Gene Expression Profiling Predicts Clinical Outcome of Breast Cancer - - PowerPoint PPT Presentation

▶
gene expression profiling predicts clinical outcome of
SMART_READER_LITE
LIVE PREVIEW

Gene Expression Profiling Predicts Clinical Outcome of Breast Cancer - - PowerPoint PPT Presentation

Dec 11, 2023 368 likes •658 views

Gene Expression Profiling Predicts Clinical Outcome of Breast Cancer Motivation Observation : breast-cancer patients at same stage have different outcomes Problem : existing outcome predictors are poor - lymph nodes - histological grade

slide-1
SLIDE 1

Gene Expression Profiling Predicts Clinical Outcome of Breast Cancer

slide-2
SLIDE 2

Motivation

Observation: breast-cancer patients at same stage have different outcomes Problem: existing outcome predictors are poor


  • lymph nodes

  • histological grade

Goals:


  • identify categories of breast cancer

  • predict outcome based on gene expression

  • decide therapy accordingly
slide-3
SLIDE 3

Treatment

“Chemotherapy or hormonal therapy reduces risk of distant metastases by approximately one-third; however 70-80% of patients receiving this treatment would have survived without it.” A main contribution of this method is lower false positives.

slide-4
SLIDE 4

Result Overview

slide-5
SLIDE 5

Result Overview

needs therapy gets therapy

slide-6
SLIDE 6

Recommend Therapy % 25 50 75 100 St Gallen NIH Prognosis Profile Recommend Therapy % 25 50 75 100 St Gallen NIH Prognosis Profile

needs therapy does not need therapy

Result Overview

slide-7
SLIDE 7

Study

  • 1. unsupervised clustering, look for tumor categories
  • 2. supervised learning, find prognosis reporter genes
slide-8
SLIDE 8

Study

  • 1. unsupervised clustering, look for tumor categories
  • 2. supervised learning, find prognosis reporter genes
slide-9
SLIDE 9

Unsupervised Hierarchical Clustering: Dendogram

http://youtu.be/XJ3194AmH40?t=5m

slide-10
SLIDE 10
slide-11
SLIDE 11

2 categories

slide-12
SLIDE 12

4 categories

slide-13
SLIDE 13

page 531

slide-14
SLIDE 14

page 531

co-regulates with ER-α

slide-15
SLIDE 15

page 531

slide-16
SLIDE 16

page 531

co-regulates with lymphocytic infiltrate

slide-17
SLIDE 17

page 531

slide-18
SLIDE 18

Study

  • 1. unsupervised clustering, look for tumor categories
  • 2. supervised learning, find prognosis reporter genes
slide-19
SLIDE 19

Study

  • 1. unsupervised clustering, look for tumor categories
  • 2. supervised learning, find prognosis reporter genes
slide-20
SLIDE 20

Method

Start with 25K genes (some double counting) From these, identify 5K that are significantly regulated From these, identify 231 significantly associated with disease outcome From these, identify 70 as classification features

slide-21
SLIDE 21

Supervised Learning on Prognosis Signatures

slide-22
SLIDE 22

page 532

Training Data (78 tumors)

slide-23
SLIDE 23

page 532

Test Data (19 tumors)

slide-24
SLIDE 24

Supervised Learning on ER and BRCA1 Signatures

slide-25
SLIDE 25

page 533

slide-26
SLIDE 26

page 533

slide-27
SLIDE 27

Summary

New ideas (for 826):


  • new label: clinical outcome

  • use of unsupervised learning

  • accuracy vs. sensitivity tradeoffs