To evaluate the dynamic effects of the dockless bike-sharing scheme on the demand of the London Cycle Hire (LCH) scheme at the station level, a novel bicycle demand prediction model is proposed using the deep learning approach, based on the transaction records at 645 docking stations of LCH in the period between July 2017 and March 2018. First, an intervention response module (IRM) is established to model the time-series trends of bicycle demands at individual LCH docking stations, with and without the dockless bike-sharing scheme. Then, the Graph Neural Networks (GNN) predictors are adopted to predict the demand for LCH, incorporating the learned effects from IRM. Results indicate that the proposed bicycle demand prediction model can achieve promising prediction performances, with higher R-squared (R), lower Root Mean Squared Errors (RMSE) and lower Mean Absolute Errors (MAE), compared to conventional prediction models. More importantly, the proposed model can recognize the dynamic effects of the dockless bike-sharing scheme on the demand for LCH. For instance, there are possible spillover effects for the influence area of dockless bike-sharing scheme, especially for the neighboring areas that have well-integrated bicycle facilities (e.g., cycle lanes). In addition, the effect of dockless bike sharing on the demand for LCH can magnify over time. Moreover, influences on the demands on weekends are more remarkable than that on weekdays. Findings should improve the understanding on the interdependency between the demands of dockless and docked bike-sharing systems. This should shed light to the optimal management strategy for the docked bike-sharing system that can maximize the operational efficiency and cost-effectiveness.