This study presents the first synthetic Cancer Dependency Map (DepMap) using a novel deep generative model. Harnessing the extensive DepMap multi-omic cancer datasets, we designed a variational auto-encoder capable of generating accurate and complete multi-omic profiles across over 1,500 cancer cell lines. It encompasses a wide range of data, including genomics, proteomics, metabolomics, and also includes data from pharmacological and CRISPR-Cas9 screens. Notably, the model excels at leveraging orthogonal multi-omic layers to provide complete de novo multi-omic characterization for cell lines lacking particular data types. Thorough benchmarking of the synthetic data, involving comparisons with independent datasets from various laboratories, confirmed its quality and ability to accurately reflect true biological associations. Additionally, this process highlighted new potential mechanisms involved in drug resistance.

Reference

Cai, Zhaoxiang, Sofia Apolinário, Ana R. Baião, Clare Pacini, Miguel D. Sousa, Susana Vinga, Roger R. Reddel, Phillip J. Robinson, Mathew J. Garnett, Qing Zhong, and Emanuel Gonçalves. 2024. ‘Synthetic Augmentation of Cancer Cell Line Multi-Omic Datasets Using Unsupervised Deep Learning’. Nature Communications 15 (1). Nature Publishing Group: 1–12.