GenCoder: A Novel Convolutional Neural Network Based Autoencoder for Genomic Sequence Data Compression

Area of Science:

• Genomics
• Bioinformatics
• Data Compression

Background:

• Advances in DNA sequencing generate vast amounts of genomic data.
• Efficient storage and analysis of this data require effective compression algorithms.
• Deep learning, particularly autoencoders, shows promise for data compression applications.

Purpose of the Study:

• To develop a novel, reference-free compression algorithm for genomic sequences.
• To address information loss issues in autoencoder-based genomic data compression.
• To achieve high compression ratios while ensuring lossless decompression.

Main Methods:

• Proposed a new algorithm, GenCoder, utilizing a convolutional autoencoder.
• Implemented a scheme for regenerating genomic sequences from a latent code.
• Ensured lossless retrieval of original genomic data post-decompression.

Main Results:

• The GenCoder algorithm demonstrated effective generalization across various genomes and datasets.
• Achieved a significant compression gain of 27% compared to state-of-the-art methods.
• Validated the capability of lossless decompression for compressed genomic data.

Conclusions:

• GenCoder offers a powerful deep learning-based solution for genomic sequence compression.
• The proposed method overcomes limitations of autoencoders by ensuring data integrity.
• This approach contributes to efficient management of large-scale genomic datasets.