Unboxing Cluster Heatmaps Sophie Engle Sean Whalen Alark Joshi - PowerPoint PPT Presentation

Unboxing Cluster Heatmaps Sophie Engle Sean Whalen Alark Joshi Katherine Pollard vgl.cs.usfca.edu docpollard.org Visualization and Graphics Lab Pollard Group University of San Francisco Gladstone Institutes, UCSF

2 Contributions • Practitioner Survey • 45 practitioners • Heatmap & dendrograms • Symmetric & asymmetric • 100 to 250,000 cells • R & Cytoscape Unboxing Cluster Heatmaps by Engle, Whalen, Joshi, and Pollard Visualization and Graphics Lab, Department of Computer Science, University of San Francisco 6th Symposium on Biological Data Visualization (BioVis), October 23, 2016 Pollard Group, Gladstone Institutes, University of California San Francisco

3 Contributions • Practitioner Survey • Unboxing Approach Unboxing Cluster Heatmaps by Engle, Whalen, Joshi, and Pollard Visualization and Graphics Lab, Department of Computer Science, University of San Francisco 6th Symposium on Biological Data Visualization (BioVis), October 23, 2016 Pollard Group, Gladstone Institutes, University of California San Francisco

4 Contributions • Practitioner Survey • Unboxing Approach • Pair Analytics Unboxing Cluster Heatmaps by Engle, Whalen, Joshi, and Pollard Visualization and Graphics Lab, Department of Computer Science, University of San Francisco 6th Symposium on Biological Data Visualization (BioVis), October 23, 2016 Pollard Group, Gladstone Institutes, University of California San Francisco

5 Contributions • Practitioner Survey • Unboxing Approach • Pair Analytics • Practitioner Interviews Unboxing Cluster Heatmaps by Engle, Whalen, Joshi, and Pollard Visualization and Graphics Lab, Department of Computer Science, University of San Francisco 6th Symposium on Biological Data Visualization (BioVis), October 23, 2016 Pollard Group, Gladstone Institutes, University of California San Francisco

6 Contributions • Practitioner Survey • Unboxing Approach • Pair Analytics • Practitioner Interviews Which two elements are • Large-Scale User Study more closely clustered? Unboxing Cluster Heatmaps by Engle, Whalen, Joshi, and Pollard Visualization and Graphics Lab, Department of Computer Science, University of San Francisco 6th Symposium on Biological Data Visualization (BioVis), October 23, 2016 Pollard Group, Gladstone Institutes, University of California San Francisco

7 Contributions • Practitioner Survey 45 • Unboxing Approach • Pair Analytics • Practitioner Interviews 5 • Large-Scale User Study 200 Unboxing Cluster Heatmaps by Engle, Whalen, Joshi, and Pollard Visualization and Graphics Lab, Department of Computer Science, University of San Francisco 6th Symposium on Biological Data Visualization (BioVis), October 23, 2016 Pollard Group, Gladstone Institutes, University of California San Francisco

8 Contributions • Practitioner Survey 45 • Unboxing Approach • Pair Analytics • Practitioner Interviews 5 • Large-Scale User Study 200 Unboxing Cluster Heatmaps by Engle, Whalen, Joshi, and Pollard Visualization and Graphics Lab, Department of Computer Science, University of San Francisco 6th Symposium on Biological Data Visualization (BioVis), October 23, 2016 Pollard Group, Gladstone Institutes, University of California San Francisco

Unboxing Cluster Heatmaps github.com/usfvgl/unboxing-cluster-heatmaps git.io/vw 0 t3 Visualization and Graphics Lab, Department of Computer Science, University of San Francisco Pollard Group, Gladstone Institutes, University of California San Francisco

10 Left: Cluster Heatmap/Gapmap, Middle: Sunburst/Circle Packing, Right: Radial Dendrogram/Force-Directed Tree Unboxing Cluster Heatmaps by Engle, Whalen, Joshi, and Pollard Visualization and Graphics Lab, Department of Computer Science, University of San Francisco 6th Symposium on Biological Data Visualization (BioVis), October 23, 2016 Pollard Group, Gladstone Institutes, University of California San Francisco

11 Left: Cluster Heatmap/Gapmap, Middle: Sunburst/Circle Packing, Right: Radial Dendrogram/Force-Directed Tree Unboxing Cluster Heatmaps by Engle, Whalen, Joshi, and Pollard Visualization and Graphics Lab, Department of Computer Science, University of San Francisco 6th Symposium on Biological Data Visualization (BioVis), October 23, 2016 Pollard Group, Gladstone Institutes, University of California San Francisco

12 mTurk Study Design Which two of the highlighted elements are more closely clustered? 5 by 7 circle packing Unboxing Cluster Heatmaps by Engle, Whalen, Joshi, and Pollard Visualization and Graphics Lab, Department of Computer Science, University of San Francisco 6th Symposium on Biological Data Visualization (BioVis), October 23, 2016 Pollard Group, Gladstone Institutes, University of California San Francisco

