TL;DR
OncotreeVIS is a new interactive web-based tool that allows researchers to visualize and compare complex mutation trees from cancer cohorts. It addresses the critical need for better tools to analyze tumor evolution at a population level, moving beyond single-tumor analysis to understand patterns across many patients.
What Problem Does This Paper Address?
Cancer is a complex, evolving disease driven by accumulating genetic mutations. Recent advances in sequencing technology allow us to reconstruct the evolutionary history of tumors as "mutation trees," where nodes represent genetic events and branches show their temporal order. While many tools can infer these trees, analyzing them collectively across many patients (a "cohort") has been challenging. This is crucial for understanding common evolutionary patterns, identifying drivers of drug resistance, and ultimately informing clinical decisions. Existing visualization tools are often limited to individual trees, lack cohort-level comparison capabilities, and require local installation, hindering collaboration and accessibility.
Methods
OncotreeVIS is built as an interactive graphical user interface (GUI) using HTML5 and JavaScript, leveraging the D3.js library for data visualization. Key methodological contributions include:
- Cohort Visualization: The tool can display multiple mutation trees side-by-side, optionally grouped by a provided clustering. It highlights matching subclones (nodes with similar genetic profiles) and conserved evolutionary trajectories (shared branches) across trees.
- Interactive Exploration: Users can zoom, pan, and explore individual trees. Clicking on nodes reveals detailed information about gene events (point mutations, copy number alterations), subclone sizes, and even drug-gene interactions.
- Pairwise Tree Comparison: OncotreeVIS can visualize pairwise tree distances using a heatmap and project trees into a 2D latent space using Multidimensional Scaling (MDS), allowing for visual identification of clusters and relationships between tumors.
- Drug-Gene Interaction Integration: The tool integrates with the DGIdb database to highlight how specific mutations in a tumor's evolutionary history might relate to drug targets and potential therapeutic resistance.
- Efficient Rendering: To handle large cohorts, OncotreeVIS employs asynchronous computations and lazy loading of offscreen elements, preventing browser slowdowns.
- JSON Input Format: A standardized JSON format is used for inputting mutation tree data, making it easier to integrate with various mutation tree inference tools. Scripts are provided to convert outputs from common tools into this format.
Datasets
The paper showcases OncotreeVIS using seven publicly available mutation tree cohorts. These include datasets from studies on:
- Acute Myeloid Leukemia (AML) (Morita et al. 2020, Sollier et al. 2023)
- Non-small cell lung cancer (TRACERX)
- Breast cancer (Razavi et al. 2018)
- Various cancer types (Noble et al. 2022)
The tool is designed to accept new mutation tree data in a predefined JSON format, allowing users to visualize their own cohorts.
Key Insights and Ideas
- Bridging the Gap: The primary insight is the development of a tool that bridges the gap between complex computational inference of tumor evolution and the practical needs of visualization and comparative analysis at the cohort level.
- Visualizing Heterogeneity: OncotreeVIS makes it easier to visually grasp the heterogeneity of cancer evolution across a population of tumors, revealing common patterns and unique evolutionary trajectories.
- Connecting Evolution to Treatment: By integrating drug-gene interaction data, the tool provides a powerful way to explore how evolutionary events might influence therapeutic response or resistance, opening avenues for personalized medicine.
- Accessibility: Offering both a JavaScript library for local use and a web application significantly lowers the barrier to entry for researchers and clinicians.
Key Results
The paper demonstrates the utility of OncotreeVIS through several use cases, including:
- Visualizing longitudinal AML samples to track tumor evolution during treatment and identify mechanisms of resistance.
- Comparing mutation trees from different cancer types to understand diverse modes of spatial tumor evolution (linear, branching, punctuated).
- Exploring clusters of AML mutation trees governed by specific primary mutations.
- Displaying alternative phylogenies inferred by a mutation tree inference tool, highlighting conserved trajectories.
While specific quantitative performance metrics (e.g., speedup, accuracy) are not the primary focus, the tool's ability to handle and visualize large cohorts interactively is its main result.
Limitations and Future Work
- Limitations:
- The paper mentions that the output of different mutation tree inference tools doesn't always agree, and OncotreeVIS visualizes these potentially differing reconstructions. The tool itself doesn't resolve these discrepancies but helps in their visual comparison.
- The visualization of conserved edges is not computed when sorting options are selected instead of clustering.
- Future Work:
- The authors suggest that OncotreeVIS can be integrated into more complex pipelines using its JavaScript library.
- The tool's ability to visualize drug-gene interactions and identify potential treatment strategies points towards future applications in clinical decision support.
- Further development could involve more sophisticated methods for comparing and clustering trees, potentially incorporating new tree distance metrics.
Summary updated