ProgrammableAging

Reimplementation of Sir2-HAP Potential Landscape

Current Status

Objectives

• Use CNN to simulate stationary Fokker-Planck Equation
• Speed up and exploring high dimension.

Results and future work

The proposed method fails to simulate the FP stationary state.

• Issues

  • ModelingToolkit.jl fails to apply chain rule for the partial differentiation. Therefore, I used SymPy.jl to do the math and figure out the FP equation.
  • The script is compiled in single core - ~1 minute/step.
    • Multiprocessing is not implemented yet.
    • GPU process remains a issue 2020/10/26
  • The initial condition is not included in the equation.

• Alternatives

  • Since the steady state can be easily derived from ODE, the SDE and potential landscape can be further derived from the minimum action as described in the article below:

    • Tang, Ying, et al. "Potential landscape of high dimensional nonlinear stochastic dynamics with large noise." Scientific reports 7.1 (2017): 1-11.

---

Video 1. CNN estimated Potential landscape of Sir2-HAP model. (This is the preliminary version with compiling success but failed in using Multi-CPU and setting of initial values.) Produced by sir2-hap-stationary.jl.

Results

• Soling Fokker-Planck is computational demending (O(n^2)).
• Try A-type Integration (O(n)).

---

Reference

1. Li, Yang, et al. "A programmable fate decision landscape underlies single-cell aging in yeast." Science 369.6501 (2020): 325-329.

2. Repository: ProgrammableAging