Scalable and universal prediction of cellular phenotypes

Input embeddings for Prophet, containing 115k treatments and 1.4k cell lines. Treatments are identified by their SMILES string for compounds or gene target for genetic perturbations. Cell lines are identified by their common name.Prophet模型输入嵌入集，包含11.5万种处理方式和1.4千种细胞系。处理方式通过化合物的SMILES字符串或基因靶点进行识别，而细胞系则通过其通用名称进行标识。