Neural Upsampling of Decompressed Genomic Data After Lossy Compression

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing data compression techniques, particularly lossy compression methods like HEVC, result in data loss that is challenging to recover, especially in biological data where information retention is critical for accurate analysis.

Innovation Solution

A system and method using a neural network with a novel AI deblocking network, incorporating recurrent layers for feature extraction and a channel-wise transformer with attention, to upsampling decompressed biological data after lossy compression, effectively recovering lost information.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of substance

If lossy compression is applied to genomic data, then compression ratio is improved, but information loss increases

Engineering Contradiction:

Improvedata sizeVSAvoidgenomic information

Core Design Contradiction:

Loss of substanceVSLoss of information

Solution Approach 1:

A neural network is introduced as an intermediary component between the lossy compression and decompression processes. The neural network learns to predict and recover lost genomic information by analyzing patterns in the compressed data, effectively mediating the information loss caused by aggressive compression techniques.

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The system changes the parameters of the decompression process by using learned models instead of traditional deterministic decompression. The neural network adjusts its internal parameters (weights and biases) during training to optimize the recovery of genomic information from compressed representations, enabling flexible adaptation to different compression levels.

Inventive Principle:

Principle #35Parameter changes

2Speed

If traditional decompression is used, then processing speed is improved, but data quality deteriorates

Engineering Contradiction:

Improvedecompression speedVSAvoiddata reconstruction quality

Core Design Contradiction:

SpeedVSManufacturing precision

Solution Approach 1:

The neural network performs preliminary learning during a training phase before actual decompression operations. By pre-learning the relationships between compressed and original genomic data through extensive training on representative datasets, the network is prepared to quickly and accurately reconstruct data during operational decompression without requiring complex real-time computations.

Inventive Principle:

Principle #10Preliminary action

3Manufacturing precision

If compression artifacts are reduced, then data quality is improved, but computational complexity increases

Engineering Contradiction:

Improvedecompressed data qualityVSAvoidneural network complexity

Core Design Contradiction:

Manufacturing precisionVSDevice complexity

Solution Approach 1:

The neural network architecture applies different processing strategies to different parts of the genomic data. By focusing computational resources on recovering locally important genomic features and patterns while using simpler interpolation for less critical regions, the system achieves high overall data quality without requiring uniformly complex processing across all data points.

Inventive Principle:

Principle #3Local quality

Data Source

PatentUS20250191697A1System and methods for upsampling of decompressed genomic data after lossy compression using a neural network

Publication Date: 2025.06.12 ATOMBEAM TECH INC

AI summary

A system and methods for upsampling of decompressed biological data after lossy compression using a neural network integrates AI-based techniques to enhance compression quality. It incorporates a novel deep-learning neural network that upsamples decompressed data to restore information lost during lossy compression, taking advantage of cross-correlations between biological data sets.