Flow cytometry

You’ll learn how to manage a growing number of flow cytometry data batches as a single queryable dataset.

Specifically, you will

1. read a single .fcs file as an AnnData and seed a versioned dataset with it

2. append a new data batch (a new .fcs file) to create a new version of the dataset (here)

3. query individual files and cell markers (here)

4. analyze the dataset and store results as plots (here)

Setup

!lamin init --storage ./test-facs --schema bionty

Show code cell output Hide code cell output

✅ saved: User(id='DzTjkKse', handle='testuser1', email='testuser1@lamin.ai', name='Test User1', updated_at=2023-10-02 10:21:00)
✅ saved: Storage(id='4Rk0y6XZ', root='/home/runner/work/lamin-usecases/lamin-usecases/docs/test-facs', type='local', updated_at=2023-10-02 10:21:00, created_by_id='DzTjkKse')

💡 loaded instance: testuser1/test-facs
💡 did not register local instance on hub (if you want, call `lamin register`)

import lamindb as ln
import lnschema_bionty as lb
import readfcs

lb.settings.species = "human"

💡 loaded instance: testuser1/test-facs (lamindb 0.54.4)

ln.track()

💡 notebook imports: lamindb==0.54.4 lnschema_bionty==0.31.2 pytometry==0.1.4 readfcs==1.1.6 scanpy==1.9.5

💡 Transform(id='OWuTtS4SAponz8', name='Flow cytometry', short_name='facs', version='0', type=notebook, updated_at=2023-10-02 10:21:04, created_by_id='DzTjkKse')

💡 Run(id='WVFjYsTXmmlztFLeq1e5', run_at=2023-10-02 10:21:04, transform_id='OWuTtS4SAponz8', created_by_id='DzTjkKse')

Ingest a first file

Access

We start with a flow cytometry file from Alpert et al., Nat. Med. (2019).

Calling the following function downloads the file and pre-populates a few relevant registries:

ln.dev.datasets.file_fcs_alpert19(populate_registries=True)

PosixPath('Alpert19.fcs')

We use readfcs to read the raw fcs file into memory and create an AnnData object:

adata = readfcs.read("Alpert19.fcs")
adata

AnnData object with n_obs × n_vars = 166537 × 40
    var: 'n', 'channel', 'marker', '$PnB', '$PnE', '$PnR'
    uns: 'meta'

It has the following features:

adata.var.head(10)

	n	channel	marker	$PnB	$PnE	$PnR
Time	1	Time		32	0,0	2097152
Cell_length	2	Cell_length		32	0,0	128
CD57	3	(In113)Dd	CD57	32	0,0	8192
Dead	4	(In115)Dd	Dead	32	0,0	4096
(Ba138)Dd	5	(Ba138)Dd		32	0,0	4096
Bead	6	(Ce140)Dd	Bead	32	0,0	16384
CD19	7	(Nd142)Dd	CD19	32	0,0	4096
CD4	8	(Nd143)Dd	CD4	32	0,0	4096
CD8	9	(Nd144)Dd	CD8	32	0,0	4096
IgD	10	(Nd146)Dd	IgD	32	0,0	8192

Transform: normalize

In this use case, we’d like to ingest & store curated data, and hence, we split signal and normalize using the pytometry package.

import pytometry as pm

First, we’ll split the signal from heigh and area metadata:

pm.pp.split_signal(adata, var_key="channel", data_type="cytof")

'area' is not in adata.var['signal_type']. Return all.

adata

AnnData object with n_obs × n_vars = 166537 × 40
    var: 'n', 'channel', 'marker', '$PnB', '$PnE', '$PnR', 'signal_type'
    uns: 'meta'

Normalize the dataset:

pm.tl.normalize_arcsinh(adata, cofactor=150)

Note

If the dataset was a flow dataset, you’ll also have to compensate the data, if possible. The metadata should contain a compensation matrix, which could then be run by the pytometry compensation function. In the case here, its a cyTOF dataset, which doesn’t (really) require compensation.

