VARI Internship Presentation

Ally Rogers
July 24, 2014

Overview

Lab Projects

  • Comparing CA 19-9 with a multi-marker panel

  • Processing data from MSS

Xenograft Tumor Modeling

qPCR Analysis

Intro to R – Markdown

LAB PROJECTS

Lab Project 1 - Compare Biomarkers

  • CA 19-9
    • Best current biomarker
      • False positives
  • Multi-marker panel
    • Correct detection in more patients

Methods

  • ROC Curves
    • Comparing the AUC

Sensitivity vs. Specificity

sens-spec-plot

Compare Area Under Curves


  • Biomarker Panels
    • Integrals
  • CA19-9
    • R function for ROC objects


AUC
Panel 4516 0.9101317
Panel 4819 0.9373238
Panel 41618 0.9407960
Panel 41923 0.9121831
CA 19-9 0.9165469

Compare Accuracy

  • Sensitivity and Specificity
    • McNemar
      • Sensitivity using cancer samples
      • Specificity using non-cancer samples
    • Confidence Intervals
      • Sesp.diff.ci() in the DTComPair package
  • Accuracy
    • Comparison of two proportions
      • Binomial (exact) CI's

Results - Panel vs CA 19-9

Panel CA199 Diff. 95% CI PValue
4516 Sensitivity 0.8793 0.8362 0.043 (-0.03,0.116) 0.3588
Specificity 0.9583 0.8611 0.0972 (0.0098,0.185) 0.07044
4819 Sensitivity 0.8793 0.8362 0.043 (-0.0219,0.108) 0.3017
Specificity 0.9444 0.8611 0.083 (-0.0006,0.167) 0.1138
41618 Sensitivity 0.8793 0.8362 0.043 (-0.0219,0.108) 0.3017
Specificity 0.9444 0.8611 0.083 (-0.0006,0.167) 0.1138
41923 Sensitivity 0.8879 0.8362 0.052 (-0.006,0.109) 0.1489
Specificity 0.9306 0.8611 0.069 (-0.011,0.150) 0.1824
5Marker Sensitivity 0.9397 0.8362 0.10345 (0.0385,0.1684) 0.00596
Specificity 0.9306 0.8611 0.0694 (-0.011,0.1495) 0.1824

Take-Away

  • Understand the data
    • Communication
  • Sensitivity and Specificity
    • Better understanding of how they are calculated
  • Bootstrap Methods

Processing MSS Data

  • Usually done in Excel

  • R Scripts

    • Cross Validation
    • Marker Selection
    • Threshold Selection

Cross Validation

cross-valid

  • Data set of over 500,000 entries
    • VERY slow
      cross-valid

Marker & Threshold Selection


  • Marker Selection
    • Overall accuracy, keeps top 2, counts number of times panels occur
  • Threshold Determination
    • Determines best thresholds for each marker
    • Produces all combinations of top thresholds


thresh-plot1

thresholds-output

Detailed Files

thresholds-output thresholds-output

Detailed File Output

thresholds-output

Detailed File Again

  • Huiyuan wanted a cleaner file to sort through

cleaner

Detailed File Again


  • Brian wanted something that was easier to view
    • Less information to look through
    • Top 2 accuracy values only


Count

Final Output

detailedTop2

Take-away

  • Data manipulations with R

  • Creating output files with R

Xenograft Tumor Modeling

  • Effect of treatment on tumor growth

  • Four groups, 31 mice total

Treatment Count
Erlotinib (75mg/kg) 8
MGCD + Erlotinib 8
MGCD265 (60 mg/kg) 8
Vehicle 7

Exploratory Plot

individ-mice

Linear Mixed Effects Model


  • Fixed effects estimates for average tumor size
  • Model fits line even when there are no mice left to observe


fit-model

Take-Away

  • Introduced to ggplot
    • Creating plots in R
  • Linear mixed effects
  • R-markdown

qPCR Analysis

1st Project:

  • 2 Timepoints
    • Compare populations within each gene separately

2nd Project:

  • One set of data
    • Compare populations within each gene separately

Project 1

time-9-data

time-11-data

Project 1 - Continued


  • Not enough power for statistical significance
  • Kruskal-Wallis
    • Pairwise comparisons with FDR correction
      • Wilcoxon
      • T-test

barplot

Project 2

proj2data

Project 2 - Continued

  • Not enough power for statistical significance
  • Kruskal-Wallis
    • Pairwise comparisons with FDR correction
      • Wilcoxon
      • T-test

Take-Away

  • More introduction to ggplot
  • Non-parametric tests

Final Project

Intro to R web-page with R Markdown

Take-Away

  • R-MARKDOWN
  • Understanding of ggplot
  • Statistical tests using R