Utilizing Machine Learning for Gene Expression Data: Incorporating Gene Sequencing, K-Mer Counting and Asymmetric N-Grams Features

Comprehending the sequence of deoxyribonucleic acid sequence is a pivotal aspect of bio-informatics research. With the substantial surge in biological data, there arises a necessity of effective methodologies to address the crucial challenge within the overarching computation framework of deoxyribonucleic acid sequence classification. Numerous machine learning’s can be used to complete these tasks compared to a manual technique that is followed for ages. The aim of this project is to perform effective approaches for pre-processing deoxyribonucleic acid sequences (K562 ChIP-seq) and use machine learning’s to train the model by using deoxyribonucleic acid sequences to create judgments, predictions and classifications deoxyribonucleic acid sequences into known categories. As the machine learning may analyze large and complex datasets to identify patterns and trends of deoxyribonucleic acid sequences that may not be apparent to humans. In this study, the pre-processing methods are k-mers and N-grams. The machine learning classifiers, we employ Naïve Bayes classifier, K-nearest neighbors classifier and Random Forest classifier to train model and to evaluate the accuracy of predicting the K562 ChIP-seq. All the classifiers achieve overall accuracy over 80%. We show that Naïve Bayes classifier based on k-mers encoding which gives good result for both symmetric and asymmetric N-grams. The Naïve Bayes classifier based on 3-mers until 6-mers encoding have the highest accuracy of all the classifiers tested at 89.8% for symmetric N-grams and 90% for asymmetric N-grams. The Random Forest based on 3-mers and (2,3)-grams achieves the second highest accuracy in predicting the K562 ChIP-seq with 87.1%. The highest accuracy for K-nearest neighbors classifier based on 2-mers until 6-mers encoding for asymmetric N-grams tested at 86.9%. The results also reveal that the performance of the novelty classifiers’ finding based on asymmetric N-grams are better if compared with symmetric N-grams. Naïve Bayes classifiers based on k-mers encoding and N-grams features shows significant performance in term of accuracy, precision, recall and F1-score.