Gene Ontology (GO)#

Pathways represent interconnected molecular networks of signaling cascades that govern critical cellular processes. They provide understandings cellular behavior mechanisms, insights of disease progression and treatment responses. In an R&D organization, managing pathways across different datasets are crucial for gaining insights of potential therapeutic targets and intervention strategies.

In this notebook we manage a pathway registry based on “2023 GO Biological Process” ontology. We’ll walk you through the following steps:

Register pathways and link them to genes.
Perform a pathway enrichment analysis on an interferon-beta treated dataset and track the dataset with LaminDB.
Demonstrate how datasets are queryable by pathways and genes using LaminDB.

Setup#

Warning

Please ensure that you have created or loaded a LaminDB instance before running the remaining part of this notebook!

# A lamindb instance containing bionty schema (skip if you already loaded your own instance)
!lamin init --storage ./enrichr --schema bionty

import lamindb as ln
import lnschema_bionty as lb
from lamin_usecases import datasets as ds
import gseapy as gp
import scanpy as sc
import matplotlib.pyplot as plt

lb.settings.species = "human"  # globally set species

💡 loaded instance: testuser1/enrichr (lamindb 0.56a1)

Fetch GO pathways annotated with human genes using Enrichr#

First we fetch the “GO_Biological_Process_2023” pathways for humans using GSEApy which wraps GSEA and Enrichr.

go_bp = gp.get_library(name="GO_Biological_Process_2023", organism="Human")
print(f"Number of pathways {len(go_bp)}")

2023-10-16 21:43:01,677:INFO - Downloading and generating Enrichr library gene sets...

2023-10-16 21:43:04,035:INFO - 0001 gene_sets have been filtered out when max_size=2000 and min_size=0

Number of pathways 5406

go_bp["ATF6-mediated Unfolded Protein Response (GO:0036500)"]

['MBTPS1', 'MBTPS2', 'XBP1', 'ATF6B', 'DDIT3', 'CREBZF']

Parse out the ontology_id from keys, convert into the format of {ontology_id: (name, genes)}

def parse_ontology_id_from_keys(key):
    """Parse out the ontology id.

    "ATF6-mediated Unfolded Protein Response (GO:0036500)" -> ("GO:0036500", "ATF6-mediated Unfolded Protein Response")
    """
    id = key.split(" ")[-1].replace("(", "").replace(")", "")
    name = key.replace(f" ({id})", "")
    return (id, name)

go_bp_parsed = {}

for key, genes in go_bp.items():
    id, name = parse_ontology_id_from_keys(key)
    go_bp_parsed[id] = (name, genes)

go_bp_parsed["GO:0036500"]

('ATF6-mediated Unfolded Protein Response',
 ['MBTPS1', 'MBTPS2', 'XBP1', 'ATF6B', 'DDIT3', 'CREBZF'])

Register pathway ontology in LaminDB#

pathway_bionty = lb.Pathway.bionty()  # equals to bionty.Pathway()

pathway_bionty

Pathway
Species: all
Source: go, 2023-05-10
#terms: 47514

📖 Pathway.df(): ontology reference table
🔎 Pathway.lookup(): autocompletion of terms
🎯 Pathway.search(): free text search of terms
✅ Pathway.validate(): strictly validate values
🧐 Pathway.inspect(): full inspection of values
👽 Pathway.standardize(): convert to standardized names
🪜 Pathway.diff(): difference between two versions
🔗 Pathway.ontology: Pronto.Ontology object

Next, we register all the pathways and genes in LaminDB to finally link pathways to genes.

Register pathway terms#

To register the pathways we make use of .from_values to directly parse the annotated GO pathway ontology IDs into LaminDB.

pathway_records = lb.Pathway.from_values(go_bp_parsed.keys(), lb.Pathway.ontology_id)

lb.Pathway.from_bionty(ontology_id="GO:0015868")

Pathway(uid='SMqshx3Y', name='purine ribonucleotide transport', ontology_id='GO:0015868', description='The Directed Movement Of A Purine Ribonucleotide, Any Compound Consisting Of A Purine Ribonucleoside (A Purine Organic Base Attached To A Ribose Sugar) Esterified With (Ortho)Phosphate, Into, Out Of Or Within A Cell.', bionty_source_id=39, created_by_id=1)

ln.save(pathway_records, parents=False)  # not recursing through parents

Register gene symbols#

Similarly, we use .from_values for all Pathway associated genes to register them with LaminDB.

