Introduction#

LaminDB is an open-source Python framework to manage biological data & analyses in generic backends:

Access data & metadata across storage (files, arrays) & database (SQL) backends.
Track data flow across notebooks, pipelines & UI.
Manage registries for experimental metadata & in-house ontologies, import public ontologies.
Validate, standardize & annotate data using registries.
Organize and share data across a mesh of LaminDB instances.
Manage data access with an auditable system of record.

LaminHub is a data collaboration hub built on LaminDB similar to how GitHub is built on git.

Basic features of LaminHub are free. Enterprise features hosted in your or our infrastructure are available on a paid plan!

Quickstart#

Warning

Public beta: Close to having converged a stable API, but some breaking changes might still occur.

Setup LaminDB#

Install the lamindb Python package:
```
pip install 'lamindb[jupyter,bionty]'
```
Sign up for a free account (see more info) and copy the API key.
Log in on the command line (data remains in your infrastructure, with Lamin having no access to it):
```
lamin login <email> --key <API-key>
```

You can now init LaminDB instances like you init git repositories:

!lamin init --schema bionty --storage ./lamin-intro  # or s3://my-bucket, gs://my-bucket as default storage

Because we passed --schema bionty, this instance mounted plug-in lnschema_bionty.

Register a file#

Track files using the File registry:

import lamindb as ln
import pandas as pd

# track run context
ln.track()

# access a batch of data
df = pd.DataFrame(
    {"CD8A": [1, 2, 3], "CD4": [3, 4, 5], "CD14": [5, 6, 7]},
    index=["observation1", "observation2", "observation3"],
)

# create a file (versioning is optional)
file = ln.File(df, description="my RNA-seq", version="1")

# register file
file.save()

💡 lamindb instance: testuser1/lamin-intro

💡 notebook imports: lamindb==0.63.4 lnschema_bionty==0.35.3 pandas==1.5.3

💡 saved: Transform(uid='FPnfDtJz8qbEz8', name='Introduction', short_name='introduction', version='0', type=notebook, updated_at=2023-12-08 11:33:05 UTC, created_by_id=1)

💡 saved: Run(uid='BKP3qYT8hwel1W7rxXQP', run_at=2023-12-08 11:33:05 UTC, transform_id=1, created_by_id=1)

Access a file#

# search a file
ln.File.search("RNAseq")

# filter a file
file = ln.File.filter(description__contains="RNA-seq").first()

# view data flow
file.view_flow()

# describe metadata
file.describe()

# load the file
df = file.load()

_images/ca6c8e97434e872b940c853948198b2d1f8b2549ef642766bbe252ec9ec39a75.svg

File(uid='6yMEbv1eZs1t6YaijPcZ', suffix='.parquet', accessor='DataFrame', description='my RNA-seq', version='1', size=3506, hash='I3bJ9fOfamH1pAcJ4kydIg', hash_type='md5', visibility=1, key_is_virtual=True, updated_at=2023-12-08 11:33:05 UTC)

Provenance:
  🗃️ storage: Storage(uid='aXvyR27m', root='/home/runner/work/lamindb/lamindb/docs/lamin-intro', type='local', updated_at=2023-12-08 11:33:02 UTC, created_by_id=1)
  📔 transform: Transform(uid='FPnfDtJz8qbEz8', name='Introduction', short_name='introduction', version='0', type='notebook', updated_at=2023-12-08 11:33:05 UTC, created_by_id=1)
  👣 run: Run(uid='BKP3qYT8hwel1W7rxXQP', run_at=2023-12-08 11:33:05 UTC, transform_id=1, created_by_id=1)
  👤 created_by: User(uid='DzTjkKse', handle='testuser1', name='Test User1', updated_at=2023-12-08 11:33:02 UTC)

Define features & labels#

Define features and labels using Feature and ULabel:

# define features
features = ln.Feature.from_df(df)
ln.save(features)

# define tissue label
tissue = ln.ULabel(name="umbilical blood")
tissue.save()

# define a parent label
is_tissue = ln.ULabel(name="is_tissue")
is_tissue.save()
is_tissue.children.add(tissue)

# view hierarchy
tissue.view_parents()

_images/61c47e6f6f33035e27b5816e26e3a6110ec9ad03d17b37d29fa2165c9d8f4000.svg

Validate & annotate data#

# create file & validate features
file = ln.File.from_df(df, description="my RNA-seq")

# register file & link validated features
file.save()

# annotate with a label
file.labels.add(tissue)

# show metadata
file.describe()

❗ returning existing file with same hash: File(uid='6yMEbv1eZs1t6YaijPcZ', suffix='.parquet', accessor='DataFrame', description='my RNA-seq', version='1', size=3506, hash='I3bJ9fOfamH1pAcJ4kydIg', hash_type='md5', visibility=1, key_is_virtual=True, updated_at=2023-12-08 11:33:05 UTC, storage_id=1, transform_id=1, run_id=1, created_by_id=1)

