Skip to main content

Integrating Kinase Bioactivity with Expression Profiles​

Here I showcase a translational knowledge graph that integrates reference datasets using semantic web technologies to support cross-domain questions in oncology target discovery.

As a proof of concept, it connects:

  • Kinase-targeted bioactivity (ChEMBL; compound–target potency and assay context)
  • Normal tissue expression (GTEx)
  • Tumour expression (TCGA)
  • Identifier harmonisation (UniProt-centered mappings to Ensembl and anatomy terms)

All data are represented as RDF and loaded into a GraphDB triple store as named graphs, enabling reproducible semantic joins across datasets without pre-materialising every combination.

Pipeline Overview​

Integration Pipeline

Each source dataset (ChEMBL, GTEx, TCGA, UniProt) is processed independently and exported to Turtle (TTL), then loaded into GraphDB as separate named graphs. Datasets are harmonised through ontology-aligned identifiers (e.g. UniProt ↔ Ensembl; anatomy terms such as UBERON) to make cross-dataset joins explicit and inspectable.

Compound → Kinase Target → Gene → Expression (GTEx / TCGA)

This enables questions such as:

  • "Which compounds inhibit kinases whose genes are overexpressed in kidney tumours but not in healthy kidney cortex?"
  • "Which kinase targets show tumour-specific expression patterns across multiple tissues?"

The Human Protein Atlas is included as a planned extension to add protein-level validation and reduce reliance on transcript-only signals.

Use Cases​

This integrated knowledge graph supports semantic exploration across compounds, kinase targets, gene identifiers, and expression profiles in normal and tumour tissues. Example use cases include:

  • Prioritise kinase targets using tumour-specific overexpression signals
  • De-risk targets by filtering for low expression in critical healthy tissues (toxicity proxy)
  • Rank compounds by the expression of their target set in a selected cancer type
  • Compare expression contrast across tissues (e.g. GTEx vs. TCGA) to identify specificity
  • Assess compound–target coverage across tumour types for combination or repositioning strategies