Validate: cell markers

First, we validate features in .var using CellMarker:

validated = lb.CellMarker.validate(adata.var.index)

❗ 13 terms (32.50%) are not validated for name: Time, Cell_length, Dead, (Ba138)Dd, Bead, CD19, CD4, IgD, CD11b, CD14, CCR6, CCR7, PD-1

We see that many features aren’t validated because they’re not standardized.

Hence, let’s standardize feature names & validate again:

adata.var.index = lb.CellMarker.standardize(adata.var.index)
validated = lb.CellMarker.validate(adata.var.index)

❗ 5 terms (12.50%) are not validated for name: Time, Cell_length, Dead, (Ba138)Dd, Bead

The remaining non-validated features don’t appear to be cell markers but rather metadata features.

Let’s move them into adata.obs:

adata.obs = adata[:, ~validated].to_df()
adata = adata[:, validated].copy()

Now we have a clean panel of 35 validated cell markers:

validated = lb.CellMarker.validate(adata.var.index)
assert all(validated)  # all markers are validated

Register: metadata

Next, let’s register the metadata features we moved to .obs.

For this, we create one feature record for each column in the .obs dataframe:

features = ln.Feature.from_df(adata.obs)
ln.save(features)

We use the Experimental Factor Ontology through Bionty to create a “FACS” label:

lb.ExperimentalFactor.bionty().search("FACS").head(2)  # search the public ontology

	ontology_id	definition	synonyms	parents	molecule	instrument	measurement	__ratio__
name
fluorescence-activated cell sorting	EFO:0009108	A Flow Cytometry Assay That Provides A Method ...	FACS|FAC sorting	[]	None	None	None	100.0
BALB/c	EFO:0000602	Balb/C Is A Mouse Strain Of Albion Mice.	BALB/cJ|C|BALBc	[]	None	None	None	90.0

We found one for “FACS”, let’s save it to our in-house registry:

# import the FACS record from the public ontology and save it to the registry
facs = lb.ExperimentalFactor.from_bionty(ontology_id="EFO:0009108")
facs.save()

We don’t find one for “CyToF”, however, so, let’s create it without importing from a public ontology but label it as a child of “is_cytometry_assay”:

cytof = lb.ExperimentalFactor(name="CyTOF")
cytof.save()
is_cytometry_assay = lb.ExperimentalFactor(name="is_cytometry_assay")
is_cytometry_assay.save()
cytof.parents.add(is_cytometry_assay)
facs.parents.add(is_cytometry_assay)

is_cytometry_assay.view_parents(with_children=True)

Let us look at the content of the registry:

lb.ExperimentalFactor.filter().df()

	name	ontology_id	abbr	synonyms	description	molecule	instrument	measurement	bionty_source_id	updated_at	created_by_id
id
lh5Cxy8w	fluorescence-activated cell sorting	EFO:0009108	None	FACS|FAC sorting	A Flow Cytometry Assay That Provides A Method ...	None	None	None	W6xl	2023-10-02 10:21:10	DzTjkKse
QljhEHxP	CyTOF	None	None	None	None	None	None	None	None	2023-10-02 10:21:10	DzTjkKse
ZyETAsPo	is_cytometry_assay	None	None	None	None	None	None	None	None	2023-10-02 10:21:10	DzTjkKse

Register: data & annotate with metadata

modalities = ln.Modality.lookup()
features = ln.Feature.lookup()
experimental_factors = lb.ExperimentalFactor.lookup()
species = lb.Species.lookup()

file = ln.File.from_anndata(
    adata, description="Alpert19", field=lb.CellMarker.name, modality=modalities.protein
)

... storing '$PnE' as categorical

... storing '$PnR' as categorical

file.save()

Inspect the registered file

Inspect features on a high level:

file.features

Features:
  var: FeatureSet(id='kPeqi2DXDNakhHqlXQN8', n=35, type='number', registry='bionty.CellMarker', hash='ldY9_GmptHLCcT7Nrpgo', updated_at=2023-10-02 10:21:10, modality_id='EFBdjpXy', created_by_id='DzTjkKse')
    'Ccr7', 'CD27', 'CD33', 'CD3', 'CD16', 'CXCR3', 'CD38', 'CD161', 'DNA1', 'CD45RA', 'CD24', 'CD11B', 'HLADR', 'Cd4', 'CD85j', 'CD11c', 'CD94', 'PD1', 'CD56', 'CD57', ...
  obs: FeatureSet(id='D9IelF210m2KvoLE4tz2', n=5, registry='core.Feature', hash='kN_l0cF14_oL_mMi1lHi', updated_at=2023-10-02 10:21:10, modality_id='shAk3Do6', created_by_id='DzTjkKse')
    Time (number)
    Dead (number)
    Bead (number)
    Cell_length (number)
    (Ba138)Dd (number)

