Neighbor preferences of amino acids and context-dependent effects of amino acid substitutions in human, mouse, and dog

A reactive center loop-based prediction platform to enhance the design of therapeutic SERPINs

Serine proteases are essential for many physiological processes and require tight regulation by serine protease inhibitors (SERPINs). A disturbed SERPIN-protease balance may result in disease. The reactive center loop (RCL) contains an enzymatic cleavage site between the P1 through P1' residues that controls SERPIN specificity. This RCL can be modified to improve SERPIN function; however, a lack of insight into sequence-function relationships limits SERPIN development. This is complicated by more than 25 billion mutants needed to screen the entire P4 to P4' region. Here, we developed a platform to predict the effects of RCL mutagenesis by using α1-antitrypsin as a model SERPIN. We generated variants for each of the residues in P4 to P4' region, mutating them into each of the 20 naturally occurring amino acids. Subsequently, we profiled the reactivity of the resulting 160 variants against seven proteases involved in coagulation. These profiles formed the basis of an in silico prediction platform for SERPIN inhibitory behavior with combined P4 to P4' RCL mutations, which were validated experimentally. This prediction platform accurately predicted SERPIN behavior against five out of the seven screened proteases, one of which was activated protein C (APC). Using these findings, a next-generation APC-inhibiting α1-antitrypsin variant was designed (KMPR/RIRA; / indicates the cleavage site). This variant attenuates blood loss in an in vivo hemophilia A model at a lower dosage than the previously developed variant AIKR/KIPP because of improved potency and specificity. We propose that this SERPIN-based RCL mutagenesis approach improves our understanding of SERPIN behavior and will facilitate the design of therapeutic SERPINs.

Convolutional Neural Network Based Approach to in Silico Non-Anticipating Prediction of Antigenic Distance for Influenza Virus

Evaluation of the antigenic similarity degree between the strains of the influenza virus is highly important for vaccine production. The conventional method used to measure such a degree is related to performing the immunological assays of hemagglutinin inhibition. Namely, the antigenic distance between two strains is calculated on the basis of HI assays. Usually, such distances are visualized by using some kind of antigenic cartography method. The known drawback of the HI assay is that it is rather time-consuming and expensive. In this paper, we propose a novel approach for antigenic distance approximation based on deep learning in the feature spaces induced by hemagglutinin protein sequences and Convolutional Neural Networks (CNNs). To apply a CNN to compare the protein sequences, we utilize the encoding based on the physical and chemical characteristics of amino acids. By varying (hyper)parameters of the CNN architecture design, we find the most robust network. Further, we provide insight into the relationship between approximated antigenic distance and antigenicity by evaluating the network on the HI assay database for the H1N1 subtype. The results indicate that the best-trained network gives a high-precision approximation for the ground-truth antigenic distances, and can be used as a good exploratory tool in practical tasks.

Distribution of dipeptides in different protein structural classes: an effort to find new similarities

Finding any regularity in the sequences of proteins and determining their correlation with structural features are of great interest for an understanding of molecular biology. We statistically analyzed the relative frequencies of all 400 possible dipeptides in a data set containing randomly selected proteins of different defined structural classes including all-alpha, all-beta, alpha + beta and alpha/beta families. We found that the distribution of dipeptides is not the same for different structural classes, and some of them are significantly far from a random distribution. A tendency of a given amino acid to localize in the first or second position of a dipeptide depending on the structural class of protein was also found. Interestingly, some amino acids may be substituted for each other in the first or second positions of specific dipeptides in each structural class. This finding apparently contrasts with the routine expectation from the viewpoint of amino acid properties, as classically understood.

Chameleon sequences in neurodegenerative diseases

Chameleon sequences can adopt either alpha helix sheet or a coil conformation. Defining chameleon sequences in PDB (Protein Data Bank) may yield to an insight on defining peptides and proteins responsible in neurodegeneration. In this research, we benefitted from the large PDB and performed a sequence analysis on Chameleons, where we developed an algorithm to extract peptide segments with identical sequences, but different structures. In order to find new chameleon sequences, we extracted a set of 8315 non-redundant protein sequences from the PDB with an identity less than 25%. Our data was classified to "helix to strand (HE)", "helix to coil (HC)" and "strand to coil (CE)" alterations. We also analyzed the occurrence of singlet and doublet amino acids and the solvent accessibility in the chameleon sequences; we then sorted out the proteins with the most number of chameleon sequences and named them Chameleon Flexible Proteins (CFPs) in our dataset. Our data revealed that Gly, Val, Ile, Tyr and Phe, are the major amino acids in Chameleons. We also found that there are proteins such as Insulin Degrading Enzyme IDE and GTP-binding nuclear protein Ran (RAN) with the most number of chameleons (640 and 405 respectively). These proteins have known roles in neurodegenerative diseases. Therefore it can be inferred that other CFP's can serve as key proteins in neurodegeneration, and a study on them can shed light on curing and preventing neurodegenerative diseases.