AC2

AC2 compresses protein (amino acid) sequence data using artificial neural networks and cache-hash memory models to provide lossless compression for bioinformatics analyses.

Key Features:

• Lossless Compression: AC2 performs lossless compression of protein (amino acid) sequences, preserving all original sequence information.
• Neural Network Integration: AC2 employs an artificial neural network with a stacked generalization approach that combines multiple expert models to enhance predictive accuracy and compression efficiency.
• Cache-Hash Memory Models: AC2 incorporates cache-hash memory models extending up to the highest-context orders to compactly represent sequence information and facilitate efficient retrieval.
• Reference-free and Reference-based Modes: AC2 supports both reference-free and reference-based compression modes with reported gains over its predecessor AC.
• Performance Improvements: AC2 achieves compression ratio gains of 2–9% in reference-free modes and 6–7% in reference-based modes relative to AC.
• Resource Efficiency: AC2 requires approximately three times longer processing time than AC while reducing memory usage by about sevenfold, independent of input sequence size.

Scientific Applications:

• Compression for storage and analysis: AC2 enables compact storage of large protein sequence datasets while preserving sequence integrity for downstream bioinformatics analyses.
• SARS-CoV-2 protein analysis: AC2 has been applied to measure similarity of SARS-CoV-2 protein sequences against viral proteins in the UniProt database, revealing higher similarities with pangolin coronaviruses than with bat and human coronaviruses.

Methodology:

AC2 uses an artificial neural network implementing stacked generalization that combines multiple expert models together with cache-hash memory models extending to the highest-context orders.