Inspect low-level features in .var:

file.features["var"].df().head()

	name	synonyms	gene_symbol	ncbi_gene_id	uniprotkb_id	species_id	bionty_source_id	updated_at	created_by_id
id
sYcK7uoWCtco	Ccr7		CCR7	1236	P32248	uHJU	3b1l	2023-10-02 10:21:06	DzTjkKse
uThe3c0V3d4i	CD27		CD27	939	P26842	uHJU	3b1l	2023-10-02 10:21:06	DzTjkKse
c3dZKHFOdllB	CD33		CD33	945	P20138	uHJU	3b1l	2023-10-02 10:21:06	DzTjkKse
a4hvNp34IYP0	CD3		None	None	None	uHJU	3b1l	2023-10-02 10:21:06	DzTjkKse
bspnQ0igku6c	CD16		FCGR3A	2215	O75015	uHJU	3b1l	2023-10-02 10:21:06	DzTjkKse

Use auto-complete for marker names in the var featureset:

markers = file.features["var"].lookup()

markers.cd14

CellMarker(id='roEbL8zuLC5k', name='Cd14', synonyms='', gene_symbol='CD14', ncbi_gene_id='4695', uniprotkb_id='O43678', updated_at=2023-10-02 10:21:06, species_id='uHJU', bionty_source_id='3b1l', created_by_id='DzTjkKse')

In a plot, we can now easily also show gene symbol and Uniprot ID:

import scanpy as sc

sc.pp.pca(adata)
sc.pl.pca(
    adata,
    color=markers.cd14.name,
    title=(
        f"{markers.cd14.name} / {markers.cd14.gene_symbol} /"
        f" {markers.cd14.uniprotkb_id}"
    ),
)

file.view_flow()

Create a dataset from the file

dataset = ln.Dataset(file, name="My versioned cytometry dataset", version="1")

dataset

Dataset(id='UYxBk5c2glqsNytkYhWE', name='My versioned cytometry dataset', version='1', hash='e1rTes7lUXcu_bndvHhLbg', transform_id='OWuTtS4SAponz8', run_id='WVFjYsTXmmlztFLeq1e5', file_id='UYxBk5c2glqsNytkYhWE', created_by_id='DzTjkKse')

Let’s inspect the features measured in this dataset which were inherited from the file:

dataset.features

Features:
  var: FeatureSet(id='kPeqi2DXDNakhHqlXQN8', n=35, type='number', registry='bionty.CellMarker', hash='ldY9_GmptHLCcT7Nrpgo', updated_at=2023-10-02 10:21:10, modality_id='EFBdjpXy', created_by_id='DzTjkKse')
    'Ccr7', 'CD27', 'CD33', 'CD3', 'CD16', 'CXCR3', 'CD38', 'CD161', 'DNA1', 'CD45RA', 'CD24', 'CD11B', 'HLADR', 'Cd4', 'CD85j', 'CD11c', 'CD94', 'PD1', 'CD56', 'CD57', ...
  obs: FeatureSet(id='D9IelF210m2KvoLE4tz2', n=5, registry='core.Feature', hash='kN_l0cF14_oL_mMi1lHi', updated_at=2023-10-02 10:21:10, modality_id='shAk3Do6', created_by_id='DzTjkKse')
    Time (number)
    Dead (number)
    Bead (number)
    Cell_length (number)
    (Ba138)Dd (number)

This looks all good, hence, let’s save it:

dataset.save()

Annotate by linking cytof & species labels:

dataset.labels.add(experimental_factors.cytof, features.assay)
dataset.labels.add(species.human, features.species)

dataset.view_flow()