Predicting Human Acceptor Splice Site Regions using Text Distance Methods

As genomic non-coding regions have gained significant research interest recently, the necessity to understand intronic sequences has become more stringent. This paper presents a study in which three computational text distance methods (Levenshtein, Hamming, Needleman-Wunsch) were used for predicting human acceptor splice site sequences. The Homo Sapiens Splice Site Dataset was used for evaluating the predicting performance of the methods. A python library was used for the implementation of the methods. Sensitivity, specificity, and the mean between the sensitivity and specificity were considered as performance metrics. For each method, several configurations were scanned. The first parameter in the configuration was the number of neighboring sequences. For this parameter, the odd values between one and eleven were considered. The second parameter represented the length of the DNA sequences (between nine and thirty-five). The overall predicting performance of Hamming and Levenshtein was lower than that of Needleman-Wunsch. The optimal configuration for Needleman-Wunsch was within nine neighboring sequences and twenty nucleotides, in which the method obtained an accuracy of 85.6%. The results can be further used for identifying cryptic acceptor splice sites in the deep intronic area of genes that occur due to genetic variants.