biosimulant 0.0.10

Open-source local simulation runtime and CLI for Biosimulant

biosimulant

Composable simulation runtime + UI layer for orchestrating runnable biomodules.

biosimulant is the primary package and CLI name. The existing biosim Python import path and python -m biosim command remain supported for existing model packages during the migration.

---

Executive Summary & System Goals

Vision

Provide a small, stable composition layer for simulations: wire reusable components ("biomodules") into a BioWorld, run them with a single orchestration contract, and visualize/debug runs via a lightweight web UI (SimUI). Biomodules are self-contained Python packages that can wrap external simulators internally (SBML/NeuroML/CellML/etc.) without a separate adapter layer.

Core Mission

• Compose simulations from reusable, interoperable biomodules.
• Make "run + visualize + share a config" the default workflow (local-first; hosted later).
• Keep the runtime small and predictable while letting biomodules embed their own simulator/tooling.

Primary Users

• Developers and researchers who need composable simulation workflows and fast iteration.
• Near-term beachhead: neuroscience demos (single neuron + small E/I microcircuits) with strong visuals and reproducible configs.

---

Installation

Preferred (pinned GitHub ref):

pip install "biosimulant @ git+https://github.com/<org>/biosim.git@<ref>"

Alternative (package index):

pip install biosimulant

For the shared ONNX biomodule helpers:

pip install "biosimulant[ml]"

Compatibility and command ownership

The biosimulant package still ships the biosim Python import path so existing model packages keep working:

import biosim

Use biosimulant for new CLI examples:

biosimulant --help
python -m biosimulant --help

python -m biosim remains available as a compatibility command. If a machine also has the Desktop/product CLI installed, PATH decides which biosimulant binary runs. Use python -m biosimulant ... to force the Python package CLI. The Python package owns local open-source workflows; Desktop/product extensions own Hub, auth, cloud, app state, and managed-service workflows.

Publishing to PyPI

See the release guide: docs/releasing.md.

Packaging Models And Labs

biosim can package one model or one lab into a single archive for portability, upload, caching, and validation.

Common commands:

# Build a package from a directory that contains model.yaml or lab.yaml
biosimulant pack build path/to/model-or-lab

# Validate an existing package file
biosimulant pack validate dist/local__counter-1.0.0.bsimodel

# Build a self-contained lab package (.bsilab)
biosimulant pack build path/to/lab

Notes:

• build prefers package: and version: from model.yaml or lab.yaml when present.
• model dependencies in manifests must use exact == pins.
• lab builds are always self-contained and preserve the full runnable source tree inside the .bsilab.
• nested lab dependencies must use relative path refs and must already exist inside the packaged lab directory.
• validate prints human-readable success or failure output by default; add --json for machine-readable output.

See docs/packaging.md for the full package layout, recommended authoring flow, and CLI examples.

Provisional Runtime Helpers

biosim.runtime is the provisional public home for package interpretation helpers shared by the open-source CLI and Biosimulant platform executors. It owns entrypoint loading, typed runtime.initial_inputs coercion, communication-step resolution, and source-neutral lab flattening. Import these helpers from biosim.runtime; they are not exported from top-level biosim while the API settles.

BioModule Convenience Layers

BioModule remains the minimal full-control runtime contract. For common model adapters, biosim also exports opt-in helpers:

• SignalEmitterBioModule: output storage, source-name resolution, and raw value to typed BioSignal wrapping.
• StatefulBioModule: fixed-step window advancement, input override storage, bounded history, and output publishing hooks.

Signal helper functions are available from biosim.signals and top-level biosim: unwrap_payload, coerce_float, scalar_or_record_input, and make_signal.

