Abstract: Gene silencing is a well-known method to study how cell lines behave in the absence of one speci c gene. Recently, large scale experiments managed to knock-out each one of the human genes in hundreds of cell lines (Tsherniak et al., 2017; Dwane et al., 2020). These screenings generate a large amount of data that should be analyzed with appropriate methods.
As an example, classical statistical methods on cancer dependency maps lead to the identi cation of TRIM8 as an essential gene in fusion-driven Ewing sarcomas. Coupled with an experimental approach, it results in the biological explanation of the dependency and a better understanding of this pediatric cancer (Seong et al., 2021).
Also, machine learning methods such as random forests or penalized linear regressions could be used to predict gene dependencies, leading to potential biomarkers for tumor vulnerabilities (Dempster, et al., 2020).
In this talk, I will present how data is generated, and how machine learning can analyze these data to perform disease understanding, target identi cation or even indication selection.

Keywords: Omics Data; Machine Learning; Public Data; Oncology.

Dempster, et al. Gene expression has more power for predicting in vitro cancer cell vulnerabilities than genomics. BioRxiv (2020).
Dwane, et al. Project Score database: a resource for investigating cancer cell dependencies and prioritizing therapeutic targets. Nucleic Acids Research 49.D1 (2021): D1365-D1372.
Seong, et al. TRIM8 modulates the EWS/FLI oncoprotein to promote survival in Ewing sarcoma. Cancer cell 39.9 (2021): 1262-1278.
Tsherniak, et al. De ning a cancer dependency map. Cell 170.3 (2017): 564-576.