Deep learning-based ranking method for subgroup and predictive biomarker identification in patients.

BACKGROUND: The task of identifying patient subgroups with enhanced treatment responses is important for clinical drug development. However, existing deep learning-based approaches often struggle to provide clear biological insights. This study aims to develop a deep learning method that not only captures treatment effect differences among individuals but also helps uncover meaningful biological markers associated with those differences.

METHODS: We introduce DeepRAB, a deep learning-based framework designed for exploring treatment effect heterogeneity by constructing individualized treatment rule (ITR). In addition, DeepRAB enables model interpretability by facilitating predictive biomarker identification. We evaluate its performance using simulated datasets that vary in complexity, treatment effect strength, and sample size. We also apply the method to the adalimumab (Humira, AbbVie) hidradenitis suppurativa (HS) clinical trial data, analyzing patient characteristics and treatment outcomes.

RESULTS: In analyses of simulated data under various scenarios, our findings show the effective performance of DeepRAB for subgroup exploration, and its capability to uncover predictive biomarkers when compared to existing approaches. When applied to the real clinical trial data, DeepRAB demonstrates its practical usage in identifying important predictive biomarkers and boosting model prediction performance.

CONCLUSIONS: Our research provides a promising approach for subgroup identification and predictive biomarker discovery by leveraging deep learning. This approach may support more targeted treatment strategies in clinical research and enhance decision-making in personalized medicine.