Abstract
The convex hull is outmost shell of structure. It is the part of contact environment frequently. Intuitively, The convex hull of protein’s structure affect the activity of protein. We study the features of convex hull structure with neural network algorithm.First, we get the training set Ⅰ include peptide’s structure and activity from Dr. Lin. We train this set with neural network algorithm and jacknife cross validation. It maybe applies to drug design from accurate prediction of the peptide structural activity.Second, we extract protein profiles from PDB format files. The protein profiles is the information of the convex hull of proteins. Next, we enumerate different propensity with different type amino acid appear over the convex hull of protein. The convex propensity of different type amino acid is approximate coincidence with hydrophobicity. Subsequently, the propensity value of different type amino acid with a sliding windows was created training set Ⅱ. Final, The set is applied to prediction of the convex hull residue of protein with neural network algorithm and linear regression. The result is 82.79% accuracy for neural network algorithm and 82.53% accuracy for linear regression. Our expectation is accurate prediction of protein tertiary structure protein. We also expect the profile of protein convex hull residue will be a good index for bioinformation and will improve the speed and accuracy on database search.