Molecular communication (MC) can play an indispensable role in nanonetworking and internet of bio-nano things based applications. In most MC receivers, the detection of data symbols requires the optimal threshold, which depends on the accurate diffusion channel impulse response (DCIR), and statistics of noise and interference (SNI). In order to estimate these parameters, a training phase must be carried out. Further, DCIR and SNI can change due to diffusion, drift, temperature and interference variation in the MC system, which results in complex training or sub-optimal MC performance. In this paper, we propose a coded modulation scheme (CMS) for MC, which does not require threshold estimation at the receiver. Hence, the proposed CMS is completely free from any training, and also independent of noise and interferences variation in MC systems. The proposed CMS has optimal system performance with negligible system complexity as verified through the numerical results. Further, the impact of various parameters such as diffusion coefficient, transmission distance, symbol duration, etc. are also examined for CMS.