Aggregation prediction; A bioinformatics method for studying the effects of missense variations on pathogenicity

Background and Aims: Aggregation has been shown to be an intrinsic property of many proteins including proteins not involved in amyloid diseases. The most common types of protein aggregates are amyloid fibrils and amorphous aggregates characterized by an increase in the level of β-structure. Missense variations have the potential to change the propensity of a property to aggregate. Variation research in recent times has focused on obtaining information about the effects of sequence variations on proteins. Experimental study of the possible disease association of variants is laborious and time-consuming. Computational methods on the other hand give rapid automated results for large amounts of data sets but are less reliable. To use aggregation as mechanism to study the effects of missense variations on pathogenicity, it is important to predict the change in aggregation of proteins upon aggregation. There are several aggregation prediction methods available on the Internet making it difficult to find the best methods. This study evaluates the performance of five widely used aggregation prediction methods. Results from the aggregation prediction can then be used for pathogenicity prediction to determine how they correlate. Methods: Aggrescan, AmylPred consensus, Average Packing Density, TANGO and Hexapeptide Conformational Energy were the evaluated methods. The methods were tested with a dataset of 365 missense variations. Matthews correlation Coefficient, Sensitivity, Specificity, Accuracy, Precision and Negative Predictive Value were the measures used to evaluate the performance of the prediction methods. Results: Aggrescan performed best in MCC, accuracy, sensitivity and NPV show that is the best method. Tango performed best in precision (0.92) and specificity (0.95). Conclusion: From the results, all the methods showed good MCC values of above 0.59. It is easy to conclude that Aggrescan was the best amongst all the five methods followed by Tango. It is on the other hand difficult to recommend a specific method since all the methods depend on physicochemical properties and side chains in β-sheet aggregates making the algorithms in the methods give different results. It is therefore advisable for the end user to know much about the algorithms used before choosing a particular method for prediction. Asiasanat:Aggregation, Missense variations,Pathogenicity, Amyloid fibril