Problem
Research questions and friction points this paper is trying to address.
Enhances HIV-1 therapy outcome prediction
Addresses data scarcity for new drugs
Improves robustness against imbalanced data
A graph neural network-based model with Out-of-Distribution Robustness for enhancing Antiretroviral Therapy Outcome Prediction for HIV-1
🤖 AI Summary
This work addresses poor generalizability of models for predicting HIV-1 antiretroviral therapy (ART) outcomes—caused by scarce novel-drug data and severe clinical distributional shift. We propose the first causal disentanglement framework integrating graph neural networks (GNNs) with out-of-distribution (OOD) robust learning. Our method constructs a patient–drug–virus heterogeneous relational graph, jointly employing causal representation disentanglement and neighborhood-aware adversarial training to explicitly mitigate confounding bias and enhance OOD robustness. Additionally, we incorporate an OOD detection and calibration module to improve prediction reliability. Evaluated on multicenter real-world cohorts, our model achieves an 8.3% absolute improvement in cross-institutional accuracy and a 12.7% gain in OOD-scenario AUC, significantly outperforming state-of-the-art baselines.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Graph Neural Network integration
Out-of-Distribution Robustness enhancement
Stanford drug-resistance mutation utilization
🔎 Similar Papers