all_genes = {g for genes in go_bp.values() for g in genes}

gene_records = lb.Gene.from_values(all_genes, lb.Gene.symbol)

gene_records[:3]

[Gene(uid='4DDdKM0LEjQ0', symbol='SNRNP70', ensembl_gene_id='ENSG00000104852', ncbi_gene_ids='6625', biotype='protein_coding', description='small nuclear ribonucleoprotein U1 subunit 70 [Source:HGNC Symbol;Acc:HGNC:11150]', synonyms='RPU1|RNPU1Z|SNP1|U1-70K|SNRP70', species_id=1, bionty_source_id=8, created_by_id=1),
 Gene(uid='TbmZksiqeaPj', symbol='NEFM', ensembl_gene_id='ENSG00000104722', ncbi_gene_ids='4741', biotype='protein_coding', description='neurofilament medium chain [Source:HGNC Symbol;Acc:HGNC:7734]', synonyms='NEF3|NF-M|NFM', species_id=1, bionty_source_id=8, created_by_id=1),
 Gene(uid='tsm5nM4xcyuK', symbol='KCNJ3', ensembl_gene_id='ENSG00000162989', ncbi_gene_ids='3760', biotype='protein_coding', description='potassium inwardly rectifying channel subfamily J member 3 [Source:HGNC Symbol;Acc:HGNC:6264]', synonyms='GIRK1|KIR3.1|KGA', species_id=1, bionty_source_id=8, created_by_id=1)]

ln.save(gene_records);

Link pathway to genes#

Now that we are tracking all pathways and genes records, we can link both of them to make the pathways even more queryable.

gene_records_ids = {record.symbol: record for record in gene_records}

for pathway_record in pathway_records:
    pathway_genes = go_bp_parsed.get(pathway_record.ontology_id)[1]
    pathway_genes_records = [gene_records_ids.get(gene) for gene in pathway_genes]
    pathway_record.genes.set(pathway_genes_records)

Now genes are linked to pathways:

pathway_record.genes.list("symbol")

['CARD18', 'CARD8', 'CST7', 'CAST', 'XIAP']

A interferon-beta treated dataset#

We will now conduct a pathway enrichment analysis on a small peripheral blood mononuclear cell dataset that is split into control and stimulated groups. The stimulated group was treated with interferon beta.

Let’s load the dataset and look at the cell type annotations.

adata = ds.anndata_seurat_ifnb()

adata

AnnData object with n_obs × n_vars = 13999 × 14053
    obs: 'orig.ident', 'nCount_RNA', 'nFeature_RNA', 'stim', 'seurat_annotations'
    var: 'features'
    uns: 'log1p'

adata.obs["seurat_annotations"].value_counts()

CD14 Mono       4362
CD4 Naive T     2504
CD4 Memory T    1762
CD16 Mono       1044
B                978
CD8 T            814
NK               633
T activated      619
DC               472
B Activated      388
Mk               236
pDC              132
Eryth             55
Name: seurat_annotations, dtype: int64

For simplicity, we subset to “B Activated” cells:

adata_ba = adata[adata.obs.seurat_annotations == "B Activated"].copy()
adata_ba

AnnData object with n_obs × n_vars = 388 × 14053
    obs: 'orig.ident', 'nCount_RNA', 'nFeature_RNA', 'stim', 'seurat_annotations'
    var: 'features'
    uns: 'log1p'

Pathway enrichment analysis using Enrichr#

This analysis is based on the GSEApy scRNA-seq Example.

First, we compute differentially expressed genes using a Wilcoxon test between stimulated and control cells.

# compute differentially expressed genes
sc.tl.rank_genes_groups(
    adata_ba,
    groupby="stim",
    use_raw=False,
    method="wilcoxon",
    groups=["STIM"],
    reference="CTRL",
)

rank_genes_groups_df = sc.get.rank_genes_groups_df(adata_ba, "STIM")

... storing 'orig.ident' as categorical

... storing 'seurat_annotations' as categorical

rank_genes_groups_df.head()

	names	scores	logfoldchanges	pvals	pvals_adj
0	ISG15	16.881584	5.923428	6.147295e-64	6.536230e-60
1	ISG20	16.857113	4.167713	9.302256e-64	6.536230e-60
2	IFIT3	14.587233	31.232290	3.386569e-48	1.586382e-44
3	IFI6	14.128634	6.471180	2.530019e-45	8.888589e-42
4	MX1	13.442097	6.241539	3.425901e-41	9.628837e-38

