Integrating single-cell atlases and machine learning to construct 'in silico patients' for predicting individualized drug responses.

Intratumoral cellular heterogeneity is a core challenge that drives drug resistance and hinders the advancement of precision oncology. Single-cell RNA sequencing (scRNA-seq) has revealed the complexity of the tumor ecosystem at unprecedented resolution, offering new opportunities for predicting therapeutic responses. This review synthesizes the emerging concept of the "in silico patient", a predictive framework that integrates multi-source data. This framework leverages large-scale single-cell atlases as references for cellular identity, combines massive pharmacogenomic databases to train models, and incorporates patient-specific scRNA-seq data to achieve individualized predictions. Artificial intelligence (AI), particularly deep learning and transfer learning algorithms, acts as the core driver of this framework, effectively applying knowledge gained from cell line data to clinically relevant patient single-cell data. By integrating the impact of the tumor microenvironment (TME) and using advanced preclinical models that preserve tissue architecture (such as acute tissue slice cultures) for rapid experimental validation, a critical "prediction-validation-optimization" closed loop is being formed. This review systematically outlines the data foundations, core computational strategies, current challenges, and future directions, including multi-omics and spatial information integration, necessary to construct "in silico patients", aiming to provide a comprehensive conceptual blueprint for developing the next generation of individualized drug response prediction tools.