GitHub

Plan: Parcel redesign and multi-unit multi-allele RNA

1. Current scope (single-entity trace and dwelltime)

1.1 Trace parcel (single entity)

  • unit: single unit index
  • emission: reporter sum, specific R step, G state, or small struct
  • frame_interval, start_frame, end_frame
  • Optional: active_fraction, background_mean, reporter / output index

1.2 Dwell-time parcel (single entity)

  • unit: single unit index
  • state_spec: (kind = :G, state = k) or (kind = :R, step = k) (same idea as trace)
  • sojourn_kind: :in_state | :out_of_state (replaces ON/OFF/ONG/OFFG)
  • bins: histogram bins for this dwell type
  • Optional: output index

1.3 Implementation tasks (now)

  • Define structs/named tuples for trace and dwell parcels.
  • Replace or wrap current onstates / bins / dttype with lists of these parcels for trace and dwell.
  • Simulator: loop over parcels; use parcel.unit to index state; use parcel fields for emission, statespec, sojournkind, bins, frames.
  • Likelihood / prediction: same parcel-driven logic for trace and dwell.
  • Hidden units: no parcel for that unit ⇒ no indexing in trace/dwell.
  • Leave RNA and nalleles unchanged for now.

2. Deferred: multi-unit multi-allele RNA

2.1 RNA histogram parcel (entity)

  • units: which units contribute, e.g. [1] or [1, 2]
  • nalleles: per contributing unit, e.g. 2 or [2, 1]
  • yield: per contributing unit, e.g. 1.0 or [1.0, 0.8]
  • Other RNA-specific fields (bins, nRNA, etc.) as needed.

2.2 Prediction pipeline (chemical_master.jl)

  1. Joint steady state: solve combined master equation for coupled units → P(n₁, n₂, …).
  2. Marginals: Pᵢ(nᵢ) from joint for each unit.
  3. Per-unit allele convolution: for unit i, convolve in that coordinate (kᵢ − 1) times with Pᵢ (1D convolutions per dimension).
  4. Aggregation: Y = ∑ᵢ yᵢ Sᵢ; compute P(Y = n) from the convolved joint.
  5. API: accept RNA parcel(s); from each parcel use units, nalleles, yield and run the above.

2.3 Simulator

  • For each RNA parcel: use units, nalleles, yield to define what is summed into each histogram (already compatible with current simulator idea).

2.4 Optional later: multi-unit trace/dwell

  • Source for trace/dwell: SingleUnit(unit, nalleles=1) vs AggregatedUnits(units, aggregation) if we ever need combined trace or dwell from several units.

3. Order of work (later)

  1. RNA parcel type: define struct with units, nalleles, yield (and existing RNA fields).
  2. chemical_master.jl:
    • Joint steady state for coupled units.
    • Marginals from joint.
    • Convolution loop: per unit, (kᵢ − 1) 1D convolutions with Pᵢ.
    • Aggregation: form Y and P(Y = n) from convolved joint.
  3. Wire RNA prediction: call the new pipeline from RNA parcel(s); deprecate or replace old nalleles/units wiring.
  4. Simulator and I/O: use RNA parcel(s) so multi-unit multi-allele runs match the new prediction.