File(uid='6yMEbv1eZs1t6YaijPcZ', suffix='.parquet', accessor='DataFrame', description='my RNA-seq', version='1', size=3506, hash='I3bJ9fOfamH1pAcJ4kydIg', hash_type='md5', visibility=1, key_is_virtual=True, updated_at=2023-12-08 11:33:05 UTC)

Provenance:
  🗃️ storage: Storage(uid='aXvyR27m', root='/home/runner/work/lamindb/lamindb/docs/lamin-intro', type='local', updated_at=2023-12-08 11:33:02 UTC, created_by_id=1)
  📔 transform: Transform(uid='FPnfDtJz8qbEz8', name='Introduction', short_name='introduction', version='0', type='notebook', updated_at=2023-12-08 11:33:05 UTC, created_by_id=1)
  👣 run: Run(uid='BKP3qYT8hwel1W7rxXQP', run_at=2023-12-08 11:33:05 UTC, transform_id=1, created_by_id=1)
  👤 created_by: User(uid='DzTjkKse', handle='testuser1', name='Test User1', updated_at=2023-12-08 11:33:02 UTC)
Features:
  columns: FeatureSet(uid='WTgR8NcuFs0q5wJj9iXK', n=3, registry='core.Feature', hash='UnW_pPBmoCTHePSnwWUB', updated_at=2023-12-08 11:33:05 UTC, created_by_id=1)
    CD8A (number)
    CD4 (number)
    CD14 (number)
Labels:
  🏷️ ulabels (1, core.ULabel): 'umbilical blood'

Query for annotations#

# a look-up object for all the children of "is_tissue" in ULabel registry
tissues = is_tissue.children.lookup()

# query for exactly one result annotated with umbilical blood
dataset = ln.File.filter(ulabels=tissues.umbilical_blood).one()

# permanently delete the file (without the permanent flag, moves to trash)
file.delete(permanent=True)

Use biological types#

The generic Feature and ULabel will get you pretty far.

But if you use an entity many times, you typically want a dedicated registry, which you can use to type your code & as an interface for public ontologies.

Let’s do this with Gene and Tissue from plug-in lnschema_bionty:

import lnschema_bionty as lb

# create gene records from the public ontology as features
genes = lb.Gene.from_values(df.columns, organism="human")
ln.save(genes)

# query the entire Gene registry content as a DataFrame
lb.Gene.filter().df()

# create file & validate features using the symbol field of Gene
file = ln.File.from_df(
    df, description="my RNA-seq", field=lb.Gene.symbol, organism="human"
)
file.save()

# search the public tissue ontology from the bionty store
lb.Tissue.bionty().search("umbilical blood").head(2)

# define tissue label
tissue = lb.Tissue.from_bionty(name="umbilical cord blood")
tissue.save()

# ontological hierarchy comes by default
tissue.view_parents(distance=2)

# annotate with tissue label
file.labels.add(tissue)

# show metadata
file.describe()

❗ now recursing through parents: this only happens once, but is much slower than bulk saving

_images/49df3a36b7aea7fe3288a885d24b7286bc1d0d83904e54332164e980245d5fad.svg

File(uid='lDyYMTRQXd2N5vJ3gzZ8', suffix='.parquet', accessor='DataFrame', description='my RNA-seq', size=3506, hash='I3bJ9fOfamH1pAcJ4kydIg', hash_type='md5', visibility=1, key_is_virtual=True, updated_at=2023-12-08 11:33:10 UTC)

Provenance:
  🗃️ storage: Storage(uid='aXvyR27m', root='/home/runner/work/lamindb/lamindb/docs/lamin-intro', type='local', updated_at=2023-12-08 11:33:02 UTC, created_by_id=1)
  💫 transform: Transform(uid='FPnfDtJz8qbEz8', name='Introduction', short_name='introduction', version='0', type=notebook, updated_at=2023-12-08 11:33:05 UTC, created_by_id=1)
  👣 run: Run(uid='BKP3qYT8hwel1W7rxXQP', run_at=2023-12-08 11:33:05 UTC, transform_id=1, created_by_id=1)
  👤 created_by: User(uid='DzTjkKse', handle='testuser1', name='Test User1', updated_at=2023-12-08 11:33:02 UTC)
Features:
  columns: FeatureSet(uid='lOYmKviSYNrmjDRT2sIs', n=3, type='number', registry='bionty.Gene', hash='7wsvCyRhtmkNeD2dpTHg', updated_at=2023-12-08 11:33:10 UTC, created_by_id=1)
    'CD8A', 'CD4', 'CD14'
Labels:
  🏷️ tissues (1, bionty.Tissue): 'umbilical cord blood'

