This is the python package for the drugst.one platform. It offers tools for network-based drug repurposing and is a programmatic interface to Drugst.one.
- Website: https://drugst.one/
- Source code: https://github.com/drugst-one/python-package
- Bug reports: https://github.com/drugst-one/python-package/issues
Install drugstone.
pip install drugstoneImport drugstone in python
import drugstoneDrugstone officially supports Python 3.6+.
Drugstone offers a toolbox for drug repurposing applications.
- Search for drugs, interacting with a list of genes
- Search for drug targets, for a list of genes
- Visualize data in common formats like JSON or CSV
- create interaction graphs for drug and gene interactions
Protein-protein interactions (ppi_dataset):
NeDRex, BioGRID, IID, IntAct, STRING, APID
Protein-drug interactions (pdi_dataset):
NeDRex, DrugBank, Drug Central, ChEMBL, DGIdb
Please note that some of the datasets require you to accept their terms and conditions before usage. DrugBank can only be used if the license has been agreed to and since NeDRex includes DrugBank data, only a part of NeDRex is available without agreeing to our license.
The terms and conditions can be read by calling
drugstone.print_license()
and can be accepted after reading with
drugstone.accept_license().
In the the main Drugst.One database is updated automatically weekly. This is the database used by default (https://api.drugst.one/). For more stable results, change the api endpoint to our stable API (https://api.stable.drugst.one/) using the function below. The Drugst.One stable version is updated yearly (on the first of January).
drugstone.set_api('https://api.stable.drugst.one/')With Drugstone it is easy and convenient to search for drugs or drug-targets, starting with a list of genes.
genes = [
"CFTR", "TGFB1", "SCNN1B",
"DCTN4", "SCNN1A", "SCNN1G",
"CLCA4", "TNFRSF1A", "FCGR2A"
]
parameters = {
"target": "drug",
"algorithm": "trustrank"
}
task = drugstone.new_task(genes, parameters)
r = task.get_result()
genes = r.get_genes()
drugs = r.get_drugs()
# save directly to files
r.download_json()
r.download_graph()You can start multiple tasks at once, either with completely independent parameters or with same parameters and different algorithms.
By defining an algorithms value in the parameters dictionary, you can pass a list of algorithm values. For every algorithm, a task will be started, with otherwise same parameter values.
genes = [
"CFTR", "TGFB1", "SCNN1B",
"DCTN4", "SCNN1A", "SCNN1G",
"CLCA4", "TNFRSF1A", "FCGR2A"
]
parameters = {
"target": "drug",
"algorithms": ["trustrank", "closeness", "degree"]
}
tasks = drugstone.new_tasks(genes, parameters)
r = task.to_dict()
r.download_json()new_tasks() accepts a list of parameter dictionaries.
For every dictionary a task will be started.
genes = [
"CFTR", "TGFB1", "SCNN1B",
"DCTN4", "SCNN1A", "SCNN1G",
"CLCA4", "TNFRSF1A", "FCGR2A"
]
p1 = {
"target": "drug",
"ppiDataset": 'nedrex',
"pdiDataset": "drugcentral"
}
p2 = {
"target": "drug",
"ppiDataset": 'IID',
"pdiDataset": "chembl"
}
p3 = {
"target": "drug",
"ppiDataset": 'apid',
"pdiDataset": "dgidb"
}
tasks = drugstone.new_tasks(genes, [p1, p2, p3])
r = tasks.get_result()
r.to_dict()
r.download_json()
# only with Python 3.6
r.create_upset_plot() You can get the union or intersection of tasks. That returns a TaskResult with the according result.
genes = [
"CFTR", "TGFB1", "SCNN1B",
"DCTN4", "SCNN1A", "SCNN1G",
"CLCA4", "TNFRSF1A", "FCGR2A"
]
parameters = {
"target": "drug",
"algorithms": ["trustrank", "closeness", "degree"]
}
tasks = drugstone.new_tasks(genes, parameters)
u = tasks.get_union()
u.download_json()
i = tasks.get_intersection()
i.download_json()You can add custom nodes and edges to the network, which may be considered for the drug-target or the drug search. Also, it is possible to ONLY consider your provided PPI-network and exclude PPI-edges provdided by Drugst.One. Please note that this is an experimental feature and is still being tested.
