LaminDB - A data framework for biology

Makes your data queryable, traceable, reproducible, and FAIR. One API: lakehouse, lineage, feature store, ontologies, LIMS, ELN.

Why?

Reproducing analytical results or understanding how a dataset or model was created can be a pain. Training models on historical data, LIMS & ELN systems, orthogonal assays, or datasets from other teams is even harder. Even maintaining an overview of a project's datasets & analyses is more difficult than it should be.

Biological datasets are typically managed with versioned storage systems, GUI-focused platforms, structureless data lakes, rigid data warehouses (SQL, monolithic arrays), or tabular lakehouses.

LaminDB extends the lakehouse architecture to biological registries & datasets beyond tables (DataFrame, AnnData, .zarr, .tiledbsoma, …) with enough structure to enable queries and enough freedom to keep the pace of R&D high. Moreover, it provides context through data lineage -- tracing data and code, scientists and models -- and abstractions for biological domain knowledge and experimental metadata.

Highlights:

• lineage → track inputs & outputs of notebooks, scripts, functions & pipelines with a single line of code
• lakehouse → manage, monitor & validate schemas; query across many datasets
• feature store → manage features & labels; leverage batch loading
• FAIR datasets → validate & annotate DataFrame, AnnData, SpatialData, parquet, zarr, …
• LIMS & ELN → manage experimental metadata, ontologies & markdown notes
• unified access → storage locations (local, S3, GCP, …), SQL databases (Postgres, SQLite) & ontologies
• reproducible → auto-version & timestamp execution reports, source code & environments
• zero lock-in & scalable → runs in your infrastructure; not a client for a rate-limited REST API
• integrations → vitessce, nextflow, redun, and more
• extendable → create custom plug-ins based on the Django ORM

If you want a GUI, you can connect your LaminDB instance to LaminHub and close the drylab-wetlab feedback loop: lamin.ai.

Who uses it?

Scientists & engineers in pharma, biotech, and academia, including:

• Pfizer – A global BigPharma company with headquarters in the US
• Ensocell Therapeutics – A BioTech with offices in Cambridge, UK, and California
• DZNE – The National Research Center for Neuro-Degenerative Diseases in Germany
• Helmholtz Munich – The National Research Center for Environmental Health in Germany
• scverse – An international non-profit consortium for open-source omics data tools
• The Global Immunological Swarm Learning Network – Research hospitals at U Bonn, Harvard, MIT, Stanford, ETH Zürich, Charite, Mount Sinai, and others

Docs

Copy summary.md into an LLM chat and let AI explain or read the docs.

Setup

Install the lamindb Python package:

pip install lamindb

Create a LaminDB instance:

lamin init --modules bionty --storage ./quickstart-data  # or s3://my-bucket, gs://my-bucket

Or if you have write access to an instance, connect to it:

lamin connect account/name

Quickstart

Lineage

Create a dataset while tracking source code, inputs, outputs, logs, and environment:

import lamindb as ln

ln.track()  # track execution of source code as a run
open("sample.fasta", "w").write(">seq1\nACGT\n")  # create a dataset
ln.Artifact("sample.fasta", key="sample.fasta").save()  # save dataset as an artifact
ln.finish()  # mark the run as finished

Running this snippet as a script (python create-fasta.py) produces the following data lineage.

artifact = ln.Artifact.get(key="sample.fasta")  # get artifact by key
artifact.view_lineage()

You'll know how that artifact was created and what it's used for. Basic metadata was captured in fields:

artifact.size        # access the size
artifact.created_at  # access the timestamp
# etc.
artifact.describe()  # describe metadata

Here is how to access the content of the artifact:

local_path = artifact.cache()  # return a local path from a cache
object = artifact.load()       # load object into memory
accessor = artifact.open()     # return a streaming accessor

And here is how to access its data lineage context:

run = artifact.run         # get the run record
transform = run.transform  # get the transform record

Just like artifacts, runs & transforms are SQLRecord objects and follow the same API, just with different fields.

run.describe()	transform.describe()

Lake: annotation & queries

You can annotate datasets and samples with features. Let's define some:

from datetime import date

ln.Feature(name="gc_content", dtype=float).save()
ln.Feature(name="experiment_note", dtype=str).save()
ln.Feature(name="experiment_date", dtype=date).save()

During annotation, feature names and data types are validated against these definitions:

artifact.features.add_values({
    "gc_content": 0.55,
    "experiment_note": "Looks great",
    "experiment_date": "2025-10-24",
})

Now that the data is annotated, you can query for it:

ln.Artifact.filter(experiment_date="2025-10-14").to_dataframe()  # query all artifacts annotated with `experiment_date`

