CTCPI – Convolution Tree Kernel-based Chemical-Protein Interaction Detection

In this paper we introduce a chemical-protein interaction detection system called CTCPI, which uses Convolution Tree Kernel to separate various levels of interaction mechanisms exhibited by chemicals while interacting with proteins/genes. Our system enlists a novel feature engineering method based on Algebraic Invariance to identify and consolidate distinct linguistic features for each class from the candidate sets and use these feature patterns as a screening function for generating the feature tree for SVM-CTK classifier. Our system achieved about 30.92% performance for chemical protein interaction class identification task.