Critical transition is ubiquitous in real-world systems; thus, being able to predict the timing of transition (or bifurcation) is a pressing concern. Various early warning signals (EWS) have been developed and successfully applied in models and natural systems. However, the current EWS are developed without explicitly considering detail causal mechanisms behind. Consequently, these EWS can easily get false positive alarm for the bifurcations. In this study, we aim to develop the EWS derived from mechanistic interaction network. Based on our analysis, we found several network topological properties changed when approaching to the tipping points of bifurcation.