jinko-sdk 1.5.1rc1

Official typed Python SDK for the Jinkō clinical trial simulation platform (jinko.ai) — script models, trials, virtual populations, and results.

Jinkō Python SDK

Jinkō is a complete solution for clinical trial simulation and protocol design optimization, developed by Nova In Silico. It combines mechanistic ("white-box") modeling, virtual populations, in-silico trials, and analytics in a collaborative platform used by modelers, scientists, and trial managers to accelerate drug development and de-risk clinical decisions.

The Jinkō Python SDK is the official, typed Python client for the Jinkō API. It lets you script, automate, and integrate Jinkō workflows — from browsing project items and editing computational models, to running trials on virtual populations and retrieving simulation results — directly from Python.

Main features

• QSP model development — Create, edit, and version quantitative systems pharmacology models with typed APIs for parameters, events, reactions, and compartments
• Virtual population management — Work with patient populations and generators for trial simulation at scale
• In-silico trial orchestration — Design protocols, run simulations, and analyze outcomes to optimize trial designs before clinical execution
• Collaborative project navigation — Browse folders, search across model libraries, and manage versioned assets in team environments
• Results analytics — Extract simulation summaries, tabular data, and visualizations with optional pandas integration
• Programmatic workflows — Automate repetitive modeling tasks, batch operations, and integrate with existing R&D pipelines

Learn more about the platform at doc.jinko.ai.

Installation

pip install jinko-sdk

Requirements:

• Python 3.11+
• A Jinko API key
• A target Jinko project id

Setup and authentication

The SDK reads configuration from environment variables by default:

export JINKO_API_KEY="..."
export JINKO_PROJECT_ID="..."
export JINKO_BASE_URL="https://api.jinko.ai"  # optional

For local developer convenience, keep secrets in a local .env and load them in your shell (for example with direnv allow if you use direnv).

You can also pass values explicitly:

from jinko import JinkoClient

client = JinkoClient(
    api_key="...",
    project_id="...",
    base_url="https://api.jinko.ai",  # optional
    timeout=30.0,
)

Validate credentials early:

check = client.auth_check()
print(check.status, check.api_key.role)

Quickstart

from jinko import JinkoClient

client = JinkoClient()

model = client.get_model("cm-...")
print(model.name)

model.rename("Retuned PK model")
model.components.get_parameter("k_clearance").set_formula("CL / V")

trial = client.create_trial(model, name=f"{model.name} - smoke run")
trial.run()
trial.wait_until_completed(timeout=600, poll_interval=5.0)

summary = trial.results.summary()
print(summary.get("status"))

Core concepts

• JinkoClient: primary user-facing entrypoint
• Client direct methods: client.get_model(...), client.list_trials(...), client.create_trial(...), ...
• Domain wrappers (Model, Trial, Vpop, ...): typed objects with behavior methods (for example model.rename(...), trial.run())
• types.ProjectItemMetadata: common typed metadata envelope (sid, type, core_id, folders, version, ...)
• Page[T]: paginated list() result (items, next_cursor, has_next)
• iter(): auto-paginated iterator for bulk traversal

Most user-facing resources follow this pattern:

• list_<types>(...) -> Page[T]
• iter_<types>(...) -> Iterator[T]
• get_<type>(sid, revision=None)
• create_<type>(...) or create_raw_<type>(...) when available
• delete(sid)
• rich object methods on returned items (.versions, rename, set_description, run, wait_until_completed, ...)

The intended SDK usage is:

• enter through client.<direct_method>(...)
• continue through the typed object returned by that method
• use object-level services only when explicitly exposed (currently model.components)

Folder handling follows a single-folder public interface:

• public filters and create methods accept folder= with either a Folder object or folder id string
• project-item wrappers expose .folder for the common single-folder case
• .folders remains available for full API compatibility when an existing item belongs to multiple folders
• use move_to_folder(folder) as the public folder mutator to place an item in one folder, and move_to_folder(None) to clear folder assignments

Public client surface

Project-wide services

• client.folders: folder CRUD and folder-tree exploration
• client.raw_request(...): authenticated low-level HTTP escape hatch
• client.delete(sid): delete any project item by SID

Typed project item methods

• client.list_models(...), client.iter_models(...), client.get_model(...)
• client.list_trials(...), client.iter_trials(...), client.get_trial(...)
• client.list_calibrations(...), client.iter_calibrations(...), client.get_calibration(...)
• client.list_simple_output_sets(...), client.iter_simple_output_sets(...), client.get_simple_output_set(...)
• client.list_protocol_designs(...), client.iter_protocol_designs(...), client.get_protocol_design(...)
• client.list_advanced_output_sets(...), client.iter_advanced_output_sets(...), client.get_advanced_output_set(...)
• client.list_vpops(...), client.iter_vpops(...), client.get_vpop(...)
• client.list_vpop_designs(...), client.iter_vpop_designs(...), client.get_vpop_design(...)
• client.list_extracts(...), client.iter_extracts(...), client.get_extract(...)
• client.list_data_tables(...), client.iter_data_tables(...), client.get_data_table(...)
• client.list_documents(...), client.iter_documents(...), client.get_document(...)
• client.list_raw_files(...), client.iter_raw_files(...), client.get_raw_file(...)
• client.list_references(...), client.iter_references(...), client.get_reference(...)
• client.list_subsampling_designs(...), client.iter_subsampling_designs(...), client.get_subsampling_design(...)
• client.list_trial_visualizations(...), client.iter_trial_visualizations(...), client.get_trial_visualization(...)

