ANN modeling of DNA sequences: new strategies using DNA shape code

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning in Biology

Background:

• Accurate DNA sequence representation is crucial for computational modeling of biological systems.
• Existing encoding strategies for DNA sequences in artificial neural networks (ANNs) have limitations, especially with sparse datasets.

Purpose of the Study:

• To introduce and evaluate two novel DNA sequence encoding strategies: wedge and twist codes.
• To assess the performance of these new codes in ANN-based modeling and classification tasks.
• To compare the efficacy of wedge and twist codes against existing methods.

Main Methods:

• Development of two new encoding strategies, wedge and twist codes, based on DNA helical parameters.
• Application of these codes in ANN models for biological system analysis.
• Rigorous comparative evaluation through three distinct case studies involving mapping and classification.

Main Results:

• The new wedge and twist coding strategies demonstrate superior performance compared to existing methods.
• This outperformance is particularly evident in scenarios with limited data availability for ANN model construction.
• Successful application in both modeling and classification tasks within the case studies.

Conclusions:

• Wedge and twist codes offer a significant advancement in representing DNA sequences for ANNs.
• These novel strategies enhance the accuracy and robustness of biological models, especially under data constraints.
• The findings suggest broader applicability of these codes in computational biology and bioinformatics.