Examples

• See examples/ for quick-start scripts. Try:

pip install -e .
python examples/basic_usage.py

For advanced curated demos (neuro/ecology), wiring configs, and model-pack templates, see the companion repo:

• https://github.com/Biosimulant/models

Quick Start: BioWorld

Minimal usage:

import biosim
from biosim import ScalarSignal, SignalSpec


class Counter(biosim.BioModule):
    def __init__(self):
        self.value = 0
        self._t = 0.0

    def outputs(self):
        return {"count": SignalSpec.scalar(dtype="int64", emitted_unit="1")}

    def advance_window(self, start: float, end: float) -> None:
        _ = start
        self.value += 1
        self._t = end

    def get_outputs(self):
        return {
            "count": ScalarSignal(
                source="counter",
                name="count",
                value=self.value,
                emitted_at=self._t,
                spec=self.outputs()["count"],
            )
        }

    def snapshot(self) -> dict:
        return {"value": self.value, "t": self._t}

    def restore(self, snapshot: dict) -> None:
        self.value = int(snapshot.get("value", 0))
        self._t = float(snapshot.get("t", 0.0))


world = biosim.BioWorld(communication_step=0.1)
world.add_biomodule("counter", Counter())
world.run(duration=1.0)

Outputs produced during a communication window are committed at the end of that window and become visible to downstream modules on a later communication turn. For final report, export, or visualisation modules in workflow-style graphs, call world.settle(steps=1) after world.run(...) to propagate final outputs without advancing simulated time.

Visuals from Modules

Modules may optionally expose visuals via visualize(), returning a dict or list of dicts with keys render and data. The world can collect them without any transport layer:

class MyModule(biosim.BioModule):
    def advance_window(self, start: float, end: float) -> None:
        _ = start, end

    def get_outputs(self):
        return {}

    def snapshot(self) -> dict:
        return {}

    def restore(self, snapshot: dict) -> None:
        _ = snapshot

    def visualize(self):
        return {
            "render": "timeseries",
            "data": {"series": [{"name": "s", "points": [[0.0, 1.0]]}]},
        }

world = biosim.BioWorld(communication_step=0.1)
world.add_biomodule("module", MyModule())
world.run(duration=0.1)
print(world.collect_visuals())  # [{"module": "module", "visuals": [...]}]

If visuals are generated by a separate downstream module wired to another producer's final outputs, run one or more settle turns before collecting visuals: world.run(duration=...); world.settle(1); world.collect_visuals().

See examples/visuals_demo.py for a minimal end-to-end example.

ONNX Modules

biosim can host ONNX-backed modules without changing the core runtime. Install the ML extras and wrap the ONNX model behind the standard BioModule interface:

from biosim import OnnxClassifierModule, ScalarSignal, SignalSpec

classifier = OnnxClassifierModule(
    model_path="artifacts/model.onnx",
    class_labels=["quiescent", "subthreshold", "spiking"],
    input_port="state_vector",
    probabilities_port="state_probabilities",
    predicted_port="predicted_state",
    input_vector_length=4,
)

classifier.set_inputs(
    {
        "state_vector": ScalarSignal(
            source="adapter",
            name="state_vector",
            value=-64.0,
            emitted_at=0.0,
            spec=SignalSpec.scalar(dtype="float64"),
        )
    }
)
classifier.advance_window(0.0, 0.001)
print(classifier.get_outputs()["predicted_state"].value)

Model packs can subclass OnnxClassifierModule to set model-relative model_path, port names, and label sets while keeping the inference logic in the shared library.

SimUI (Python-Declared UI)

SimUI lets you build and launch a small web UI entirely from Python (similar to Gradio's ergonomics), backed by FastAPI and a prebuilt React SPA that renders visuals from JSON. The frontend uses Server-Sent Events (SSE) for real-time updates.

• User usage (no Node/npm required):

  • Install UI extras: pip install 'biosimulant[ui]'

  • Try the demo: python examples/ui_demo.py then open http://127.0.0.1:7860/ui/.

  • From your own code:

    from biosim.simui import Interface, Number, Button, EventLog, VisualsPanel
    world = biosim.BioWorld(communication_step=0.1)
    ui = Interface(
        world,
        controls=[Number("duration", 10), Button("Run")],
        outputs=[EventLog(), VisualsPanel()],
    )
    ui.launch()
    

  • The UI provides endpoints under /ui/api/...:

    • GET /api/spec – UI layout (controls, outputs, modules)
    • POST /api/run – Start a simulation run
    • GET /api/status – Runner status (running/paused/error + optional progress fields)
    • GET /api/state – Full state (status + last step + modules)
    • GET /api/events – Buffered world events (?since_id=&limit=)
    • GET /api/visuals – Collected module visuals
    • GET /api/snapshot – Full snapshot (status + visuals + events)
    • GET /api/stream – SSE endpoint for real-time event streaming
    • POST /api/pause – Pause running simulation
    • POST /api/resume – Resume paused simulation
    • POST /api/reset – Stop, reset, and clear buffers
    • Editor sub-API (/api/editor/...): visual config editor for loading, saving, validating, and applying YAML wiring configs as node graphs. Endpoints include modules, current, config, apply, validate, layout, to-yaml, from-yaml, and files.

