A data compression system, method, terminal and medium based on a dynamic mask self-encoding architecture

The data compression system, based on a dynamic mask autoencoder architecture, solves the problems of low efficiency and insufficient accuracy in genomic data compression, achieving efficient and cross-platform compatible genomic data processing, and supporting precision medicine and global biomedical collaboration.

CN122245427APending Publication Date: 2026-06-19PENG CHENG LAB
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN ยท China
Patent Type
Applications(China)
Current Assignee / Owner
PENG CHENG LAB
Filing Date
2026-04-08
Publication Date
2026-06-19

AI Technical Summary

Technical Problem

Existing genome data compression technologies cannot effectively handle the k-mer distribution and local repetitive patterns of DNA sequences, ignore the dynamic range of quality values, resulting in low compression efficiency and accuracy loss in complex scenarios, high query latency in cross-platform scenarios, and failure to meet the needs of different sequencing devices.

Method used

A data compression system based on a dynamic mask autoencoder architecture is adopted, including a genome-aware dynamic masking module, a multi-scale Transformer encoding module, a dual-path residual decoding module, a context-aware probability calibration module, and an intelligent asynchronous group encoding and decoding module. Through a three-level dynamic masking strategy, a hierarchical attention mechanism, and asynchronous pipeline technology, efficient compression and decoding are achieved.

Benefits of technology

It achieves efficient genomic data compression, supports multi-platform data compatibility, preserves key biological information, reduces storage and transmission costs, improves data processing speed and accuracy, and supports precision medicine and global biomedical collaboration.

โœฆ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122245427A_ABST
    Figure CN122245427A_ABST
Patent Text Reader

Abstract

This invention discloses a data compression system, method, terminal, and storage medium based on a dynamic mask autoencoder architecture. The system includes: a genome-aware dynamic masking module, a multi-scale Transformer encoding module, a dual-path residual decoding module, a context-aware probability calibration module, and an intelligent asynchronous block encoding / decoding module. The genome-aware dynamic masking module executes a three-level dynamic masking strategy on the raw genome sequence to be compressed. The multi-scale Transformer encoding module extracts global features. The dual-path residual decoding module outputs the initial base probability distribution. The context-aware probability calibration module outputs stable calibration probabilities. The intelligent asynchronous block encoding / decoding module executes an intelligent block encoding strategy and an asynchronous pipeline, combined with stable calibration probabilities, to perform arithmetic encoding, obtaining a compressed whole-genome bitstream. This invention enables deep mining and efficient compression of whole-genome semantic features.
Need to check novelty before this filing date? Find Prior Art