Fast Fourier transform is a training-free, ultrafast, highly efficient, and fully interpretable approach for...

Area of Science:

• Bioinformatics
• Genomic Data Analysis
• Signal Processing

Background:

• Data compression is crucial for feature selection and modeling in genomics.
• Current autoencoder methods for epigenomic data are computationally intensive and lack interpretability.

Purpose of the Study:

• To develop a training-free, efficient, and interpretable data compression method for epigenomic data.
• To overcome the limitations of autoencoder-based compression techniques.

Main Methods:

• A novel Fast Fourier Transform (FFT)-based approach was developed for data compression.
• The method was applied to compress epigenomic data, specifically histone modification data.

Main Results:

• The FFT method achieved up to 1,000-fold compression of epigenomic data.
• High reconstruction fidelity was maintained (cosine similarity of 99.7%).
• Compression was ultrafast (70 ms on GPU, 20 s on CPU) and required no training.

Conclusions:

• The FFT method provides a significant improvement over autoencoders for epigenomic data compression.
• The approach offers full interpretability of compressed features, unlike black-box autoencoders.
• This method enables high accuracy in classification models (AUC 0.960), representing a paradigm shift in data compression.