SLInterpreter: An Exploratory and Iterative Human-AI Collaborative System for GNN-based Synthetic Lethal Prediction

TL;DR

SLInterpreter presents an iterative Human-AI collaboration framework to improve interpretability and mechanism analysis in GNN-based synthetic lethal prediction. It combines Human-Engaged Knowledge Graph Refinement via metapath strategies with Cross-Granularity SL Interpretation to align AI outputs with domain knowledge. The approach is instantiated in a backend GNN atop a SynLethDB/ProteinKG25 knowledge graph and a frontend with Session, Embedding, Interpretation, and Modifier views, enabling targeted data cleaning, path-level reasoning, and granularity-aware exploration. Case studies and expert interviews demonstrate improved interpretability, more efficient expert workflows, and the potential to identify novel SL relationships while mitigating noise and data reliability concerns.

Abstract

Synthetic Lethal (SL) relationships, though rare among the vast array of gene combinations, hold substantial promise for targeted cancer therapy. Despite advancements in AI model accuracy, there is still a significant need among domain experts for interpretive paths and mechanism explorations that align better with domain-specific knowledge, particularly due to the high costs of experimentation. To address this gap, we propose an iterative Human-AI collaborative framework with two key components: 1) Human-Engaged Knowledge Graph Refinement based on Metapath Strategies, which leverages insights from interpretive paths and domain expertise to refine the knowledge graph through metapath strategies with appropriate granularity. 2) Cross-Granularity SL Interpretation Enhancement and Mechanism Analysis, which aids experts in organizing and comparing predictions and interpretive paths across different granularities, uncovering new SL relationships, enhancing result interpretation, and elucidating potential mechanisms inferred by Graph Neural Network (GNN) models. These components cyclically optimize model predictions and mechanism explorations, enhancing expert involvement and intervention to build trust. Facilitated by SLInterpreter, this framework ensures that newly generated interpretive paths increasingly align with domain knowledge and adhere more closely to real-world biological principles through iterative Human-AI collaboration. We evaluate the framework's efficacy through a case study and expert interviews.

Figures (7)

• Figure 1: SL gene pairs refer to a pair of genes where inhibiting one does not impact cancer cell survival, but inhibiting both leads to cell death.
• Figure 2: SLInterpreter comprises data processing module, backend engine, and frontend interface, facilitating an iterative workflow.
• Figure 3: Glyph Designs for Embedding View. is the overview of Radar in disease searching. is the enlarged Radar. is the Rank Indicator.
• Figure 4: Design alternatives for Interpretation View. is an alternative based on Sankey and Packchart. is an alternative based on Sankey and node matrix. is the final design based on Voronoi Treemap.
• Figure 5: Metapath Modifier design for Modifier View. Design for the Entity-level Summary. Design for the Metapath Strategy. Design for the Path-level Summary. Granularity Level Flow for the Path-level Summary, H represents High Granularity and L represents Low Granularity.