Enhanced Multi-Criteria DNA Sequence Analysis Using  Neutrosophic Logic and Deep Learning: An Integrated  Approach for Comparison and Classification

Romany M. Farag; Mahmoud Y. Shams; Dalia A. Aldawody; Hazem M. El-Bakry; Huda E. Khalid; Ahmed K. Essa; A. A. Salama

Authors

Romany M. Farag Dept. of Math and Computer Science, Faculty of Science, Port Said Univ., Egypt;
Mahmoud Y. Shams Dept. of Machine Learning, Faculty of Artificial Intelligence, Kafrelsheikh University, Egypt;
Dalia A. Aldawody Dept. of Math and Computer Science, Faculty of Science, Port Said Univ., Egypt;
Hazem M. El-Bakry Dept. of Information Systems, Faculty of Computer and Information Sciences Mansoura University Egypt;
Huda E. Khalid University of Telafer, The Administration Assistant for the President of the Telafer University, Telafer, Iraq;
Ahmed K. Essa University of Telafer, The Administration Assistant for the President of the Telafer University, Telafer, Iraq;
A. A. Salama Dept. of Math and Computer Science, Faculty of Science, Port Said University, Egypt;

Keywords:

DNA Sequence Analysis; Neutrosophic Logic; Deep Learning; Enhanced Multi-Criteria

Abstract

This paper presents a comprehensive approach for analyzing DNA sequences and attempts to find a practical
solution to the problems of comparing and classifying genomic data, which have become increasingly complex
and noisy. The approach we propose combines neutrosophic logic which primarily deals with data confusion
and uncertainty with modern machine learning techniques to achieve more realistic and flexible analysis.
Traditionally, traditional methods such as multiple sequence alignment (MSA) algorithms or the Needleman
Winch algorithm become extremely cumbersome when faced with large or unclear data. Performance degrades,
and results are sometimes inaccurate. Therefore, we propose a system we call the "Advanced DNA
Comparator," which converts DNA sequences to digital representations using Word2Vec (similar to how
humans understand words), uses neutrosophic logic as a tool for understanding ambiguity, and then applies
classification to a random forest algorithm. The results we achieved were not only promising, but also truly
powerful: 99.83% accuracy, 100% accurate classification, and a ROC-AUC of 99.92%. This simply means that
the system was able to handle sequences and detect similarities even when the world was a little luttered.This
framework could be a real turning point, because it is not only accurate but also easy to interpret and expand,
making it a powerful tool for applications such as personalized medicine or early disease prediction.

DOI: 10.5281/zenodo.15806690