simulation and study of large-scale bacteria-materials interactions via bioscape enabled by gpus
TRANSCRIPT
Challenge: At any given +me, each par+cle can react with any number of other par+cles within a neighborhood. How to choose reagents independently, autonomously and consistently across the whole system within a parallel computa8onal framework? Example: Par+cles of type A and type B can react within a neighborhood to produce Par+cle C. Mul+ple objects of same species can be present within the neighborhood. Each thread has to be aware of the selec8on of other threads in the system and adjust its own selec8on accordingly.
Solu8on: We introduce a novel and efficient Parallel Selec+on Algorithm that maintains the state of the system and selects reagents and carries out reac+ons in parallel.
Massively Parallel Implementa8on On a massively parallel plaEorm each par+cle is assigned a work-‐unit(thread) which independently and autonomously simulates each par+cle’s behavior, by broadly following the steps:
1) Ini+alize par+cle posi+ons and movements. 2) Choose reagents and carryout reac+ons in parallel. 3) Update the state of reac+ons, new par+cles and par+cle life spans. 4) Update par+cle co-‐ordinates by avoiding collisions. 5) Increment fixed global +me step. 6) Repeat steps 2 through 5 un+l simula+on +me is reached.
Simula8on and Study of Large-‐Scale Bacteria-‐Materials Interac8ons via BioScape Enabled by GPUs Jie Li, Vishakha Sharma, Narayan Ganesan, and Adriana Compagnoni
Department of Electrical and Computer Engineering Department of Computer Science, Stevens Ins9tute of Technology
Results and Conclusion
BioScape in the context of a real applica+on: bio-‐triggered drug delivery system for infec+on-‐resistant medical implants. The four objects: • Planktonic bacteria (BACF) • Adsorbed bacteria (BACB) • Bound an+bacterial agent (MOLB) • Released an+bacterial agent (MOLF)
Conclusions: Realis+c modeling and simula+on studies of complex biochemical pathways, such as glycosis, signal transduc+on, cancer, PI3K-‐Mtor etc., with millions of interac+ng par+cles is en+rely possible via GPU-‐Bioscape. • Virtual silicon-‐based experimental test-‐bed can greatly help narrow down and reduce the +me needed for costly wet-‐lab experiments and accelerate finding cure for diseases.
Concurrency model for Bacteria-‐Materials interac5ons depic5ng the objects and their A<ributes.
Example: Modeling Large-‐Scale Bacteria-‐Materials Interac8ons
Figure(Top) (a) CPU simula5on of Gillespie Algorithm for 1000 par5cles (Compare behavior with Fig (b)). (b) GPU-‐Bioscape simula5on of a total of 100,000 par5cles under same seJngs. (c) GPU-‐BioScape simula5on of a total of 500,000 par5cles. The seJngs were chosen such that a delicate balance between bacteria and AmA leads to pH oscilla5ons. (d) Higher ini5al bacterial concentra5on leads to bacterial prolifera5on.
Introduc8on and Mo8va8on • Biological systems encompass complexity that far surpasses most man-‐made systems.
• Modeling and simula+on of large and complex biological systems is computa+onally intensive.
• Need a formal modeling language and large computa+onal power in order to simulate the behavior of a complex biological system.
• We present a general language for modeling and describing biological processes, BioScape.
• The interac+on and dynamics of the biological system are implemented on GPUs in order to leverage the power of massively parallel processors.
• BioScape is based on the Stochas5c Pi-‐Calculus, and it is mo+vated by the need for individual-‐based, con+nuous mo+on and con+nuous space in modeling complex bacteria materials interac+ons in a reac+ve environment in 3D space.
• Our driving example is a bio-‐triggered drug delivery system for infec+on-‐resistant medical implants.
• The modeling and simula+on framework helps in iden+fying biological targets and materials to treat bacterial infec+ons.
Parallel and Consistent Reac8on Selec8on
Computa+onal Workgroup
Par+cle A: Par+cle B:
Challenge: How to manage and update millions of par8cles that are dynamically created and consumed in various reac8ons? Solu8on: The efficient dynamic reconfigura+on technique collects resources from inac+ve/dead par+cles and reuses them for newly created par+cles with very liale overhead.
*Bound AmA par+cles and bound Bacteria not shown
Figure(Right): Shows the concentra+on profile of par+cles in the bio-‐film, (a) Ini+ally at +me t=0, (b) Aeer +me t=100, the diffusion of AmA par+cles is triggered by pH change due to metabolizing bacteria.
(Boaom): Performance of GPU-‐Bioscape (ms/+mestep) with increasing number of par+cles.
Bioscape example [email protected], 2.0 [email protected], 0.5 [email protected], 0.2 BacF()@resBF, stepBF, shapeBF = mov.BacF() + !bind.BacB() + [email protected].(BacF() | BacF()) + [email protected].(BacF() | HIon()) + ?kill.DeadBac()
Reaction rates and radius
Movement Bind Reproduction Metabolism Destroy
ACM Conference on Bioinforma8cs, Computa8onal Biology and Biomedical Informa8cs (ACM BCB), Orlando, FL, October 7 – 10, 2012