iDeLUCS: a deep learning interactive tool for alignment-free clustering of DNA sequences

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Unsupervised clustering of DNA sequences is crucial for genomic analysis.
• Existing methods often require sequence alignment or taxonomic information, limiting their applicability.
• Scalable and accurate tools are needed for large, diverse DNA sequence datasets.

Purpose of the Study:

• To introduce iDeLUCS, an interactive deep learning-based software tool for unsupervised DNA sequence clustering.
• To enable the detection of genomic signatures for clustering without sequence alignment or taxonomic identifiers.
• To provide a user-friendly and scalable solution for DNA sequence analysis.

Main Methods:

• Developed an interactive deep learning software tool, iDeLUCS.
• Implemented a graphical user interface with hardware acceleration support.
• Evaluated performance on diverse real, viral, simulated metagenomic, and synthetic DNA sequence datasets.
• Compared iDeLUCS against k-means++, GMM, MeShClust v3.0, and DeLUCS using intrinsic and external evaluation metrics.

Main Results:

• iDeLUCS achieved superior unsupervised clustering accuracy compared to classical and specialized algorithms.
• Outperformed classical algorithms by an average of ~20% and specialized algorithms by ~12% on real DNA sequence datasets.
• Demonstrated robustness across various genomic datasets, including microbial, viral, and synthetic sequences.

Conclusions:

• iDeLUCS is a robust and accurate method for unsupervised clustering of large and diverse unlabeled DNA sequences.
• The tool's user-friendly interface and deep learning approach facilitate genomic signature detection and clustering.
• iDeLUCS offers a significant advancement over existing methods for DNA sequence analysis.