Predicting human immunodeficiency virus protease cleavage sites in proteins by a discriminant function method

Based on the sequence‐coupled (Markov chain) model and vector‐projection principle, a discriminant function method is proposed to predict sites in protein substrates that should be susceptible to cleavage by the HIV‐1 protease. The discriminant function is defined by Δ = ϕ+ – ϕ−, where ϕ+ and ϕ− are the cleavable and noncleavable attributes for a given peptide, and they can be derived from two complementary sets of peptides, S+ and S−, known to be cleavable and noncleavable, respectively, by the enzyme. The rate of correct prediction by the method for the 62 cleavable peptides and 239 noncleavable peptides in the training set are 100 and 96.7%, respectively. Application of the method to the 55 sequences which are outside the training set and known to be cleaved by the HIV‐1 protease accurately predicted 100% of the peptides as substrates of the enzyme. The method also predicted all but one of the sites hydrolyzed by the protease in native HIV‐1 and HIV‐2 reverse transcriptases, where the HIV‐1 protease discriminates between nearly identical sequences in a very subtle fashion. Finally, the algorithm predicts correctly all of the HIV‐1 protease processing sites in the native gag and gag/pol HIV‐1 polyproteins, and all of the cleavage sites identified in denatured protease and reverse transcriptase. The new predictive algorithm provides a novel route toward understanding the specificity of this important therapeutic target.