KIMI: Knockoff Inference for Motif Identification from molecular sequences with controlled false discovery rate

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Metagenomic sequencing generates vast amounts of data, requiring efficient methods for classifying microbial sequences.
• Current k-mer frequency methods often use all k-mers, potentially including irrelevant ones, limiting prediction accuracy.
• Identifying biologically significant nucleotide patterns (k-mers) is crucial for distinguishing microbial groups.

Purpose of the Study:

• To develop a general framework, KIMI, for selecting relevant k-mers with guaranteed false discovery rate (FDR) control.
• To enable motif discovery for distinguishing microbial sequence groups at a user-defined FDR level.
• To improve the accuracy of metagenomic sequence classification by utilizing selected, biologically significant k-mers.

Main Methods:

• Developed KIMI, a framework utilizing the model-X Knockoffs statistical method for FDR control.
• Applied KIMI to select significant k-mers for sequence motif discovery.
• Evaluated KIMI's performance through simulation studies and real-world viral and bacterial contig datasets.

Main Results:

• KIMI effectively controls FDR while maintaining high statistical power in simulations.
• KIMI outperforms established methods like Benjamini-Hochberg and q-value in FDR control.
• Using KIMI-selected k-mers for prediction models significantly increases the accuracy of classifying viral and bacterial contigs.

Conclusions:

• KIMI provides a theoretically guaranteed and effective approach for selecting biologically relevant k-mers.
• The framework enhances the accuracy of microbial sequence classification by focusing on significant nucleotide patterns.
• KIMI offers a reproducible and powerful tool for metagenomic data analysis and motif discovery.