HiViewDDRAM is an interactive webware for exploration of the DNA Damage Response Assemblies Map (DDRAM).
DDRAM is a data-driven, multi-scale map describing the organisation of 605 individual DDR proteins into 109 named assemblies, and the nested relationship between these assemblies.
The DDR Association Score (DAS) network is the integration of 114 input features of physical interaction, co-dependency, co-abundance, co-expression on a genome-wide scale. Protein assemblies in DDRAM, and the assignment of individual proteins to their respective assemblies, are the result of hierarchical clustering of the DAS network.
HiViewDDRAM displays the hierarchical structure of DDRAM on the left panel in a model view using a circle-packing layout, and the underlying DAS network in the right panel in the data view using a network layout that corresponds to the circle-packing layout.
Below the data view is the control panel where certain protein interactions and properties are under the users control. The control panel also contains the SHapley Additive eXplanations (SHAP) framework as well as convenience functions (copying protein names, exporting the current network as a vector image).
Proteins can be found through the search panel. The search supports multiple proteins in the same search (space- or comma-separated); found proteins are distinguished by color).
Interactions in the DAS network can be explained using the SHAP scores.
There is an extensive help system in the form of a guided tour, tool tips and this documentation. The data view has a context-sensitive legend.
HiViewDDRAM reads and displays data structures stored in NDEx. A toolkit and documentation exist to make data structures representing multi-scale maps other than DDRAM and to upload them into NDEx for display in HiViewDDRAM: ddramutils.
A strength of HiViewDDRAM is the ability to explain any interaction in the DAS network using SHapley Additive eXplanations (SHAP) scores, grouped by class of evidence (physical, co-expression, co-abundance, co-dependency). Click any interaction in the DAS network to see an explanation of the DAS score; it will appear below the controls of the data view. The explanation comes in the form of Shapley values [Lundberg, Scott M., and Su-In Lee. "A unified approach to interpreting model predictions." Advances in neural information processing systems 30 (2017)]. These SHAP scores shown indicate the contribution of the respective input feature to the DAS score. SHAP and DAS scores are on the same scale. Only the most important SHAP scores are shown (those that are more than one standard deviation from the mean).
For features embedded with node2vec, we also show the subnetwork supporting the interaction [Grover, Aditya, and Jure Leskovec. "node2vec: Scalable feature learning for networks." Proceedings of the 22nd ACM SIGKDD international conference on Knowledge discovery and data mining. 2016.]. Note that you can expand each class to see explanations at a higher granularity.
Proteins in the current assembly can be queried for term enrichment against selected pathway databases via Enrichr [Xie 2021; PMID: 33780170]. By default, this option is turned off. To enable it, turn on the Automatically run gene set analysis with Enricher function in the Control Panel (hamburger icon left to the search panel). Enrichment results are displayed on a separate bottom panel.
HiView is implemented using React
- Checkout this repository
- Clone CyViewer module
- yarn install
- yarn link
- In the frontend directory, run
yarn link cy-viewer yarn start
© 2017 - 2022 UC, San Diego Trey Ideker Lab
HiView Application is designed and implemented by Keiichiro Ono (kono ucsd edu).