ANFIS ve SBERT Yaklaşımlarının Hibrit Kullanımı ile DNA Dizilimleri Üzerinde Ekson ve İntron Bölgelerinin Sınıflandırılması

Abstract

DNA is the part of the genome that contains enormous amounts of information related to life. Amino acids are formed by coding three nucleotides in this genome part, and the encoded amino acids are called codes in DNA. The frequency of the triple nucleotide in the DNA sequence allows for the evaluation of protein-coding (exon) and non-protein-coding (intron) regions. Distinguishing these regions enables the analysis of vital functions related to life. This study provides the classification of exon and intron regions for BCR-ABL and MEFV genes obtained from NCBI and Ensemble datasets, respectively. Then, existing DNA sequences are clustered using pretrained models in the scope of the SBERT approach. In the clustering process, K-Means and Agglomerative Clustering approaches are used consecutively. The frequency of repetition of codes is calculated with a representative sample selected from each cluster. The matrix is created using the frequencies of 64 different codons that constitute genetic code. This matrix is given as input to the ANFIS structure. The %88.88 accuracy rate is obtained with the ANFIS approach to classify exon and intron DNA sequences. As a result of this study, a successful result was produced independently of DNA length.

Human exons and introns classification using pre-trained Resnet-50 and GoogleNet models and 13-layers CNN model

“…They reached an accuracy rate value of 93.5% for their proposed S-Resnet model. The results are also comparable with results in 29 where they preprocessed their exon and intron dataset using clustering technique and they proposed ANFIS model (a combination between an artificial neural network and a fuzzy inference system) for the classification process. They obtained a value of 88.88% for the accuracy rate on the test phase.…”

“… Spec. Resnet-50 92% 94.2% 90.4% GoogleNet 91.5% 95.9% 90.2% 13-layers CNN 91.6% 92.4% 94.2% S-Resnet 28 93.5% 92.9% 93.8% ANFIS 29 88.88% 75% 100% …”

Detection of Exon and Intron Regions in Dna Sequences by the Proposed Hashing Function

“…In this direction, it has been stated that inferences based on nucleotide length are produced in the [27] study, which uses the signal processing approach that has been prominent, especially since the 2000s. On the other hand, a study independent of nucleotide length was done in the [28] study, in which classification was provided with the structural and statistical features extracted from the sequences. The obtained results were more successful than the outputs produced using the signal-processing approach.…”

“…The integer mapping technique is a 1-dimensional mapping technique [28]. In order to digitize the DNA structure with this technique, firstly, the number of organic bases in the sequences is examined and then the assignment is done.…”

“…Especially the developments in the field of digital signal processing [3,4] have emerged as a preferred strong field for exploring the relationships, patterns and periodicity states in the data [10]. However, the [27] and [28] studies using the digital signal processing approach stated that digital signal processing approaches not only serve a successful decision process but also provide an inference based on DNA length. Therefore, it was planned to construct a structure independent of DNA length.…”