Dependable properties such as self-stabilization are crucial requirements in sensor networks. One way to achieve these properties is to utilize the vast literature on distributed systems where such self-stabilizing algorithms have been designed. Since these existing algorithms are designed in read/write model (or variations thereof), they cannot be directly applied in sensor networks. For this reason, we consider a new atomicity model, write all with collision (WAC), that captures the computations of sensor networks and focus on transformations from read/write model to WAC model and vice versa. We show that the transformation from WAC model to read/write model is stabilization preserving, and the transformation from read/write model to WAC model is stabilization preserving for timed systems. In the transformation from read/write model to WAC model, if the system is untimed (asynchronous) and processes are deterministic then under reasonable assumptions, we show that (1) the resulting program in WAC model can allow at most one process to execute, and (2) the resulting program in WAC model cannot be stabilizing.