Skip to content

TxGNN Knowledge Graph Module

Overview

The TxGNN module maps altered genes onto a TxGNN-derived biomedical knowledge graph to identify drug candidates through disease-drug, drug-target, and FDA approval relationships.

How It Works

  1. The TxGNN knowledge graph is loaded (17,079 diseases, 7,957 drugs, 27,610 genes/proteins)
  2. For a given cancer type, drugs are identified through:
    • Indication edges: current treatments for the cancer type
    • Drug-protein edges: drugs targeting the mutated genes
    • FDA approval: drugs approved for other diseases (repurposing candidates)
  3. Candidates are scored using mutation-aware heuristic rules

Drug Categories

Category Description Base Score
Repurposing Priority FDA-approved for other diseases + targets mutated gene 20.0
Investigational Targeting Targets mutated gene, not FDA-approved 10.0
Current Indication Already indicated for this cancer type 5.0
Disease-Related Connected to cancer in the knowledge graph 2.0

Multi-gene targeting bonuses are added when a drug targets more than one mutated gene.

Key Output Fields

Field Description
drug Drug name (uppercase)
txgnn_score Heuristic prioritization score
category Drug category label
repurposing Whether it is a repurposing candidate
mutation_target Whether drug targets a mutated gene
fda_approved Whether drug is FDA-approved for any indication
connected_genes Mutated genes targeted by this drug

Note

This module uses the TxGNN knowledge graph as a structured resource for candidate discovery, not as a variant-specific deep-learning inference engine.

Usage

from txgnn_module import run_txgnn

txgnn_df = run_txgnn(
    maf_path="sample.maf",
    cancer_type="lung cancer",
    output_path="txgnn_output.tsv",
    data_folder="./data",
    txgnn_root="./TxGNN"
)