A Deep-Learning Pipeline for TSS Coverage Imputation From Shallow Cell-Free DNA Sequencing.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Cell-free DNA (cfDNA) analysis offers noninvasive insights into health and disease by reflecting nucleosome occupancy at transcription start sites (TSSs).
• Current cfDNA analysis for disease detection is limited by the high sequencing depth required, hindering widespread clinical application.
• Accurate profiling of cfDNA TSS coverage is crucial for developing effective noninvasive diagnostic tools.

Purpose of the Study:

• To develop a deep-learning pipeline, the Autoencoder of cfDNA TSS (AECT) coverage profile, for analyzing shallow cfDNA sequencing data.
• To enhance the accuracy of TSS coverage profiles from low-depth sequencing data.
• To assess the potential of AECT for noninvasive disease detection, including cancer screening.

Main Methods:

• Development of the Autoencoder of cfDNA TSS (AECT) coverage profile, a deep-learning pipeline tailored for shallow cfDNA sequencing data.
• Evaluation of AECT's performance against existing single-cell sequencing imputation algorithms regarding TSS coverage accuracy.
• Application of AECT to impute shallow cfDNA sequencing data for building cancer detection classifiers.

Main Results:

• AECT demonstrated superior performance in improving TSS coverage accuracy compared to existing imputation algorithms.
• AECT successfully captured latent biological features, such as sex and tumor status, from shallow sequencing data.
• Classifiers built using AECT-imputed data for breast and rectal cancer detection achieved performance comparable to high-depth sequencing.

Conclusions:

• The AECT pipeline enables accurate analysis of cfDNA TSS coverage from shallow sequencing, overcoming a major limitation in clinical utility.
• AECT facilitates the capture of biologically relevant information, including disease-specific signatures.
• AECT presents a promising, broadly applicable, accurate, and moderately costly approach for noninvasive cancer screening.