THBKG — Temporal Heterogeneous Biomedical Knowledge Graph

A dated biomedical knowledge graph built from Open Targets 26.09 (with Gene Ontology, Reactome, IntAct, the Human Protein Atlas and ClinicalTrials.gov), plus a clinical-advancement benchmark: rank target–disease pairs that entered Phase II by how likely they are to advance to Phase III, scored only from evidence dated strictly before each pair's decision year.

Every temporal edge carries the year its evidence first appeared, so the graph can be queried as of any historical decision point without leakage.

Files

Path	What it is
graph.safetensors	All graph tensors (node features, edge_index, edge_attr, edge_time), pickle-free.
metadata.json	Schema to reassemble the graph: node/edge types, tensor keys, shapes, dtypes.
mappings/<ntype>.parquet	node_id ↔ index for each node type (target, disease, molecule, go, reactome).
labels/train.parquet, labels/eval.parquet	Advancement labels: target_id, disease_id, transition_year, outcome.
load_thbkg.py	Standalone loader — rebuilds a PyG HeteroData, reads the label splits.
source_versions.json	Pinned upstream versions (Open Targets release, GO, IntAct, HPA).
croissant.json	MLCommons Croissant metadata.

Graph at a glance

• Nodes: 109,087 across five types — target (19,620), disease (45,896), molecule (2,596), Gene Ontology term go (38,092), Reactome pathway reactome (2,883). Feature dims: molecule 1024, disease 256, target 56, GO 64, Reactome 64.
• Edges: ~13.6M over nineteen relations (sixteen temporal, three static); each temporal edge carries edge_time (year first observed) and edge_attr = [weight, novelty].
• Clinical-trial edges are dated by outcome. Every registered trial adds an ongoing edge at its start date; a finished trial adds an outcome edge (positive, unmet efficacy, adverse effects, unknown/operational) at its actual completion date from ClinicalTrials.gov.
• Supervision is external. The graph holds only nodes and evidence edges; the clinical-advancement outcomes live in the label parquets and are joined to the graph by node ID at train/eval time.

Advancement benchmark

47,802 target–disease pairs, split temporally: 34,740 train (Phase II entry 1995–2015, 17.7% positive) / 13,062 eval (2016–2021, 7.9% positive). outcome is True iff the pair reached Phase III by a fixed horizon: 2018 for train, 2024 for eval. Eval entries stop at 2021 so every pair has at least three years of follow-up (the protocol of Czech et al.). Each pair is scored using only graph evidence dated strictly before its transition_year.

Primary metric: Relative Success @ K (an importance-weighted hit rate), reported per therapeutic area and Wilcoxon-tested against a randomized-decisions baseline.

Quick start

from huggingface_hub import snapshot_download
from load_thbkg import load_graph, load_labels, node_index

path  = snapshot_download("jspc00012567/THBKG", repo_type="dataset")
data  = load_graph(path)              # PyG HeteroData (reverse edges added, training view)
train = load_labels(path, "train")    # DataFrame: target_id, disease_id, transition_year, outcome
eval_ = load_labels(path, "eval")

# resolve a label endpoint to its graph node index
t2i, d2i = node_index(path, "target"), node_index(path, "disease")

The label parquets also load directly as a 🤗 datasets config:

from datasets import load_dataset
ds = load_dataset("jspc00012567/THBKG", "advancement")   # splits: train, eval

Versions

This is the 26.09 build. The previous 26.03 build stays available at revision v26.03:

snapshot_download("jspc00012567/THBKG", repo_type="dataset", revision="v26.03")

Results are not directly comparable across builds: Open Targets replaced many EFO disease IDs with MONDO IDs in 26.06 and merged duplicate terms in 26.09, and 26.09 has about 60% more clinical records. An earlier 23.06 build had a same-year clinical-trial-edge leak and is deprecated.

License & citation

Dataset artifacts: CC-BY-4.0. Source code (construction pipeline, benchmark harness): MIT, at https://github.com/jackysiupuichung/THBKG. Archival DOI on Zenodo: concept 10.5281/zenodo.20795231.

Siu, Cabrera, Mudaliar, and Zubiaga.
THBKG: A Temporal Biomedical Knowledge Graph for Leak-Free Clinical Advancement Prediction.