Per-run resets for clean visuals

• On each Run, the backend clears its event buffer and calls reset() on modules if they implement it.

• The frontend clears visuals/events before posting /api/run.

• To avoid overlapping charts across runs, add reset() to modules that accumulate history (e.g., time series points).

• Maintainer flow (building the frontend SPA):

  • Edit the React/Vite app under src/biosim/simui/_frontend/.
  • Build via Python: python -m biosim.simui.build (requires Node/npm). This writes src/biosim/simui/static/app.js.
  • Alternatively: bash scripts/build_simui_frontend.sh.
  • Packaging includes src/biosim/simui/static/**, so end users never need npm.

• CI packaging (recommended): run the frontend build before python -m build so wheels/sdists ship the bundled assets.

Troubleshooting:

• If you see SimUI static bundle missing at .../static/app.js, build the frontend with python -m biosim.simui.build (requires Node/npm) before launching. End users installing a release wheel won't see this.

SimUI Design Notes

• Transport: SSE (Server-Sent Events). The SPA connects to /api/stream for real-time updates. Polling endpoints (/api/status, /api/visuals, /api/events) remain available for fallback/debugging.
• Objective progress fields are based on simulation-time progress ((sim_time - sim_start) / duration), not wall-clock time.
• /api/status may include: sim_time, sim_start, sim_end, sim_remaining, progress, progress_pct (all optional/additive).
• Events API: /api/events?since_id=<int>&limit=<int> returns { events, next_since_id } where events are appended world events and next_since_id is the cursor for subsequent calls.
• VisualSpec types supported now:
  • timeseries: data = { "series": [{ "name": str, "points": [[x, y], ...] }, ...] }
  • bar: data = { "items": [{ "label": str, "value": number }, ...] }
  • scatter: data = { "points": [{ "x": number, "y": number, "label"?: str, "series"?: str }, ...] }
  • heatmap: data = { "values": [[number, ...], ...], "x_labels"?: [str, ...], "y_labels"?: [str, ...] }
  • table: data = { "columns": [..], "rows": [[..], ...] } or data = { "items": [{...}, ...] }
  • image: data = { "src": str, "alt"?: str, "width"?: number, "height"?: number }
  • graph: simple node-edge graph renderer
  • structure3d: data = { "title"?: str, "source": { "kind": "url", "url": str } | { "kind": "artifact", "artifact_id": str }, "format": "mmcif" | "pdb", "annotations"?: [{ "label": str, "value": str|number|bool }], "initial_view"?: {...} }
• VisualSpec may also include an optional description (string) for hover text or captions.
• SimUI serves artifact-backed structure3d files through /api/artifacts/{artifact_id} so browser clients do not receive raw local filesystem paths.

Terminology

Understanding the core concepts is essential for working with biosim effectively.

Term	Description
BioWorld	Runtime container that orchestrates multi-rate biomodules, routes signals, and publishes lifecycle events.
BioModule	Pluggable unit of behavior with local state. Implements the runnable contract (setup/reset/advance_to/...).
BioSignal	Typed, versioned data payload exchanged between modules via named ports.
WorldEvent	Runtime events emitted by the BioWorld (STARTED, TICK, FINISHED, etc.).
Wiring	Module connection graph. Defined programmatically, via WiringBuilder, or loaded from YAML/TOML configs.
VisualSpec	JSON structure returned by module.visualize() with render type and data payload.

Event Lifecycle

Every simulation follows this sequence:

STARTED -> TICK (xN) -> FINISHED

PAUSED, RESUMED, STOPPED, and ERROR may also be emitted depending on runtime control flow.

License

MIT. See LICENSE.txt.