Identification of intelligence-related proteins through a robust two-layer predictor

Abstract

In this study, we advance a robust methodology for identifying specific intelligence-related proteins across phyla. Our approach exploits a support vector machine-based classifier capable of predicting intelligence-related proteins based on a pool of meaningful protein features. For the sake of illustration of our proposed general method, we develop a novel computational two-layer predictor, Intell_Pred, to predict query sequences (proteins or transcripts) as intelligence-related or non-intelligence-related proteins or transcripts, subsequently classifying the former sequences into learning and memory-related classes. Based on a five-fold cross-validation and independent blind test, Intell_Pred obtained an average accuracy of 87.48 and 88.89, respectively. Our findings revealed that a score >0.75 (during prediction by Intell_Pred) is a well-grounded choice for predicting intelligence-related candidate proteins in most organisms across biological kingdoms. In particular, we assessed seismonastic movements and associate learning in plants and evaluated the proteins involved using Intell_Pred. Proteins related to seismonastic movement and associate learning showed high percentages of similarities with intelligence-related proteins. Our findings lead us to believe that Intell_Pred can help identify the intelligence-related proteins and their classes using a given protein/transcript sequence.

Keywords: Intell_pred; learning; protein features; support vector machine.