Understanding the mechanism of conformational changes in macromolecules requires the knowledge of the intermediate states. A version of the string method, which uses multiple short dynamics trajectories to propagate the pathway, was recently developed by Pan et al. Here we use data from swarms of trajectories calculated at discrete points in phase space to interpolate the average displacement and variance at arbitrary points. This is tested on model potentials using statistics from actual swarms of trajectories. We use the interpolated parameters to compute the Markovian propagators from one point on the transition path to the next. We use them to obtain a time-dependent action of a path, which can be optimized to produce the highest probability pathway. We describe the optimization protocol and demonstrate that in artificial flat potentials the existing string method cannot correct problems such as loops in the initial path, while the new method produces the correct pathway (Figure shows pathway in 2D potential). We further illustrate the utility of our method by applying it to protein conformational transitions, such as the KcsA potassium channel, and comparing its performance to existing transition pathway methods.