MCP: a multi-component learning machine for prediction of protein secondary structure
share
Proteins biological function is tightly connected to its specific 3D structure. Prediction of protein secondary structure is a crucial intermediate step towards elucidating its 3D structure and function. The average accuracy of the previous machine learning approaches has hardly reached beyond 80 possible underlying reasons are abstruse sequence structure relation, noise, class imbalance and high dimensionality of encoding schemes which represent protein sequences. In this paper, we have developed an accurate multi-component prediction machine to overcome challenges of protein secondary structure prediction. The principal tenet behind the proposed approach is to directly process amino-acid sequences to reveal deeper and, simultaneously, more biologically meaningful sequence structure relation. Taking this approach, it is possible to prevent losing rich information hidden in sequence data, which is biologically believed to be sufficient for structure adoption. Additionally, it facilitates resolving the high dimensionality of the numerical representation for protein sequences. Moreover, the multi-component designation can better address the high complexity of the relation between sequence and structure. To pursue these objectives, we have employed various components in our framework. The learning components of our framework are SVM and fuzzy KNN along with edit kernel and a compound dissimilarity measure symbolized as e d. Also a fuzzy aggregation pool and a biological altering module are utilized. Utilizing the popular RS126 dataset, our multi-component framework demonstrates a prediction accuracy of 87.29 effectiveness can be further improved via parameter optimization.
READ FULL TEXT