IN SEARCH OF A SEQUENCE CLASSIFIER FOR A SYSTEM EMPLOYING NLP ENCODING

The evolving field of natural language processing (NLP) is increasingly intersecting with bioinformatics, particularly in DNA/RNA analysis. Previous studies highlighted the potential for classifying sequences containing selected genes. By applying NLP encoding, nucleotide strings are transformed into numerical representations, serving as inputs for predictive models. This paper analyzes the effectiveness of 25 machine learning algorithms in such classification. Impressively, 20 of them achieved a balanced accuracy (BACC) above 90%. The top performers: the MLP Classifier neural network and the SVC from the SVM family, surpassed 98% BACC and underwent hyperparameter tuning. Additionally, the performance of Automated Machine Learning (AutoML) was evaluated, allowing for the selection of optimized predictive algorithm. The preferred Hist Gradient Boosting Classifier, after hyperparameter adjustment, achieved a positive predictive value (PPV) of approximately 97.30%, outperforming the neural network and SVM representatives. The classifiers with the best BACCs exhibited negative predictive values (NPVs) higher than PPVs, reaching over 99% effectiveness. The results indicate that most models successfully handled sequence classification, with NLP encoding having a significant role. This provides flexibility in classifier selection based on complexity, score and speed. It also confirms that AutoML can adapt to data, and appropriately adjusting a model’s default values enhances classifier efficiency.