Next, we filter out up/down-regulated differentially expressed gene sets:

degs_up = rank_genes_groups_df[
    (rank_genes_groups_df["logfoldchanges"] > 0)
    & (rank_genes_groups_df["pvals_adj"] < 0.05)
]
degs_dw = rank_genes_groups_df[
    (rank_genes_groups_df["logfoldchanges"] < 0)
    & (rank_genes_groups_df["pvals_adj"] < 0.05)
]

degs_up.shape, degs_dw.shape

((89, 5), (47, 5))

Run pathway enrichment analysis on DEGs and plot top 10 pathways:

enr_up = gp.enrichr(degs_up.names, gene_sets="GO_Biological_Process_2023").res2d

gp.dotplot(enr_up, figsize=(2, 3), title="Up", cmap=plt.cm.autumn_r);

https://d33wubrfki0l68.cloudfront.net/98ee88a4dfd9bb9a68cb2891e25a10c573534262/ac5de/_images/f6e62d4860e9e087d96d2d9839454e975af46f11f34b8567f3214a6b77e39852.png

enr_dw = gp.enrichr(degs_dw.names, gene_sets="GO_Biological_Process_2023").res2d

gp.dotplot(enr_dw, figsize=(2, 3), title="Down", cmap=plt.cm.winter_r, size=10);

https://d33wubrfki0l68.cloudfront.net/27f214ecc93d0e7d2821b6ba01d7f78b837e80e2/b60e7/_images/74559dded90618c90183cd29f8c3d51bb543e8adecdfb27d4716eba1b5192f69.png

Track datasets containing annotated pathways in LaminDB#

Let’s enable tracking of the current notebook as the transform of this file:

ln.track()

💡 notebook imports: gseapy==1.0.6 lamin_usecases==0.0.1 lamindb==0.56a1 lnschema_bionty==0.32.0 matplotlib==3.8.0 scanpy==1.9.5

💡 Transform(id=1, uid='6oxEIEduvo6wz8', name='Gene Ontology (GO)', short_name='enrichr', version='0', type=notebook, updated_at=2023-10-16 21:43:48, created_by_id=1)

💡 Run(id=1, uid='VLHBYD0jcN6LNJCwRoXu', run_at=2023-10-16 21:43:48, transform_id=1, created_by_id=1)

We further create a File object to track the dataset.

file = ln.File.from_anndata(
    adata_ba, description="seurat_ifnb_activated_Bcells", field=lb.Gene.symbol
)
file.save()

❗    4729 terms (33.70%) are not validated for symbol: AL627309.1, RP11-206L10.2, LINC00115, KLHL17, C1orf159, ACAP3, CPSF3L, GLTPD1, RP4-758J18.2, AL645728.1, RP11-345P4.9, SLC35E2B, SLC35E2, RP5-892K4.1, C1orf86, AL590822.2, MORN1, RP3-395M20.12, RP3-395M20.9, FAM213B, ...

❗    5 terms (100.00%) are not validated for name: orig.ident, nCount_RNA, nFeature_RNA, stim, seurat_annotations

❗    no validated features, skip creating feature set

We further create two feature sets for degs_up and degs_dw which we can later associate with the associated pathways:

degs_up_featureset = ln.FeatureSet.from_values(degs_up.names, lb.Gene.symbol)

degs_dw_featureset = ln.FeatureSet.from_values(degs_dw.names, lb.Gene.symbol)

Link the top 10 pathways to the corresponding differentially expressed genes:

# get ontology ids for the top 10 pathways
enr_up_top10 = [
    pw_id[0] for pw_id in enr_up.head(10).Term.apply(parse_ontology_id_from_keys)
]
enr_dw_top10 = [
    pw_id[0] for pw_id in enr_dw.head(10).Term.apply(parse_ontology_id_from_keys)
]

# get pathway records
enr_up_top10_pathways = lb.Pathway.from_values(enr_up_top10, lb.Pathway.ontology_id)
enr_dw_top10_pathways = lb.Pathway.from_values(enr_dw_top10, lb.Pathway.ontology_id)