Exploring a project

1) Start broad: project-items search

models_page = client.list_models(name="PK")
for model in models_page:
    print(model.sid, model.type, model.name)

Use iter() when you want all pages:

all_trial_sids = [trial.sid for trial in client.iter_trials()]

2) Explore folders and folder trees

print(client.folder_tree())

root = client.get_folder_by_name("Program A", exact_match_only=True)
if root:
    print(root.tree(max_depth=2, include_project_items=True))
    reports = root.find_child("Reports", exact_match_only=True)
    children = root.children()
    tree_data = root.tree_dict(include_project_items=False)
    tree_nodes = root.tree_nodes(include_project_items=False)

For a project-wide structured forest:

tree_nodes = client.folder_tree_nodes()
tree_data = client.folder_tree_dict()

You can also list by folder:

modeling_folder = client.get_folder_by_name("Modeling", exact_match_only=True)
if modeling_folder:
    models = client.list_models(folder=modeling_folder, limit=25)

3) Fetch typed resources and inspect content

model = client.get_model("cm-...")
content = model.content()  # typed model interface
print(content.model.modelName)

folder = model.folder
all_api_folders = model.folders

Versioned resources

By default, get(sid) reads the latest revision.

latest = client.get_model("cm-...")
older = client.get_model("cm-...", revision=3)

List version metadata and label snapshots:

versions_page = latest.versions.list(only_labeled=False, first=20)
for version in versions_page:
    print(version.revision, version.label, version.is_latest)

latest.versions.label(revision=3, label="baseline")
latest.versions.unlabel(revision=3)

Models and components

Model.components is the ergonomic typed API for editing components.

model = client.get_model("cm-...")

# Typed read
k_clearance = model.components.get_parameter("k_clearance")

# Immediate multi-field update in one API call
k_clearance.update(formula="CL / V", unit="L/h", description="retuned clearance")

# Typed create (immediate commit)
model.components.create_parameter(id="k_abs", formula=1.2, unit="1/h")

# Event updates are provided as a component->value mapping
model.components.create_event(
    id="dose_start",
    updates={"Dose": 100},
    condition_trigger="t >= 0",
)

# Reaction stoichiometry is also a mapping
model.components.create_mass_action_reaction(
    id="binding",
    reactants={"Drug": 1, "Target": 1},
    products={"Complex": 1},
    k_plus="kon",
    k_minus="koff",
)

When changing several components, batch commits avoid one-request-per-change:

with model.components.batch(version="retune") as batch:
    batch.edit_parameter("k_clearance").set_formula("CL2 / V")
    batch.create_parameter(id="k_new", formula=0.8, unit="1/h")

with model.components.batch(version="event and kinetics") as batch:
    batch.edit_event("dose_start").set_updates({"Dose": 120})
    batch.edit_reaction("binding").set_general_kinetics(
        reactants={"Drug": 1, "Target": 1},
        products={"Complex": 1},
        rate="kon * Drug * Target - koff * Complex",
    )

For unsupported modeling endpoints, use client.raw_request(...) rather than private object internals.

Trials and result retrieval

Create a trial from typed resources, run it, then consume results:

model = client.get_model("cm-...")
vpop = client.get_vpop("vp-...")
protocol = client.get_protocol_design("pd-...")

trial = client.create_trial(
    model,
    vpop=vpop,
    protocol=protocol,
    name="model-vpop-protocol run",
)
trial.run()
trial.wait_until_completed(timeout=1800)

summary = trial.results.summary()
scalars_csv = trial.results.scalars(["AUC", "Cmax"])  # TabularDownload

# Optional convenience if pandas is installed in your environment
df = scalars_csv.to_dataframe()
print(df.head())

Pagination patterns

list() returns a Page[T] with cursor metadata:

page = client.list_models(limit=20)
print(len(page), page.has_next, page.next_cursor)

if page.has_next:
    next_page = client.list_models(limit=20, after=page.next_cursor)

Use iter() when you want to consume all pages seamlessly.

Raw API escape hatch

Use client.raw_request(...) when an endpoint is not yet wrapped by a typed service. It reuses SDK authentication, project scoping, transport, and error handling.

payload = client.raw_request(
    "GET",
    "/app/v1/auth/check",
)
print(payload)

Example with params + JSON body:

result = client.raw_request(
    "POST",
    "/app/v1/project-item",
    params={"first": 10},
    json_body={"text": "tumor", "type": "Trial"},
    headers={"X-My-Header": "value"},
)

Use this path for uncovered routes. Prefer typed SDK methods when available.

Error handling

The SDK raises typed exceptions from jinko, including:

• ConfigurationError
• AuthenticationError, AuthorizationError
• NotFoundError, ValidationError, ConflictError
• RateLimitError, ServerError, TransportError

Minimal example:

from jinko import JinkoClient, NotFoundError

client = JinkoClient()

try:
    client.get_model("cm-does-not-exist")
except NotFoundError as exc:
    print(f"Model not found: {exc}")

Development tests

See tests/README.md.

License

This project is licensed under the MIT License. See the LICENSE file for details.

Contact & references

For support or inquiries, please contact us at [e-mail]

• Jinkō
• Nova In Silico