Addressing the Cold-Start Problem for Personalized Combination Drug Screening

2025-09-09

AI Summary

This work addresses the cold-start challenge in personalized combination therapy screening—specifically, how to efficiently select initial high-information drug combinations when no patient-specific prior information is available. Method: We propose an unsupervised experimental design framework that leverages pre-trained deep learning models to generate drug-combination embeddings and dose importance scores. It integrates clustering to ensure functional diversity and incorporates a dose-weighted mechanism to assimilate historical response data. Crucially, the framework requires no patient-specific inputs and relies solely on large-scale *in vitro* drug response datasets to build predictive and recommendation systems. Contribution/Results: Retrospective evaluations across multiple benchmark datasets demonstrate that our method significantly improves hit rates within the first N rounds of screening, outperforming random selection, synergy-driven, and single-agent sensitivity–based baselines by 23.6%–38.1% on average. The approach provides a scalable, data-efficient solution for combination drug discovery under cold-start conditions.

Technology Category

• Search and Optimization: Mixed Discrete/Continuous Search
• Machine Learning: Learning Preferences or Rankings
• Reasoning under Uncertainty: Stochastic Optimization

Application Category

• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs

Abstract

Personalizing combination therapies in oncology requires navigating an immense space of possible drug and dose combinations, a task that remains largely infeasible through exhaustive experimentation. Recent developments in patient-derived models have enabled high-throughput ex vivo screening, but the number of feasible experiments is limited. Further, a tight therapeutic window makes gathering molecular profiling information (e.g. RNA-seq) impractical as a means of guiding drug response prediction. This leads to a challenging cold-start problem: how do we select the most informative combinations to test early, when no prior information about the patient is available? We propose a strategy that leverages a pretrained deep learning model built on historical drug response data. The model provides both embeddings for drug combinations and dose-level importance scores, enabling a principled selection of initial experiments. We combine clustering of drug embeddings to ensure functional diversity with a dose-weighting mechanism that prioritizes doses based on their historical informativeness. Retrospective simulations on large-scale drug combination datasets show that our method substantially improves initial screening efficiency compared to baselines, offering a viable path for more effective early-phase decision-making in personalized combination drug screens.

Problem

• Addressing cold-start problem in personalized combination drug screening
• Selecting informative drug combinations without prior patient information
• Leveraging pretrained deep learning model for initial experiment selection

Innovation

• Pretrained deep learning model for drug embeddings
• Clustering drug combinations for functional diversity
• Dose-weighting mechanism prioritizing historically informative doses