Link feature sets to file:

file.features.add_feature_set(degs_up_featureset, slot="up-DEGs")
file.features.add_feature_set(degs_dw_featureset, slot="down-DEGs")

Associate the pathways to the differentially expressed genes:

degs_up_featureset.pathways.set(enr_up_top10_pathways)
degs_dw_featureset.pathways.set(enr_dw_top10_pathways)

degs_up_featureset.pathways.list("name")

['antiviral innate immune response',
 'defense response to symbiont',
 'response to type II interferon',
 'defense response to virus',
 'negative regulation of viral genome replication',
 'interleukin-27-mediated signaling pathway',
 'positive regulation of interferon-beta production',
 'response to interferon-beta',
 'negative regulation of viral process',
 'regulation of viral genome replication']

Querying for pathways#

Querying for pathways is now simple with .filter:

lb.Pathway.filter(name__contains="interferon-beta").df()

	uid	name	ontology_id	abbr	synonyms	description	bionty_source_id	updated_at	created_by_id
id
684	l06ZujxW	cellular response to interferon-beta	GO:0035458	None	cellular response to fibroblast interferon\|cel...	Any Process That Results In A Change In State ...	39	2023-10-16 21:43:07	1
2130	uu9GYFx2	negative regulation of interferon-beta production	GO:0032688	None	down regulation of interferon-beta production\|...	Any Process That Stops, Prevents, Or Reduces T...	39	2023-10-16 21:43:07	1
3127	SGYMKD7O	positive regulation of interferon-beta production	GO:0032728	None	positive regulation of IFN-beta production\|up-...	Any Process That Activates Or Increases The Fr...	39	2023-10-16 21:43:07	1
4334	GD9xCHBK	regulation of interferon-beta production	GO:0032648	None	regulation of IFN-beta production	Any Process That Modulates The Frequency, Rate...	39	2023-10-16 21:43:07	1
4953	mCgM7JYR	response to interferon-beta	GO:0035456	None	response to fiblaferon\|response to fibroblast ...	Any Process That Results In A Change In State ...	39	2023-10-16 21:43:07	1

Query pathways from a gene:

lb.Pathway.filter(genes__symbol="KIR2DL1").df()

	uid	name	ontology_id	abbr	synonyms	description	bionty_source_id	updated_at	created_by_id
id
1346	TSXmNUbN	immune response-inhibiting cell surface recept...	GO:0002767	None	immune response-inhibiting cell surface recept...	The Series Of Molecular Signals Initiated By A...	39	2023-10-16 21:43:07	1

Query files from a pathway:

ln.File.filter(feature_sets__pathways__name__icontains="interferon-beta").first()

File(id=1, uid='VuTyryuGPIZw6mN1q0Eu', suffix='.h5ad', accessor='AnnData', description='seurat_ifnb_activated_Bcells', size=5896640, hash='DsfKFJ906VT9UR4lrxd8Ag', hash_type='md5', updated_at=2023-10-16 21:43:49, storage_id=1, transform_id=1, run_id=1, created_by_id=1)

Query featuresets from a pathway to learn from which geneset this pathway was computed:

pathway = lb.Pathway.filter(ontology_id="GO:0035456").one()
pathway

Pathway(id=4953, uid='mCgM7JYR', name='response to interferon-beta', ontology_id='GO:0035456', synonyms='response to fiblaferon|response to fibroblast interferon|response to interferon beta', description='Any Process That Results In A Change In State Or Activity Of A Cell Or An Organism (In Terms Of Movement, Secretion, Enzyme Production, Gene Expression, Etc.) As A Result Of An Interferon-Beta Stimulus. Interferon-Beta Is A Type I Interferon.', updated_at=2023-10-16 21:43:07, bionty_source_id=39, created_by_id=1)

degs = ln.FeatureSet.filter(pathways__ontology_id=pathway.ontology_id).one()

Now we can get the list of genes that are differentially expressed and belong to this pathway:

pathway_genes = set(pathway.genes.list("symbol"))
degs_genes = set(degs.genes.list("symbol"))

pathway_genes.intersection(degs_genes)

{'BST2',
 'IFI16',
 'IFITM2',
 'IFITM3',
 'IRF1',
 'OAS1',
 'PLSCR1',
 'STAT1',
 'XAF1'}

# clean up test instance
!lamin delete --force enrichr
!rm -r ./enrichr