Interactive Dashboard for TrPLet - Transcriptional Prediction of Lethality

Large-scale functional genetic screens have enabled the discovery of many selective cancer dependencies. However, rare cancers have often been underrepresented in prior efforts, and some cancer models may not be amenable to high-throughput screening.

This project aims to predict gene dependency scores based on tumor or cell line transcriptional profile (RNA-seq).

This interactive dashboard is designed to enable users to explore predicted dependencies across 11K+ TCGA tumor samples, 850+ kidney tumors across 13 different molecular subtypes, and 600+ previously unscreened cancer cell lines.

We also provide visualizations to report performance for each gene dependency prediction model as well as to explore top features driving each model.

The application consists of multiple interactive tabs:

• Dependency Visualizations: Interactive visualization of dependency scores predicted based on RNA-seq data from the TCGA, other tumor cohorts, or cell line models. Experimentally-derived dependency scores from the Cancer DepMap are also shown for comparison by lineage (CRISPR KO, 23Q2).
• Model & Feature Selection: Compare model performance based on different number of features and gene models for a given gene of interest.
• Feature Importance: Explore the contribution of each feature to the dependency prediction for 6283 highly predictable genes (all with r ≥ 0.2 between predicted and experimentally observed).
• Prediction Consistency: Assess the performance of the best model (r) for all 16K+ genes.
• Acknowledgement & License: Show acknowledgement, data sources, and the project's copyright & license information.