Cancer classification based on chromatin accessibility profiles with deep adversarial learning model.

Area of Science:

• Genomics
• Cancer Biology
• Bioinformatics

Background:

• Cancer genomics presents complex, diverse profiles, challenging distinct clustering across tumor types.
• Existing methods struggle with high-dimensional, whole-genome omics data like ATAC-seq profiles.

Purpose of the Study:

• To develop an end-to-end deep adversarial learning approach for cancer genomics data clustering.
• To leverage high-dimensional features from ATAC-seq and RNA-seq data for improved classification.

Main Methods:

• Implemented ClusterATAC, an end-to-end deep adversarial learning framework.
• Applied ClusterATAC to ATAC-seq and RNA-seq datasets for performance evaluation.
• Analyzed clustering solutions on a pan-cancer ATAC dataset to identify non-coding region associations.

Main Results:

• ClusterATAC demonstrated excellent performance on ATAC-seq and RNA-seq datasets.
• Over 70% of clusters from the pan-cancer ATAC dataset were single-tumor-type-dominant.
• Remaining clusters predominantly grouped similar tumor types, highlighting non-coding loci's role.

Conclusions:

• Deep adversarial learning effectively clusters cancer genomics data, particularly ATAC-seq profiles.
• Identified non-coding loci and linked genes significantly impact cancer progression.
• Findings suggest potential advancements in cancer diagnosis and therapeutic strategies.