Abstract 1306: Density-based mutation clustering in 3D space

Abstract

Abstract Cancer driver mutation and gene discovery has been a major challenge in cancer research. Several computational tools focus on clustering mutations on 3D protein structures to identify cancer drivers, including HotSpot3D, CLUMPS, HotMAPS, 3DHotSpots.org and e-Driver. While other tools provide a single snapshot of mutation clusters on the 3D structure, HotSpot3D utilizes a unique density-based clustering module capable of providing a full dynamical profile of clusters with varying densities. The density module (DM) is capable of detecting subtle variation in the density of mutations in 3D structures with little computation time and complexity. DM gives users the ability to detect the bridging effect, where more dense subclusters form a less dense supercluster due to a small number of mutations in between the subclusters connecting them. In addition to clustering based on physical density of mutations, DM is capable of clustering by additional biologic properties, including mutation recurrence and pathogenicity. Moreover, the underlying OPTICS algorithm utilized in the DM is inherently a “fuzzy” clustering algorithm, allowing slight variation of clusters based on the start point of the clustering algorithm from one run to another. Therefore, the DM provides a cluster membership probability measure for each mutation, giving the user the ability to choose the preferred stringency. Additionally, DM allows users to visualize the density clusters using Pymol. Since the whole dynamical set of clusters is produced by a single run, the visualization is performed so that one can reveal other clusters with higher densities by zooming in and out. Using a curated set of experimentally validated oncogenic mutations, we evaluated the performance of DM in detecting functionally activating mutations; our tool performed well in identifying driver mutations at high density thresholds. Moreover, the DM was able to identify mutations clustering at high density in BRAF, PIK3CA and KEAP1, which were not detected by other 3D clustering tools and sequence-based tools (SIFT, PolyPhen2, CHASM, etc.). The DM was also better at detecting clusters along the interface of protein-protein complexes (e.g., BRAF-KEAP1) compared to other 3D tools. In summary, HotSpot3D-DM allows for dynamical clustering, improved visualization, and identifies novel driver mutations missed by previous tools. Citation Format: Amila Weerasinghe, Sohini Sengupta, Adam D. Scott, Maththew H. Bailey, Michael C. Wendl, Ken Chen, Gordon Mills, Li Ding. Density-based mutation clustering in 3D space [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2018; 2018 Apr 14-18; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2018;78(13 Suppl):Abstract nr 1306.