import drugstone
seeds = [
"CFTR", "TGFB1"
]
parameters = {
"target": "drug",
"algorithm": "trustrank",
"pdiDataset": 'nedrex', # drug nodes and edges will still be taken from Drugst.One
"custom_edges": [ # add here your edges for you custom PPI
{"from": "CFTR", "to": "SCNN1G"},
{"from": "CFTR", "to": "SCNN1B"},
{"from": "TGFB1", "to": "DCTN4"},
{"from": "SCNN1B", "to": "SCNN1A"},
{"from": "SCNN1B", "to": "CLCA4"},
{"from": "TNFRSF1A", "to": "FCGR2A"}
],
"exclude_drugstone_ppi_edges": True, # this parameter removes all Drugst.One PPI edges (only if edges in 'custom_edges' are provided)
"identifier": "symbol",
"include_indirect_drugs": True,
"network_nodes": ["CFTR", "TGFB1", "SCNN1B", # add here your nodes for you custom PPI
"DCTN4", "SCNN1A", "SCNN1G",
"CLCA4", "TNFRSF1A", "FCGR2A"]
}
task = drugstone.new_task(seeds, parameters)
r = task.get_result()
genes = r.get_genes()
drugs = r.get_drugs()For large networks such as a complete PPI, custom_edges can also be given in a compact format:
a list of nodes and the edges as pairs of positions in that list. It is about a third of the size
of the edge list above.
parameters["custom_edges"] = {
"nodes": ["CFTR", "SCNN1G", "SCNN1B", "TGFB1", "DCTN4"],
"edges": [[0, 1], [0, 2], [3, 4]] # CFTR-SCNN1G, CFTR-SCNN1B, TGFB1-DCTN4
}This will perform a drug-target search for the seed genes and then use the drug-target search results and the seed genes to perform a drug-search. Finally, a Task with the drug-search results will be returned.
genes = [
"CFTR", "TGFB1", "SCNN1B",
"DCTN4", "SCNN1A", "SCNN1G",
"CLCA4", "TNFRSF1A", "FCGR2A"
]
parameters = {
"algorithm": "trustrank"
}
task = drugstone.deep_search(genes, parameters)
r = tasks.get_result()
r.to_dict()
r.download_json()
# only with Python 3.6
r.create_upset_plot() parameters = {
"identifier": "symbol", #("entrez" | "uniprot" | "ensg" will be supported in future versions)
"algorithm": "adjacentDrugs" | "trustrank" | "multisteiner" | "keypathwayminer" | "closeness" | "degree" | "proximity" | "betweenness"
| "harmonic" | "first-neighbor" | "connect" | "connectSelected" | "quick" | "super"
| "pathway-enrichment" | "louvain-clustering" | "leiden-clustering" | "adjacentDisorders",
"ppiDataset": "NeDRex",
"pdiDataset": "NeDRex",
"resultSize": 20,
"target": "drug" | "drug-target" | "gene" | "clustering",
"includeIndirectDrugs": True | False,
"includeNonApprovedDrugs": True | False,
"maxDeg": sys.maxsize, # filter out nodes with high degrees
"hubPenalty": 0.0, # penalize hub nodes
"filterPaths": True | False, # include only shortest connections in the result
"damping_factor": 0.85, # only in trustrank
"num_trees": 5, # only in multisteiner
"tolerance": 10, # only in multisteiner
"k": 5, # only in keypathwayminer
}
Additional targets: gene for pathway enrichment and clustering for Louvain/Leiden.
Set the target explicitly. New options include alpha and kegg/reactome/wiki
(enrichment; KEGG only by default), seed/ignore_isolated (clustering), resolution
(Louvain), max_nodes (Leiden) and only_upstream_regulators (First Neighbor).
quick and super use fixed parameters and ignore custom edges.
Louvain and Leiden raise ValueError if no edge connects two distinct input proteins.
Enrichment, clustering, First Neighbor and adjacent disorders return raw results through
get_raw_result(); their get_genes() and get_drugs() are empty.
Use quiet=True to reduce output and drugstone_ids=True to retain internal IDs in
normalized results. Available datasets: drugstone.fetching.get_datasets().
from drugstone import fetching
task = drugstone.new_task(["EGFR", "GRB2"], {"algorithm": "pathway-enrichment", "target": "gene"})
rows = task.get_result().get_raw_result()["tableView"]
if rows:
row = rows[0]
pathway = fetching.get_pathway_network(task.get_token(), row["geneset"], row["pathway"], summary=True)Use source/pathway pairs from tableView; each call replaces the task's stored pathway
network. get_pathway_sources() lists sources. For associated disorders, use
new_task(seeds, {"algorithm": "adjacentDisorders"}); drug indications are available via
fetching.get_adjacent_disorders_of_drugs(["dr9859"]) using internal drug IDs.
from drugstone import fetching, network
fetching.get_tissues() # choose a tissue ID
fetching.get_tissue_expression(tissue="1", proteins=["EGFR"], identifier="symbol")
# Alternatively supply token=task.get_token() instead of proteins.Expression maps input identifiers to all matching proteins, returns values by internal
p… ID and uses None for missing data; unmapped inputs are logged. Also available:
get_tissue_proteins(tissue_id, threshold) and get_max_tissue_expression(tissue).
Mapped proteins already include cellular_component annotations where available.
