Prediction of protein-protein interaction sites

Protein-Protein Interaction Sites Prediction

Why a network in protein-protein interactions?

Protein is an important component of all cells and tissues of the human body and an important material basis for life activities. However, the function of living organisms is not performed independently by a single protein, but is achieved through the interaction of protein and protein. If the protein interaction is abnormal, it will affect the cell activity and function, and cause various diseases. For protein interaction, in essence, it is realized by the mutual combination of some residues (amino acids from which water molecules are removed) on the protein. These residues are called protein interaction sites. By studying the binding sites of the interaction, it is possible to predict which residues on the protein are involved in the protein interaction. The research has important influence and significance on understanding the mechanism of life activities, exploring the principle of protein interaction, and discovering new drug target protein interaction relationships.

Figure 1. Protein-protein interaction sites.

Overall solutions

For the study of protein interaction site prediction, the method of biological experiment is adopted, which not only takes a long period but also consumes a lot of manpower and material resources. Therefore, computational methods to predict protein interaction sites have become the current mainstream method.

MedAI summarized the previous calculation methods used for protein interaction site prediction, many of which usually have the following problems:

• They often take the characteristic information of multiple adjacent residues, and then vectorize them in one dimension, and then input them Different learning algorithms. However, this one-dimensional vectorization operation destroys the contextual relationship between residues, thus losing some important characteristic information.
• Most machine learning methods used in traditional methods do not have the ability to learn the residual context, which will lead to unsatisfactory prediction results.
• Traditional calculation methods will combine features of different properties indiscriminately and send them to the algorithm for learning. However, considering the differences in the biological expression of features of different properties, The undifferentiated combination of these features may affect the saliency of the original features.

Traditional Computational Algorithm

• Sequence-based methods.
  Methods based on sequence information use features extracted from protein sequences to predict protein interaction sites.
• Structure-based methods.
  Knowledge of the three-dimensional (3D) structure of the protein complex provides much valuable information on the protein interaction sites.
• Methods based on integrated information.
  Three-dimensional structure of proteins are more difficult and expensive to elucidate than protein sequences, so its magnitude in protein structure databases such as the Protein Data Bank (PDB) is remarkably smaller compared to that of sequences in protein sequence databases like UniProt.

AI-based Algorithm

• MedAI first proposed the concept of feature map to characterize the characteristic information of residues. The feature map not only contains the original characteristic information of the residue, but also contains the contextual characteristic information of the residue.
• We use a deep convolutional neural network to construct a feature map, where the local connection and weight sharing ability of the convolutional neural network can extract the contextual relationship of the feature map and the feature information of the residue. In addition, deep convolution Then the high-dimensional abstract information in the feature map can be extracted.
• Considering the different biological expressions of each type of original feature, for feature maps constructed based on different original features, we train the deep convolutional neural network learner separately, then use ensemble learning to integrate them together, which can avoid the influence of the undifferentiated combination of features of different natures, and also improve the predictive ability of the model.

The Prediction of protein-protein interaction sites provided by MedAI can help you obtain many important information in biology, because they can regulate almost all cellular processes, including metabolic cycles, DNA transcription and replication, different signal cascades and mny other processes.