Dynamic Tumor Analysis Using Deep Embedded Clustering (DEC) for Personalized Oncology Treatment

Maragoni Mahendar, Pinnapureddy Manasa, Satyanarayana Nimmala, K Raghavendar, Anjum Sheikh, Medikonda Asha Kiran

2025 · 16 citations · 11 references

Concepts

Abstract

Clustering tumors is crucial for comprehending tumor heterogeneity and for personalized treatment approaches. Conventional clustering techniques frequently inadequately represent the intricacies of high-dimensional tumor data. This study utilizes Deep Embedded Clustering (DEC), a sophisticated unsupervised learning method, to categorize tumors according to molecular and clinical characteristics. The collection, obtained from the Cancer Genome Atlas (TCGA) through the Genomic Data Commons (GDC) Portal, comprises 12,000 tumor samples spanning 15 tumor types, including breast, lung, colorectal, prostate, and ovarian cancers, along with attributes such as gene expression and clinical metadata. DEC uses a Convolutional Neural Network (CNN) to map high-dimensional data into a lower-dimensional latent space, effectively capturing essential tumor patterns. The samples are subsequently grouped in this space utilizing a clustering objective function that delineates tumor subtypes while preserving the integrity of the data structure. The model is progressively optimized to enhance the embedding and clustering assignments, reducing clustering loss and augmenting classification accuracy. The experimental findings indicate that DEC surpasses conventional approaches, with a clustering accuracy of 92.8%, in contrast to 78.5% for k-means and 81.3% for DBSCAN. The application of t-SNE and UMAP techniques substantiates the biological significance of the detected tumor clusters. Our results underscore the promise of DEC for tumor categorization, providing a means for personalized cancer therapy and predictive oncology.

References

11