From cells to drug responses - machine learning in cancer research
summary
Predicts drug response from tumor sequencing using Snakemake/Bioconda to build reproducible pipelines, fits simple linear models with elastic net and tandem two-stage residual regression, and confronts very high dimensional data ~20,000 features vs ~1,000 samples and strong cross-type correlations. Finds gene expression alone explains most predictive signal but the tandem approach increases selection of upstream DNA features by ~50% without losing predictive performance on cell‑line tumor assays.
AI-written summary · about
chapters (10)
- 0:00ML in cancer research
- 1:25Cancer biology and tumor sequencing
- 7:26Tumor sequencing data generation
- 10:33Reproducible workflows with Snakemake
- 15:30Drug response from genomics
- 16:35High-dimensional modeling challenges
- 19:27DNA and RNA data limits
- 21:58Two-stage interpretable modeling
- 23:44Future of cancer genomics
- 26:27Q&A
related talks
29mMachine Learning Techniques for Omics Data Analysis
32mHow machine learning helps cancer research
30mMachine Learning: A New Approach to Drug Discovery
30mLeveraging Molecular Data for Drug Discovery
01h 31mIntroduction to Cancer Bioinformatics I: Inferring Genomic Variation from Tumor Sequencing Data
03h 51mMachine Learning Summit
12mHow data science can make your genome actionable
20mAdvancing Epigenetic Predictors with Scalable Machine Learning