Prediction of quadruplex secondary structure topologies by means of sequence mutual information functions and interpretable machine learning models