A Pooled Cell Painting CRISPR Screening Platform Enables de novo Inference of Gene Function by Self-supervised Deep Learning

Pooled CRISPR screening enables large-scale interrogation of gene functions but typically measures simple phenotypes such as fitness. High-content methods like Perturb-seq extend dimensionality to transcriptomics but are costly and limited in scope. Optical pooled screening (OPS) combines pooled CRISPR screening with imaging to yield scalable, information-rich readouts, yet existing implementations remain pathway-specific. Here we describe an OPS-compatible Cell Painting platform that enables hypothesis-free reverse genetic screening through multiplexed morphological profiling. We validate this technique using a well-defined morphological gene set, compare classical image analysis to self-supervised learning methods using a mechanism-of-action library, and perform discovery screening with a druggable genome library. By combining rich morphological data with deep learning, gene networks emerge without the need for target-specific biomarkers, leading to unbiased discovery of gene functions.

混合CRISPR筛选（Pooled CRISPR Screening）可实现基因功能的大规模解析，但通常仅能检测诸如适应性之类的简单表型。诸如扰动测序（Perturb-seq）这类高内涵技术可将检测维度拓展至转录组学层面，但成本高昂且应用范围受限。光学混合筛选（Optical Pooled Screening, OPS）将混合CRISPR筛选与成像技术相结合，可获得兼具规模化与信息丰富度的检测结果，但现有应用仍局限于特定通路。本研究报道了一种适配OPS的细胞绘画（Cell Painting）平台，可通过多重形态谱分析实现无假设的反向遗传筛选。我们通过一组特征明确的形态学基因集对该技术进行验证，借助作用机制文库（Mechanism-of-Action Library）对比经典图像分析与自监督学习（Self-Supervised Learning）方法，并利用可成药基因组文库（Druggable Genome Library）开展发现型筛选。通过将丰富的形态学数据与深度学习相结合，无需靶向特异性生物标志物即可构建基因网络，从而实现基因功能的无偏发现。