You can also query by the metadata that lamindb automatically collects:

ln.Artifact.filter(run=run).to_dataframe()                # query all artifacts created by a run
ln.Artifact.filter(transform=transform).to_dataframe()    # query all artifacts created by a transform
ln.Artifact.filter(size__gt=1e6).to_dataframe()           # query all artifacts bigger than 1MB

If you want to include more information into the resulting dataframe, pass include.

ln.Artifact.to_dataframe(include="features")  # include the feature annotations
ln.Artifact.to_dataframe(include=["created_by__name", "storage__root"])  # include fields from related registries

Lake ♾️ LIMS ♾️ Sheets

You can create records for the entities underlying your experiments: samples, perturbations, instruments, etc.. For example:

sample_type = ln.Record(name="Sample", is_type=True).save()  # a sample type
ln.Record(name="P53mutant1", type=sample_type).save()  # sample 1
ln.Record(name="P53mutant2", type=sample_type).save()  # sample 2

Define the corresponding features and annotate:

ln.Feature(name="design_sample", dtype=sample_type).save()
artifact.features.add_values({"design_sample": "P53mutant1"})

You can query & search the Record registry in the same way as Artifact or Run.

ln.Record.search("p53").to_dataframe()

You can also create relationships of entities and -- if you connect your LaminDB instance to LaminHub -- edit them like Excel sheets in a GUI.

Lake: versioning

If you change source code or datasets, LaminDB manages their versioning for you. Assume you run a new version of our create-fasta.py script to create a new version of sample.fasta.

import lamindb as ln

ln.track()
open("sample.fasta", "w").write(">seq1\nTGCA\n")  # a new sequence
ln.Artifact("sample.fasta", key="sample.fasta", features={"design_sample": "P53mutant1"}).save()  # annotate with the new sample
ln.finish()

If you now query by key, you'll get the latest version of this artifact.

artifact = ln.Artifact.get(key="sample.fasta")  # get artifact by key
artifact.versions.to_dataframe()                # see all versions of that artifact

Lakehouse ♾️ feature store

Here is how you ingest a DataFrame:

import pandas as pd

df = pd.DataFrame({
    "sequence_str": ["ACGT", "TGCA"],
    "gc_content": [0.55, 0.54],
    "experiment_note": ["Looks great", "Ok"],
    "experiment_date": ["2025-10-24", "2025-10-25"],
})
ln.Artifact.from_dataframe(df, key="my_datasets/sequences.parquet").save()  # no validation

To validate & annotate the content of the dataframe, use a built-in schema:

ln.Feature(name="sequence_str", dtype=str).save()  # define a remaining feature
artifact = ln.Artifact.from_dataframe(
    df,
    key="my_datasets/sequences.parquet",
    schema="valid_features"  # validate columns against features
)
artifact.describe()

Now you know which schema the dataset satisfies. You can filter for datasets by schema and then launch distributed queries and batch loading.

Lakehouse beyond tables

To validate an AnnData with a built-in schema call:

import anndata as ad
import numpy as np

adata = ad.AnnData(
    X=pd.DataFrame([[1]*10]*21).values,
    obs=pd.DataFrame({'cell_type_by_model': ['T cell', 'B cell', 'NK cell'] * 7}),
    var=pd.DataFrame(index=[f'ENSG{i:011d}' for i in range(10)])
)

artifact = ln.Artifact.from_anndata(
    adata,
    key="my_datasets/scrna.h5ad",
    schema="ensembl_gene_ids_and_valid_features_in_obs"
)
artifact.describe()

To validate a spatialdata or any other array-like dataset, you need to construct a Schema. You can do this by composing the schema of a complicated object from simple pandera/pydantic-like schemas: docs.lamin.ai/curate.

Ontologies

Plugin bionty gives you >20 of them as SQLRecord registries. This was used to validate the ENSG ids in the adata just before.

import bionty as bt

bt.CellType.import_source()  # import the default ontology
bt.CellType.to_dataframe()   # your extendable cell type ontology in a simple registry

CLI

Most of the functionality that's available in Python is also available on the command line (and in R through LaminR). For instance, to upload a file or folder, run:

lamin save myfile.txt --key examples/myfile.txt

Workflow managers

LaminDB is not a workflow manager, but it integrates well with existing workflow managers and can subsitute them in some settings.

In github.com/laminlabs/schmidt22 we manage several workflows, scripts, and notebooks to re-construct the project of Schmidt el al. (2022). A phenotypic CRISPRa screening result is integrated with scRNA-seq data. Here is one of the input artifacts:

And here is the lineage of the final result:

You can explore it here.

If you'd like to integrate with Nextflow, Snakemake, or redun, see here: docs.lamin.ai/pipelines