AI-Assisted Drug Target Identification via Network Biology

Abstract

Identifying druggable targets from the vast landscape of disease-associated genes requires methods that capture the network context in which proteins function, and artificial intelligence approaches that integrate multi-omics data with biological interaction networks now offer systematic target prioritisation beyond single-gene association, yet adoption of these AI-network methods into pharmaceutical discovery pipelines varies across approach categories. We evaluated 208 AI-assisted drug target identification programmes across computational biology centres affiliated with Nordic Technical University in Stockholm between 2016 and 2021, spanning five approach categories: graph neural network-based target scoring, network propagation with machine learning ranking, knowledge graph embedding for target-disease linking, multi-omics network integration models, and causal inference on molecular networks. A Network-AI Target Identification Index (NATII) was constructed from five sub-scores -- target novelty yield, biological validation rate, network prediction accuracy, computational scalability, and pharmaceutical pipeline integration -- with weights from regression against sustained adoption into active drug discovery workflows. NATII correlated with adoption at r = +0.83 and discriminated adopted from non-adopted methods with an AUC of 0.880. Knowledge graph embedding scored highest (mean NATII 0.822), while causal inference methods trailed at 0.596. Only 34.6 percent of programmes exceeded the 0.75 threshold. Target novelty yield carried the largest regression weight (beta = +0.278), followed by biological validation rate (beta = +0.230).