DeepDNA

DeepDNA compresses human mitochondrial genome sequences using a hybrid convolutional neural network (CNN) and long short-term memory (LSTM) architecture to capture local and long-range sequence dependencies for data reduction.

Key Features:

• Hybrid Architecture: Combines CNNs and LSTMs so convolutional layers detect local sequence patterns while LSTM layers model longer-range dependencies.
• Non-Reference Based Compression: Performs compression without relying on external reference genomes and achieves effectiveness comparable to reference-based approaches.
• Versatility Across Data Types: Effective for population genomes with high redundancy as well as single genomes with lower redundancy, enabling use across diverse mitochondrial datasets.

Scientific Applications:

• Mitochondrial genome compression: Reduces the size of human mitochondrial genome datasets produced by high-throughput sequencing technologies to facilitate storage and transmission.

Methodology:

The hybrid CNN-LSTM model is trained on human mitochondrial genome sequences, with CNN components learning local sequence features and LSTM components capturing higher-level, longer-range dependencies to produce compressed sequence representations.