Deep Learning-Based Classification of CRISPR Loci Using Repeat Sequences.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• CRISPR-Cas systems are crucial for gene editing and are increasingly found in metagenomic data.
• Traditional classification relies on identifying cas genes, which is challenging in metagenomes or fragmented assemblies.
• Accurate classification of CRISPR-Cas systems is vital for understanding microbial communities and genome evolution.

Purpose of the Study:

• To develop a deep learning method for classifying CRISPR-Cas systems using only repeat sequences.
• To address the limitations of cas gene-dependent classification in metagenomic datasets.
• To provide a complementary tool for identifying orphan or distant CRISPR loci.

Main Methods:

• Developed CRISPRclassify-CNN-Att, a deep learning model utilizing convolutional neural networks (CNNs) and self-attention mechanisms.
• Employed a stacking strategy to handle imbalanced sample sizes across subtypes.
• Applied transfer learning to enhance classification accuracy for underrepresented subtypes.

Main Results:

• CRISPRclassify-CNN-Att achieved outstanding performance in classifying multiple CRISPR-Cas subtypes, especially those with larger sample sizes.
• The method successfully classified CRISPR loci based solely on repeat sequences, independent of cas gene presence.
• Demonstrated effectiveness as a complementary approach to traditional cas-based classification methods.

Conclusions:

• CRISPRclassify-CNN-Att offers a novel and effective deep learning-based approach for CRISPR-Cas system classification.
• This method enhances the analysis of metagenomic data by enabling classification of CRISPR loci lacking cas gene information.
• The tool serves as a significant advancement for CRISPR-Cas research, complementing existing methodologies.