rcCAE: a convolutional autoencoder method for detecting intra-tumor heterogeneity and single-cell copy number...

Area of Science:

• Genomics
• Computational Biology
• Cancer Research

Background:

• Intra-tumor heterogeneity (ITH) is a key challenge in cancer, leading to relapse and treatment resistance.
• Single-cell DNA sequencing (scDNA-seq) offers high resolution for studying ITH by profiling copy number alterations (CNAs) at the single-cell level.

Purpose of the Study:

• To introduce rcCAE, an integrated framework for accurate inference of cell subpopulations and single-cell CNAs from scDNA-seq data.
• To enhance the understanding of ITH for improved personalized cancer therapies.

Main Methods:

• Utilized a convolutional autoencoder (CAE) within rcCAE to learn latent cell representations and extract CNA information from noisy scDNA-seq read counts.
• Employed unsupervised representation learning for accurate cell clustering in a low-dimensional latent space.
• Developed methods for detecting single-cell CNAs from enhanced read count data.

Main Results:

• rcCAE demonstrated superior performance compared to existing CNA calling methods on simulated datasets.
• The framework proved highly effective in inferring cancer clonal architecture.
• Analysis of real datasets revealed a more refined clonal structure, capturing details missed by integer copy number-based methods.

Conclusions:

• rcCAE provides a robust and accurate approach for deciphering ITH from scDNA-seq data.
• The method offers a more granular view of clonal evolution, crucial for advancing personalized cancer treatment strategies.