Large-scale atomistic simulations provide invaluable understandings for the mechanical response of structural materials under complex loads and temperatures. Accurate interatomic potentials (IPs) are essential for obtaining reliable results. In this thesis, we introduce the extended modified embedded-atom method (XMEAM) by extending partial electron density functions and screening parameters in the classical MEAM. The XMEAM model is firstly applied to Vanadium (V), one of the body-centered cubic (BCC) transition metals (TMs). XMEAM-V is constructed on a training dataset generated from extant experiments and density-functional theory (DFT) calculations. We utilize the particle swarm optimization algorithm to determine the parameters in the XMEAM formulation. XMEAM-V is validated on a wide range of mechanical and thermodynamic properties of V, showing the good capability and transferability of the XMEAM model compared to classical EAM/MEAM and recently-developed machine learning models. Additionally, we propose a physics-based materials index 𝜒, the energy difference between the BCC and face-centered cubic (FCC) structures, to control the screw dislocation associated properties in BCC metals and alloys. The lattice friction and nucleation barrier have near-linear scaling with 𝜒 and the core transforms from non-degenerate (ND) to degenerate (D) when 𝜒 drops below certain thresholds in binary BCC TM alloys. The index 𝜒 is related to valence electron concentrations, which can be effectively tuned and quantitatively predicted by DFT calculations for average alloying effects. The ND-D transition with 𝜒 is also seen in 72 existing BCC IPs, demonstrating critical insights for further BCC potential development. Finally, XMEAM-Nb and XMEAM-Mo are developed with the focus on screw dislocation and twinning properties. These newly-developed XMEAM potentials (V, Nb and Mo) are suitable for investigating the plastic deformation and fracture behaviours in BCC TMs.