Protein Hotspot Prediction Using S-Transform

Since experimental techniques of protein hotspot prediction are still financially extremely demanding and time consuming there is a strain to produce sufficiently reliable computational techniques for this particular task. We propose an algorithm based on Resonant Recognition Model relying heavily on signal processing techniques. Processed numerical signal is obtain solely form protein sequence using physical quantity EIIP. We therefore use no information of protein structure. The key element here is a time-frequency analysis tool – S-transform. This allows us to determine exact residues responsible for majority of performance on protein’s characteristic frequency. We achieve basic sensitivity of 85 % and PPV 49 %, while demanding very little computing resources, because simplicity is one of the biggest advantages of our approach.