Query for gene sets & the linked files:

# an object to auto-complete human genes
genes = lb.Gene.filter(organism__name="human").lookup()

# all gene sets measuring CD8A
genesets_with_cd8a = ln.FeatureSet.filter(genes=genes.cd8a).all()

# all files measuring CD8A
ln.File.filter(feature_sets__in=genesets_with_cd8a).df()

	uid	storage_id	key	suffix	accessor	description	version	size	hash	hash_type	transform_id	run_id	initial_version_id	visibility	key_is_virtual	updated_at	created_by_id
id
2	lDyYMTRQXd2N5vJ3gzZ8	1	None	.parquet	DataFrame	my RNA-seq	None	3506	I3bJ9fOfamH1pAcJ4kydIg	md5	1	1	None	1	True	2023-12-08 11:33:10.423872+00:00	1

Append a new batch of data#

# assume we now run a pipeline in which we access a new batch of data
transform = ln.Transform(name="RNA-seq file ingestion", type="pipeline", version="1")
ln.track(transform)

# access a new batch of data with a different schema
df = pd.DataFrame(
    {
        "CD8A": [2, 3, 3],
        "CD4": [3, 4, 5],
        "CD38": [4, 2, 3],
    },
    index=["observation4", "observation5", "observation6"],
)

# because gene `"CD38"` is not yet registered, it doesn't yet validate
file2 = ln.File.from_df(
    df, description="my RNA-seq batch 2", field=lb.Gene.symbol, organism="human"
)

# let's add it to the `Gene` registry and re-create the file - now everything passes
lb.Gene.from_bionty(symbol="CD38", organism="human").save()

# now we can validate all features
file2 = ln.File.from_df(
    df, description="my RNA-seq batch 2", field=lb.Gene.symbol, organism="human"
)
file2.save()

💡 saved: Transform(uid='VKrFIeI5Edud27', name='RNA-seq file ingestion', version='1', type='pipeline', updated_at=2023-12-08 11:33:21 UTC, created_by_id=1)

💡 saved: Run(uid='EswO863J8aKbv7pbQSbG', run_at=2023-12-08 11:33:21 UTC, transform_id=2, created_by_id=1)

❗ 1 term (33.30%) is not validated for symbol: CD38

Create a dataset using Dataset by linking both batches in a “sharded dataset”:

dataset = ln.Dataset([file, file2], name="my RNA-seq dataset")
dataset.save()
dataset.describe()
dataset.view_flow()

Dataset(uid='gzlsEnucixYv1r2cEWMR', name='my RNA-seq dataset', hash='Kv5biKHSYT0rrCcw3TKG', visibility=1, updated_at=2023-12-08 11:33:24 UTC)

Provenance:
  🧩 transform: Transform(uid='VKrFIeI5Edud27', name='RNA-seq file ingestion', version='1', type='pipeline', updated_at=2023-12-08 11:33:21 UTC, created_by_id=1)
  👣 run: Run(uid='EswO863J8aKbv7pbQSbG', run_at=2023-12-08 11:33:21 UTC, transform_id=2, created_by_id=1)
  👤 created_by: User(uid='DzTjkKse', handle='testuser1', name='Test User1', updated_at=2023-12-08 11:33:02 UTC)
Features:
  columns: FeatureSet(uid='HaSQni4Sx75mM1c8wkPC', n=4, type='number', registry='bionty.Gene', hash='OaTsWN-lR7zDUC1bjzk8', updated_at=2023-12-08 11:33:24 UTC, created_by_id=1)
    'CD8A', 'CD4', 'CD14', 'CD38'

_images/a78b39b8721b0a7030be91b9eebc2fbdaec9b4e7c94805c88e6457d923072858.svg

You can load the entire dataset into memory as if it was one:

dataset.load()

	CD8A	CD4	CD14	CD38
observation1	1	3	5.0	NaN
observation2	2	4	6.0	NaN
observation3	3	5	7.0	NaN
observation4	2	3	NaN	4.0
observation5	3	4	NaN	2.0
observation6	3	5	NaN	3.0

Or iterate over its files:

dataset.files.df()

	uid	storage_id	key	suffix	accessor	description	version	size	hash	hash_type	transform_id	run_id	initial_version_id	visibility	key_is_virtual	updated_at	created_by_id
id
2	lDyYMTRQXd2N5vJ3gzZ8	1	None	.parquet	DataFrame	my RNA-seq	None	3506	I3bJ9fOfamH1pAcJ4kydIg	md5	1	1	None	1	True	2023-12-08 11:33:10.423872+00:00	1
3	RmDjsiSmdfzUeIkLvCIi	1	None	.parquet	DataFrame	my RNA-seq batch 2	None	3499	xFu1vFetW040mVQppcx6zw	md5	2	2	None	1	True	2023-12-08 11:33:24.103953+00:00	1

More examples#

Understand data flow#

View the sequence of data transformations (Transform) in a project (from here, based on Schmidt et al., 2022):

transform.view_parents()

Or, the generating flow of a file or dataset:

file.view_flow()

Both figures are based on mere calls to ln.track() in notebooks, pipelines & app.

