Notable Alums and Staff
Charles Darwin
James Clerk Maxwell
Alexander Graham Bell
King’s Building Campus— James Clerk Maxwell Building
House #1
Central Campus
King’s Building Campus
House #2

3. University of Edinburgh
Supervisor
Assistant Professor
Dr. Bartlomiej Waclaw
Royal Society of Edinburgh Research Fellow

4. University of Edinburgh
School of Physics and Astronomy
Institute of Condensed Matters and Complex Systems
Physics of Living Matters (Bacteria + Cancer)
Statistical Physics and Complexity
Population Dynamics Group
Dr. Bartlomiej Waclaw

5. Cancer Growth Simulation
Waclaw, B. Nature 2015

6. Cancer Growth Simulation
Paul Dr. Bartek Waclaw website

7. Cancer Growth Simulation
4 Mechanisms of Tumor Evolution
Replication – Based on available site, (normal cell/matrix), cancer cell can choose to replicate itself to (fixed position) nearest neighbors.
Mutation – The mother cell can induce genetic alteration to the daughter cell. Mutation can be driver mutation or passenger mutation.
Dispersal – Cell can detach from the current metastatic lesion and form another lesion and reattach to the main lesion. To find new nutrition resources.
Cell Death – Cell can decide to die if there is no more space or nutrition resources have been depleted. Metastatic lesion can also be targeted for treatment.

8. Cancer Growth Simulation
Primers
2D/3D Real time simulation of tumor evolution
Stochastic Simulation
Lattice-based model
Eden Growth Cluster
CPU
Randomly picked a point in linear 1d array
Randomly growing the cluster based on growth probability
Cluster can grow dynamically
GPU
Cluster is fixed size (Power of 2)
Need to ask for fixed lattice before execution.
1 Thread = 1 Cell location (2D)
1 Thread = Sub-lattice (3D)
7-10 times performance gains
Jason Miller Wired 2017

9. Cancer Growth Simulation
CPU is optimized for serialized execution (low latency/delay)
GPU is optimized for high-throughput execution (Multi-thread)
NVIDIA

10. Cancer Growth Simulation
Checkerboard decomposition 2D Lattice in GPU
yoffset = Height of block
xoffset = Width of block y
All cell read/write asynchronously. Need to call device synchronize to wait for other block update.
XOR update pattern x
Dr. Martin Weigel - Winter school on GPU computing (2011)

11. Cancer Growth Simulation
3D Lattice-based Simulation
Need to convert 3d coordinates to 2d texture map.
Threads/Blocks/Grids will now responsible for square space read/write.
Not efficient due to unable to perform coalesced memory access.
GPU Computing Gems 3rd Edition

12. Cancer Growth Simulation
Project Summary
Cancer growth + Genotype Mutation + Cell Death (Treatment)
Multi-GPU Algorithm
GPU struggles with non-deterministic execution model.
How can we reduce the non-deterministic process such as random growth to (somewhat) deterministic.
Replication
Mutation
Real-time rendering +
Tesla C2070
GTX470
Dr. Bartek Waclaw website

13. Cancer Growth Simulation
2D Lattice-based Simulation

14. Cancer Growth Simulation
3D Lattice-based Simulation

15. Cancer Growth Simulation
3D Lattice-based Simulation (Treatment)

16. Living & Working in Edinburgh
Lessons Learned
Mentorship
Creativity & Originality
Collaboration
Research Methodologies
Work Critique

17. THANK YOU ACLS Committee for this Opportunity ACLS Staff - Kumagai-san
Prof. Akihiko Konagaya
Dr. Bartlomiej Waclaw & Prof. Rosalind Allen