Domain Problem Domain Problem Domain Problem = biological receptor = biological receptor = biological receptor = biological target = biological target = biological target UNDERSTANDING DRUG DISCOVERY UNDERSTANDING DRUG DISCOVERY UNDERSTANDING DRUG DISCOVERY Scenario 1: = chemical compound Scenario 2: = chemical compound Scenario 3: = chemical compound Targeted interaction, understood Indirect interaction, understood Complex interactions, mechanism mechanism, desired outcome = potential target mechanism, desired outcome = potential target poorly understood, multiple outcomes = potential target = direct target = direct target = direct target ConTour: Data-Driven Exploration of Multi- = inhibits target = inhibits target = inhibits target Relational Datasets for Drug Discovery Result Result Result Christian Partl, Alexander Lex, Marc Streit, Hendrik Strobelt, Anne- Interaction Compound Interaction Compound Interaction Compound (Phenotype) (Phenotype) (Phenotype) MaiWassermann, Hanspeter Pfister and Dieter Schmalstieg Domain Problem ConTour Data Abstraction Data Abstraction History View Pathway View Compound View Drug Discovery Main Goals derived* • Identify a drug’s mechanism of action • Identify the biological process a drug modulates • Identify new drugs for specific therapeutic derived indications “The drug discovery domain problem can be generalized to the problem of analysing multi-relational datasets […] Consequently, we argue that our approach is applicable to many other problems .” Relationship View * Derived using a scheme propose by the Prous Integrity database Filter View Data Abstraction Task Analysis Task Analysis Task Analysis • T1 : Identify Related Items • T1 : Identify Related Items • T1 : Identify Related Items Item selection and highlighting Selection-based filters Nesting Clicking, not hovering, on an item also moves Filter choices when multiple items are selected Simple Nesting all related items in columns to the top “The multi-relational data exploration problem can be interpreted as a graph exploration problem where each item of each dataset represents a node and the relationships between the items are the edges” Task Analysis Task Analysis Task Analysis Task Analysis • T1 : Identify Related Items • T2: Identify Items that Share a • T3: Analyse Network Enrichment • T4: Rank Items Relationships with a Set of Items Nesting Nesting Enrichment Score Sorting by interest Enrichment Score Sort alpha-numerically Recursive Nesting Judging how specific two items are Sort by enrichment score Simple Nesting Recursive Nesting when compared to a third compounds compounds Where : I = clusters K = compounds J = Pathways S(i,j) = pair score clusters pathways clusters pathways *I assume they take care of divide by 0?
Task Analysis Task Analysis Task Analysis Task Analysis • T4: Rank Items • T5: Filter Items • T5: Filter Items • T6 : View items in detail Depends on tasks 1 and 2 Depends on tasks 1 and 2 Pathway View Marks Navigation Navigation Selection-based filters Sorting by interest Enrichment Score Individual Sort alpha-numerically Compound Sort by enrichment score Compounds Clusters compounds Nesting Local Filter : filter within a specific column Simple Nesting Recursive Nesting Global Filter : remove items that are not connected to the source column clusters pathways Task Analysis Task Analysis Task Analysis • T6 : View items in detail • T6 : View items in detail • T6 : View items in detail Pathway View Pathway View Compound View Channels Linked Views Implementation Details Size Total # of compounds that interact with pathway Total # of compounds that interact with pathway Saturation Hue : Elements Hue Compounds binding None Many One Highlighting Hover ConTour ConTour Relationship View History View Pathway View Compound View • Relationship View There are facets within the relationship view source code: https://github.com/Caleydo/ • Combination of tabular data and plots Algorithm Design Filter View Relationship View Relationship View Relationship View Relationship View Approximately 100 numerical values shown here
Relationship View Relationship View T1 T2 T5 Compound:Gene binding Conclusions Activating Inhibiting Binding T3 T5 T4 Approximately 100 numerical values shown here System ConTour Concluding Thoughts What : Data Multi-relational databases; node-link graph; clusters (derived) Why : Tasks Discovery; drill down; highlight relationships • Seems like a very good tool for use on How : Multiple Relationship view; pathway view; structured datasets compound view; history and filters Views Case Studies Backup Side-by-side linked views, containing • When there are indirect (inferred) relationships, How : Facet tabular data, bar plots, glyphs How : Selection it would be good to highlight this with some Linked highlighting across facets; automatic sorting & Highlighting uncertainty How : Filtering Drag and drop (nesting); user control (navigation) • What about incomplete relationships? How : Ranking Enrichment score; highlight; user control (navigation) & sorting Simple marks with manipulation of hue and How : Encode saturation (pathway view) Dozens of columns; upper limit on HD display appears to Scale : about 20. Thousands of data items. Up to 8 simultaneous views for compounds; only 1 for pathways
Recommend
More recommend