Manage biological registries#

Create a cell type registry from public knowledge and add a new cell state (from here):

import lnschema_bionty as lb

# create an ontology-coupled cell type record and save it
lb.CellType.from_bionty(name="neuron").save()

# create a record to track a new cell state
new_cell_state = lb.CellType(name="my neuron cell state", description="explains X")
new_cell_state.save()

# express that it's a neuron state
cell_types = lb.CellType.lookup()
new_cell_state.parents.add(cell_types.neuron)

# view ontological hierarchy
new_cell_state.view_parents(distance=2)

_images/10484bc13381cb2c54f3cfbaeb52e7203c6bb7635d6dbbd02f571b3d9c4c0e68.svg

Leverage a mesh of instances#

LaminDB is a distributed system like git. Similar to cloning a repository, collaborators can load your instance on the command-line using:

lamin load myhandle/myinstance

If you run lamin save <notebook_path>, you will save the notebook to your default storage location.

You can explore the notebook report corresponding to the quickstart here in LaminHub.

Manage custom schemas#

LaminDB can be customized & extended with schema & app plug-ins building on the Django ecosystem. Examples are

lnschema_bionty: Registries for basic biological entities, coupled to public ontologies.
lnschema_lamin1: Exemplary custom schema to manage samples, treatments, etc.

If you’d like to create your own schema or app:

Create a git repository with registries similar to lnschema_lamin1
Create & deploy migrations via lamin migrate create and lamin migrate deploy

It’s fastest if we do this for you based on our templates within an enterprise plan.

Design#

Why?#

We wrote a blog post about the key problems Lamin tries to solve when starting to work on it.

Schema & API#

LaminDB provides a SQL schema for common entities: File, Dataset, Transform, Feature, ULabel etc. - see the API reference or the source code.

The core schema is extendable through plug ins (see blue vs. red entities in graphic), e.g., with basic biological (Gene, Protein, CellLine, etc.) & operational entities (Biosample, Techsample, Treatment, etc.).

On top of the schema, LaminDB is a Python API that abstracts over storage & database access, data transformations, and (biological) ontologies.

The code for this is open-source & accessible through the dependencies & repositories listed below.

Dependencies#

Data is stored in a platform-independent way:
- location → local, on AWS S3 or GCP Storage, accessed through fsspec
- format → blob-like files or queryable formats like parquet, zarr, HDF5, TileDB, …
Metadata is stored in SQL: current backends are SQLite (small teams) and Postgres (any team size).
Django ORM for schema management & metadata queries.
Biological knowledge sources & ontologies: see Bionty.

For more details, see the pyproject.toml file in lamindb & the linked repositories below.

Repositories#

LaminDB and its plug-ins consist in open-source Python libraries & publicly hosted metadata assets:

lamindb: Core API, which builds on the core schema.
lnschema-bionty: Registries for basic biological entities, coupled to public ontologies.
lnschema-lamin1: Exemplary custom schema to manage samples, treatments, etc.
lamindb-setup: Setup & configure LaminDB, client for LaminHub.
lamin-cli: CLI for lamindb and lamindb-setup.
bionty: Accessor for public biological ontologies.
nbproject: Metadata parser for Jupyter notebooks.
lamin-utils: Generic utilities, e.g., a logger.
readfcs: FCS file reader.

LaminHub is not open-sourced, and neither are plug-ins that model lab operations.

Assumptions & principles#

Data is generated by instruments that process physical samples: it comes in batches stored as immutable files.
Files are transformed into more useful data representations, e.g.:
- Summary statistics, e.g., count matrices for fastq files
- Arrays of non-array-like input data (e.g., images)
- Higher-level embeddings for lower-level array, text or graph representations
- Concatenated arrays for large-scale atlas-like datasets
Semantics of high-level embeddings (“inflammatory”, “lipophile”) are anchored in experimental metadata and knowledge (ontologies)
Experimental metadata is another ontology type
Experiments measure features (Feature, CellMarker, …)
Samples are annotated by labels (ULabel, CellLine, …)
Learning and data warehousing both iterate transformations (see graphic, Transform)
Basic biological entities should have the same meaning to anyone and across any data platform
Schema migrations should be easy

Influences#

LaminDB was influenced by many other projects, see Influences.

Notebooks#

Find all tutorial & guide notebooks here and use cases here.
You can run these notebooks in hosted versions of JupyterLab, e.g., Saturn Cloud, Google Vertex AI, Google Colab, and others.