Function in drugstone.network |
Purpose |
|---|---|
create_network(net, config=None, groups=None) |
Store a network; anyone with its ID can load it |
load_network(id) / get_network_link(id) |
Load it or get its web link |
get_pruning_range(nodes, attribute) / prune_network(net, attribute, cutoff=...) |
Inspect/filter node properties; filtered network is prunedNetwork |
get_pruning_distribution(nodes, attribute, path=None) |
Inspect values; optionally save a plot |
update_result_network(token, net) |
Replace a task result's displayed network |
overlay_directed_edges(net, identifier="symbol") |
Annotate edges with OmniPath direction and activation/inhibition |
recalculate_statistics(net, config) |
Return nodes with recalculated degree, clustering coefficient and SPD |
Networks use {"nodes": [...], "edges": [{"from": ..., "to": ...}]}. For statistics,
map nodes first with drugstone.map_nodes(). Config uses identifier,
interactionProteinProtein and interactionDrugProtein; enable custom_edges=True
and exclude_drugstone_ppi_edges=True to measure PPI degree on your own graph.
Only proteins in the selected background graph are evaluated; statistics are undirected.
Overlays check all proteins a node maps to; edges to unmapped or non-protein nodes and
edges absent from OmniPath are returned unchanged. See function docstrings for full options.
For more information about the algorithms, please refer to https://drugst.one/doc#implementation_algorithms.
For more information abouyt the available dataset types, please refer to https://drugst.one/doc#implementation_datasources.
Represents a task.
get_result() -> TaskResult
Returns a TaskResult for the result of the task.
get_info() -> dict
Returns a dict with information about the task.
get_parameters() -> dict
Returns a dict with the parameters of the task.
get_token() -> str
Returns the server task token, or None if no server task was started.
Represents the results of a task.
get_genes() -> dict
Returns a dict with the genes.
get_drugs() -> dict
Returns a dict with the drugs.
to_dict() -> dict
Returns a dict with the result.
to_pandas_dataframe() -> DataFrame
Returns a pandas DataFrame of the result.
download_json(path: str, name: str) -> None
Downloads a json file with the result.
download_genes_csv(path: str, name: str) -> None
Downloads a csv file with the genes of the result.
download_drugs_csv(path: str, name: str) -> None
Downloads a csv file with the drugs of the result.
download_edges_csv(path: str, name: str) -> None
Downloads a csv file with the edges of the result.
download_graph(path: str, name: str) -> None
Downloads a html file with a graph of the nodes.
Wraps a list of Task objects.
get_result() -> TasksResult
Returns a TasksResult for the list of tasks.
get_union() -> TaskResult
Returns a TaskResult with the union of the tasks.
get_intersection() -> TaskResult
Returns a TaskResult with the intersection of the tasks.
Represents the results of a list of Task objects.
get_tasks_list() -> List[Task]
Returns the list of tasks.
to_dict() -> dict
Returns a dict with the results of the tasks.
download_json(path: str, name: str) -> None
Downloads a json file with the results.
create_upset_plot() -> None
Opens a new window with an upset plot of the results.
This will fetch all available information for the given nodes from the Drugst.One database as a list of dictionaries. Each node contains a key 'drugstoneType', where the value is 'protein' if the given node can be mapped to a protein. If a node can not be mapped, the 'drugstoneType' will be 'other'.
Be aware of the parameter dictionary with the key 'identifier', available options are one of 'symbol' (HUGO symbol), 'uniprot' (Uniprot AC), 'ensg' or 'ensembl' (Ensembl Gene ID), 'entrez' or 'ncbigene' (Entrez ID), depending on your input.
import drugstone
nodes = [
"CFTR", "TGFB1", "SCNN1B", "justatest"
]
parameters = {'identifier': 'symbol'}
drugstone_nodes = drugstone.map_nodes(nodes, parameters)This will fetch all available edges for a given list of proteins. Returned will be a list of node entities. Each of these nodes contains the key 'hasEdgesTo' with a list of all node ids this node has an edge to.
Be aware of the parameter dictionary with the key 'identifier', available options are one of 'symbol' (HUGO symbol), 'uniprot' (Uniprot AC), 'ensg' or 'ensembl' (Ensembl Gene ID), 'entrez' or 'ncbigene' (Entrez ID), depending on your input.
import drugstone
# optional
drugstone.print_license()
drugstone.accept_license()
genes = [
"CFTR","TGFB1","TNFRSF1A","FCGR2A","ENG","DCTN4","CLCA4","STX1A","SCNN1G","SCNN1A","SCNN1B", "justatest"
]
parameters = {'identifier': 'symbol', 'ppiDataset': 'IID'}
network = drugstone.build_network(genes, parameters)Copyright: 2024 - Institute for Computational Systems Biology by Prof. Dr. Jan Baumbach.