Skip to content

Latest commit

 

History

History
57 lines (47 loc) · 3.22 KB

File metadata and controls

57 lines (47 loc) · 3.22 KB

Activity Rank Exploration

Goal

Increase activity rank(99%) from baseline of 24 (out of 100 metabolites).

Baseline Config

  • n_metabolites=100, n_reactions=256, n_frames=2880, delta_t=0.1
  • cycle_fraction=1.0, cycle_length=3 (all 3-cycles, A+B->2B)
  • log_k in [-2.5, -1.0], use_mass_action=true, flux_limit=false
  • n_metabolite_types=2, lambda=[0.001, 0.002], baselines=[4.0, 6.0]
  • concentration_min=1.0, concentration_max=9.0
  • Baseline activity rank(99%) = 24

Test Results (single parameter changes)

Test Change Activity Rank Delta Notes
baseline (none) 24 -- reference
1 cycle_fraction 1.0 -> 0.7 CRASH -- Random rxns have coeff=2, c^2*c^2 explodes without flux_limit
2 log_k_min -2.5 -> -3.5 7 -17 Slow reactions (k~0.0003) are inert, reducing active count
3 n_met_types 4 + lambda=[.001,.005,.01,.02] 19 -5 Stronger homeostasis dampens oscillatory modes
4 lambda 10x [0.01, 0.02] (2 types) 12 -12 Even stronger damping kills more modes
5a log_k_min -2.5 -> -1.5 CRASH -- Too many fast reactions blow up Euler
5b log_k_min -2.5 -> -2.0 50 +26 BEST. Narrower range = all reactions active
6a cycle_length 3 -> 4 11 -13 Fewer cycles (64 vs 85), fewer modes
6b cycle_length 3 -> 2 2 -22 Symmetric pairs = 1D oscillation, all correlate
7 n_reactions 256 -> 512 47 +23 More reactions = more overlapping cycles
8a n_metabolites 100 -> 200 26/200 +2 abs Dilutes reaction budget (13% vs 24% ratio)
8b n_metabolites 100 -> 50 25/50 +1 abs 50% ratio but same absolute rank
9 lambda [0.001,0.002] -> [0.0,0.0] 37 +13 Even tiny homeostasis damps 13 modes

Key Findings

Three winning parameters (each tested in isolation):

  1. log_k_min = -2.0 (rank 50): Eliminating slow reactions ensures all 256 reactions actively drive dynamics
  2. n_reactions = 512 (rank 47): More reactions = more overlapping cycles per metabolite = higher connectivity
  3. lambda = [0, 0] (rank 37): Zero homeostasis removes all linear damping

Pattern: Activity rank is controlled by the number of ACTIVELY CONTRIBUTING reactions per metabolite

  • Slow k kills reactions → rank drops (test 2)
  • Fewer reactions → rank drops (test 7 shows converse)
  • Homeostasis opposes reaction dynamics → effective reaction count drops → rank drops
  • Shorter cycles = more correlated pairs → rank drops (test 6b)
  • cycle_length=3 is near-optimal for the autocatalytic topology

Critical insight on rate constant range:

  • log_k in [-2.5, -1.0] spans 1.5 decades → ~40% of reactions have k < 0.01, contributing weakly
  • log_k in [-2.0, -1.0] spans 1.0 decade → all reactions have k >= 0.01, all contribute
  • log_k in [-1.5, -1.0] spans 0.5 decades → all very fast, but total rates overflow Euler step

Absolute rank vs metabolite count:

  • 50 mets: rank 25 (50%)
  • 100 mets: rank 24 (24%)
  • 200 mets: rank 26 (13%)
  • Absolute rank ~25 is independent of n_metabolites, suggesting it's limited by the REACTION topology (85 cycles, 256 reactions), not the metabolite pool size.

Next: Test combinations of winning parameters