Terrarium

Terrarium is a dynamic computer-aided process planner (CAPP) for biology designed for agile manufacturing of biological products, such as E Coli & Yeast strains, or Mammalian cell lines.

This piece of software automatically plans scientific experiments in Aquarium using historical planning data and current laboratory inventory. Data can be pulled from specific researchers to emulate how that particular researcher would plan experiments.

Requirements

• development version of trident (v0.1.0)
• Python >= 3.6
• Aquarium login credentials

Usage

Installing a specific version

pip install terrarium-capp==0.1.5

New models can be built as in the following:

from pydent import AqSession
from terrarium import AutoPlannerModel
production = AqSession("login", "pass", "url")

# pull last 300 experimental to build model
model = AutoPlannerModel(production, depth=300)
model.build()
models.save('terrarium.pkl')

Saved models can be open later:

model = AutoPlannerModel.load('terrarium.pkl')

What protocols the model uses can be adjusted using filters:

ignore_ots = production.OperationType.where({"category": ["Control Blocks", "Library Cloning"], "deployed": True})
ignore_ots += production.OperationType.where({"name": "Yeast Mating"})
ignore_ots += production.OperationType.where({"name": "Yeast Auxotrophic Plate Mating"})
ignore = [ot.id for ot in ignore_ots]
model.add_model_filter("AllowableFieldType", lambda m: m.field_type.parent_id in ignore)

Sample composition:

sample_composition = nx.DiGraph()

# build a new yeast strain from a plasmid, which is comprised of several fragments
edges = [
     ('DTBA_backboneA_splitAMP', 'pyMOD-URA-URA3.A.1-pGPD-yeVenus-tCYC1'),
     ('T1MC_NatMX-Cassette_MCT2 (JV)', 'pyMOD-URA-URA3.A.1-pGPD-yeVenus-tCYC1'),
     ('BBUT_URA3.A.0_homology1_UTP1 (from genome)', 'pyMOD-URA-URA3.A.1-pGPD-yeVenus-tCYC1'),
     ('DH5alpha', 'pyMOD-URA-URA3.A.1-pGPD-yeVenus-tCYC1'),
     ('TP-IRES-EGFP-TS', 'pyMOD-URA-URA3.A.1-pGPD-yeVenus-tCYC1' ),
     ('pyMOD-URA-URA3.A.1-pGPD-yeVenus-tCYC1', 'CEN.PK2 - MAT alpha | his-pGRR-W5-W8-RGR-W36'),
]

for n1, n2 in edges:
    s1 = browser.find_by_name(n1)
    s2 = browser.find_by_name(n2)
    sample_composition.add_node(s1.id, sample=s1)
    sample_composition.add_node(s2.id, sample=s2)
    sample_composition.add_edge(s1.id, s2.id)

ignore_items = []  # optional to not include certain items in the search.
desired_object_type = production.ObjectType.find_by_name('Fragment Stock')
cost, paths, graph = network.run(desired_object_type, ignore=ignore_items)

# make a new plan
canvas = Planner(production)

# add protocols from optimized network to plan
network.plan(paths, graph, canvas)

# submit to Aquarium
canvas.create()

Example of Planning Yeast Construction

Probability Matrix of Connecting Aquarium Protocols

The autoplanner uses this type of data, in concert with the sample_composition network, to build an optimal experiment.

Top 50 Connections

Model Factory

factory = ModelFactory(session)

# make a model from a single user
model1 = factory.emulate('user1').build()

# make a model from a group of users
user_group = ['user2', 'user3']
model2 = factory.emulate(user_group).build()

# make a model from the last 100 plans
model3 = factory.new(100).build()

# compose a weighted model
model = model1 + model2 * 3

Future Version

• estimate convidence for certain inventory items or operations based on past success rate
• better api for
• using 'ghost' plans to build model
• emulating specific users / user groups ** faster execution (currently ~45-60 seconds)

License

Feb. 4, 2019 - This software is not currently licensed. The author (Justin D. Vrana of University of Washington) does not grant permission to copy or modify code base.