Minimum Functional Length Analysis of K-Mer Based on BPNN.

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Multilayer feed-forward networks, such as BP neural networks (BPNN), offer deep data cognition and high accuracy.
• Previous research has not explored BPNN for k-mer analysis in DNA sequences.

Purpose of the Study:

• To investigate the efficacy of BPNN in classifying DNA k-mers.
• To differentiate k-mers based on their composition (X+Y content vs. random).

Main Methods:

• BPNN was employed to train and test binary classification models for k-mers.
• K-mers were categorized into two groups: X+Y content and completely random.
• Classification accuracy was evaluated for varying k-mer lengths (k≤6 and k≥7).

Main Results:

• 100% classification accuracy was achieved for k-mers with X+Y content, irrespective of k-mer length (k≤6 or k≥7).
• For completely random k-mers, 100% accuracy was observed for k≤6.
• Accuracy decreased for random k-mers with k≥7, approaching 50% as k increased, indicating a loss of predictive power.

Conclusions:

• K-mers of length k≥7 appear to function as basic functional fragments in DNA sequences.
• K-mers of length k≤6 represent fundamental component fragments without inherent basic nucleic acid function.