While scientific collaboration can be critical for a scholar, some collaborator(s) can be more significant than others, a.k.a. lifetime collaborator(s). This work-in-progress aims to investigate whether it is possible to predict/identify lifetime collaborators given a junior scholar's early profile. For this purpose, we propose a supervised approach by leveraging scholars' local and network properties. Extensive experiments on DBLP digital library demonstrate that lifetime collaborators can be accurately predicted. The proposed model outperforms baseline models with various predictors. Our study may shed light on the exploration of scientific collaborations from the perspective of life-long collaboration.