Classification of Short-Interfering RNA Through Transformer Encoder Model

This paper presents the use of natural language processing (NLP) to classify short-interfering RNA (siRNA) sequences based on their knockdown efficiency. The proposed method utilizes a transformer encoder model. The siRNA sequences are treated as text input, tokenized, and embedded through positional encoding to allow the transformer encoder model to extract significant features of the siRNA nucleotides and classify them based on their silencing efficiency. The transformer encoder model achieved a high level of accuracy and performance on the test sequences with an overall accuracy of 98%. Despite the imbalanced number of samples of each class, the proposed method for classifying the knockdown efficiency is proven effective. This claim is supported by examining the results of precision, recall, and f1-score of each class with class 1 (knockdown efficiency from 50-100% having scores of precision equals 0.98, recall is 0.99, and f1-score of 0.99. Class 0 with precision, recall, and f1-score of 0.90, 0.76, and 0.82. The ability of the model to process large datasets and capture dependencies, as well as parallel data processing and positional embedding contributed to its high accuracy in predicting complex siRNA sequences.