Feature Based Method for Predicting Pharmacological Interaction

Prediction of drug target interaction is an extrusive domain discovery and repositioning drugs. Most conventional studies are carried out in early years the wet laboratory, but it very expensive time consuming. So nowadays, use machine learning techniques to predict pairs. A new method targeting drugs introduced this paper. Use Pseudo Position Specific Scoring Matrix (PsePSSM) used represent target, which generate features that describe original information protein. The chemical structure can be extracted through FP2 molecular fingerprint information. Then a network constructed using bipartite graph where each node represents or link indicates interaction. From above stages, data contains some noise redundant have negative impact on prediction output. So, LASSO (Least Absolute Shrinkage Selection Operator) handle reduce dimension feature data. But pair samples imbalanced, then cost-sensitive ensemble address imbalanced problem between positive samples, learns about minority class by assigning higher costs optimizing their cost error. Finally, processed given as input extreme gradient boosting classifier algorithm for predicting This significantly improve accuracy