13 mTurk Study Design Which two of the highlighted elements are more closely clustered? 5 by 7 radial dendrogram Unboxing Cluster Heatmaps by Engle, Whalen, Joshi, and Pollard Visualization and Graphics Lab, Department of Computer Science, University of San Francisco 6th Symposium on Biological Data Visualization (BioVis), October 23, 2016 Pollard Group, Gladstone Institutes, University of California San Francisco

14 mTurk Study Design Which two of the highlighted elements are more closely clustered? 8 by 16 cluster heatmap Unboxing Cluster Heatmaps by Engle, Whalen, Joshi, and Pollard Visualization and Graphics Lab, Department of Computer Science, University of San Francisco 6th Symposium on Biological Data Visualization (BioVis), October 23, 2016 Pollard Group, Gladstone Institutes, University of California San Francisco

15 mTurk Study Design Which two of the highlighted elements are more closely clustered? 8 by 16 gapmap Unboxing Cluster Heatmaps by Engle, Whalen, Joshi, and Pollard Visualization and Graphics Lab, Department of Computer Science, University of San Francisco 6th Symposium on Biological Data Visualization (BioVis), October 23, 2016 Pollard Group, Gladstone Institutes, University of California San Francisco

16 mTurk Study Results Q13 − Close Siblings Q14 − Close Leaves Q15 − Close Clusters Q16 − Least Similar Circle Packing Cluster Heatmap Force Directed Tree Gapmap Radial Dendrogram Sunburst 0 0.5 1 0 0.5 1 0 0.5 1 0 0.5 1 CH-SB 0.01, CH-RD 0.02 CH-RD 0.041 GM-RD 0.040 CP-SB 0.01, CP-RD 0.02 Average Score Violin Plot. Circle: Average Score, Bar: Standard Error, Dotted Line: Random Performance (w/o “Unsure”) Unboxing Cluster Heatmaps by Engle, Whalen, Joshi, and Pollard Visualization and Graphics Lab, Department of Computer Science, University of San Francisco 6th Symposium on Biological Data Visualization (BioVis), October 23, 2016 Pollard Group, Gladstone Institutes, University of California San Francisco

17 mTurk Study Results Q13 − Close Siblings Q14 − Close Leaves Q15 − Close Clusters Q16 − Least Similar Circle Packing Cluster Heatmap Force Directed Tree Gapmap Radial Dendrogram Sunburst 0 0.5 1 0 0.5 1 0 0.5 1 0 0.5 1 CH-SB 0.01, CH-RD 0.02 CH-RD 0.041 GM-RD 0.040 CP-SB 0.01, CP-RD 0.02 Average Score Violin Plot. Circle: Average Score, Bar: Standard Error, Dotted Line: Random Performance (w/o “Unsure”) Unboxing Cluster Heatmaps by Engle, Whalen, Joshi, and Pollard Visualization and Graphics Lab, Department of Computer Science, University of San Francisco 6th Symposium on Biological Data Visualization (BioVis), October 23, 2016 Pollard Group, Gladstone Institutes, University of California San Francisco

18 mTurk Study Results Q13 − Close Siblings Q14 − Close Leaves Q15 − Close Clusters Q16 − Least Similar Circle Packing Cluster Heatmap Force Directed Tree Gapmap Radial Dendrogram Sunburst 0 0.5 1 0 0.5 1 0 0.5 1 0 0.5 1 CH-SB 0.01, CH-RD 0.02 CH-RD 0.041 GM-RD 0.040 CP-SB 0.01, CP-RD 0.02 Average Score Violin Plot. Circle: Average Score, Bar: Standard Error, Dotted Line: Random Performance (w/o “Unsure”) Unboxing Cluster Heatmaps by Engle, Whalen, Joshi, and Pollard Visualization and Graphics Lab, Department of Computer Science, University of San Francisco 6th Symposium on Biological Data Visualization (BioVis), October 23, 2016 Pollard Group, Gladstone Institutes, University of California San Francisco

19 mTurk Study Results Q17 − Visual Clusters Circle Packing Cluster Heatmap Force Directed Tree Gapmap Radial Dendrogram Sunburst 2 8 14 20 Average Clusters Violin Plot. Circle: Average Estimate, Bar: Standard Error Unboxing Cluster Heatmaps by Engle, Whalen, Joshi, and Pollard Visualization and Graphics Lab, Department of Computer Science, University of San Francisco 6th Symposium on Biological Data Visualization (BioVis), October 23, 2016 Pollard Group, Gladstone Institutes, University of California San Francisco

20 Conclusions • Gapmaps are a good alternative for cluster heatmaps - Preferred by interviewed practitioners - Performed well in mTurk user study • Involving practitioners at multiple stages critical - Caught places were assumptions were incorrect • Symmetric matrices need (much) more study Unboxing Cluster Heatmaps by Engle, Whalen, Joshi, and Pollard Visualization and Graphics Lab, Department of Computer Science, University of San Francisco 6th Symposium on Biological Data Visualization (BioVis), October 23, 2016 Pollard Group, Gladstone Institutes, University of California San Francisco