Step 7 – Design Selection & Mapping

Script:

  • step_7_design_selection_and_map.py

Biological question

Among all real and simulated designs, which ones are the best candidates for experimental follow-up?

Using results from the design-space scan, we construct:

  • a design space map,
  • a ranked shortlist of top designs.

Inputs

  • parameters/design_space_scan_repression.csv
  • parameters/Hill_parameters.csv
  • parameters/half_times_upregulation.csv
  • parameters/alphamcherry.csv
  • plots/goodness_of_fit/gof_hill_fc_obs_vs_pred.pdf (to ensure we only consider well-behaved models).

Method

  1. Filter designs

    • optionally remove poor-fitting constructs or parameter sets with bad GOF,
    • focus on designs that:
      • operate in a desired dynamic-range window,
      • avoid extreme delays or unrealistic half-times.

  2. Define scoring criteria

Example components:

- dynamic range score (higher is better),
- speed score (lower t₅₀ is better),
- smoothness/overshoot penalties,
- consistency with experimentally observed behaviour.

Composite score S can be a weighted sum or multi-objective ranking.

  1. Rank designs

    • assign scores to real constructs and simulated parameter sets,
    • rank by S descending.

  2. Map design space

    • visualise designs in 2D or 3D projections, e.g.:
      • K vs n coloured by performance,
      • t₅₀ vs dynamic range,
      • delay vs half-time.

Outputs

  • plots/design_space_map.pdf – a global map of design space.
  • parameters/candidate_selection_top10.csv – table of top-ranked designs.

Each row of candidate_selection_top10.csv typically includes:

  • construct ID or synthetic design label,
  • key parameter values (K, n, half-times, delays, α),
  • performance metrics and final score.

How to interpret

This is the actionable output for wet-lab work:

  • which constructs to build/test next,
  • which parameter regimes to try to realise via new guide designs, linkers or degron variants.

It also gives a compact, visual summary of:

  • where current constructs sit in the space of all plausible CasTuner designs,
  • whether there is unused “room” for faster, cleaner, or more dynamic tuning.