We present the results of our participation in the BioCreative VI: Text mining chemical-protein interactions (CHEMPROT) track. The goal of this task is to promote the development and evaluation of systems capable of extracting relations between chemical compounds/drug and genes/proteins from biomedical literature. We participate with two systems: (1) an SVM system which relies on a rich set of features extracted from the parse graph and (2) an ensemble of neural networks that utilize LSTM networks and generate features along the shortest path of dependencies. We also combine the predictions from the two systems with the goal of increasing performance. On the development set, our system combination approach outperforms the two individual systems, achieving an F-score of 61.09 (according to the official evaluation metric). On the test set, our SVM system achieves the highest result for our submissions with​ ​an​ ​ F-score​ ​ of​ ​ 60.99.