Joint Analysis of Multiple Cancer Types for Revealing Disease-Specific Genomic Events
summary
Joint analysis uses a curated compendium of 13,000 GEO profiles (170 datasets, 17 platforms) and COSMIC whole‑exome calls (~9,000 patients plus TCGA) mapped to Disease Ontology terms to learn biomarkers that are different from tissue controls and the massive background. Key claims: separating positive vs background controls prevents tissue confounding, simple SVMs beat complex multi‑label methods on expression, and a downsampled‑background random forest performs best on mutations. Outcome: classifiers robustly separate roughly 20% of tested DO terms, but marker reproducibility remains low (Jaccard ~0.2) and more data are needed.
AI-written summary · about
chapters (23)
- 0:00Intro
- 1:02Outline
- 1:36Goals
- 2:39The analysis process
- 4:05Multi-label classification
- 4:41Evaluating the algorithms • Use leave-dataset-out cross-validation Classifier output
- 6:02An embarrassment of riches
- 8:06Solution: stricter evaluation, multiple criteria
- 9:10Comparing classifiers (2)
- 9:3124 well-classified diseases
- 9:48Independent validation: RNA-seq
- 10:37Robustness
- 12:06Selecting biomarker genes For disease D, gene G, rank samples by G's expression
- 12:57Projecting the cancer biomarker genes on the PPI network • 258 cancer-specific genes • The two largest components in the PPI graph
- 14:01Summary (1)
- 16:32Data selection
- 17:32Tested algorithms
- 18:40LDO Results
- 20:10TCGA vs. non TCGA
- 21:44Interpreting the classifiers
- 22:34Intestinal cancer: nice agreement with colon
- 24:26Benign: non TCGA example
- 25:02Summary (2) TCGA is not enough - COSMIC adds value
related talks
30mUnlocking the Genome: Exploring the Potential of CRISPR/Cas9
12mA Complexity-Theoretic Perspective on Fairness
33mEvolution of Music, Cantometrics
26mTed Willke, Senior Principal Engineer, Intel Labs
58mStrong Generalization from Small Brains and No Training Data
17mSimulation of Transient Effects in Accelerator Magnets
31mAnswering FAQs in CSPs, PGMs, Databases, Logic and Matrix Operations
01h 01mThe Gospel According to TCS