腾讯发表Cell论文：用AI打通基因与药物筛选壁垒，加速AI虚拟细胞实现
Merged summary
TL;DR - A Cell paper introduces UniPert-G2CP, an AI framework that transfers knowledge from large-scale genetic screens to predict chemical perturbation effects. It could reduce experimental drug-screening costs and advance AI virtual-cell modeling.
- UniPert aligns protein and small-molecule representations in a shared embedding space using multimodal features, known compound–protein interactions, and contrastive learning.
- G2CP pretrains on abundant CRISPR screening data, then fine-tunes on scarcer chemical data to predict drug-induced transcriptomic and morphological changes.
- The framework reportedly outperformed existing methods on datasets including LINCS and sciPlex3, with gains of up to 375% in data-scarce settings.
- A breast-cancer case study reproduced estrogen-receptor mutation-associated drug resistance and identified potentially involved pathways such as MYC and DNA repair.
Sources (1)
腾讯发表Cell论文：用AI打通基因与药物筛选壁垒，加速AI虚拟细胞实现
TL;DR - A Cell paper introduces UniPert-G2CP, an AI framework that transfers knowledge from large-scale genetic screens to predict chemical perturbation effects. It could reduce experimental drug-screening costs and advance AI virtual-cell modeling.
- UniPert aligns protein and small-molecule representations in a shared embedding space using multimodal features, known compound–protein interactions, and contrastive learning.
- G2CP pretrains on abundant CRISPR screening data, then fine-tunes on scarcer chemical data to predict drug-induced transcriptomic and morphological changes.
- The framework reportedly outperformed existing methods on datasets including LINCS and sciPlex3, with gains of up to 375% in data-scarce settings.
- A breast-cancer case study reproduced estrogen-receptor mutation-associated drug resistance and identified potentially involved pathways such as MYC and DNA repair.