Bayesian networks (BNs) and dynamic Bayesian networks (DBNs) are becoming more widely used as a way to learn various types of networks, including cellular signaling networks, from high-throughput data. Due to the high cost of performing experiments, we are interested in developing an experimental design for time series data generation. Specifically, we are interested in determining properties of time series data that make them more efficient for DBN modeling. We present a theoretical analysis on the ability of DBNs without hidden variables to learn from proteomic time series data. The analysis reveals, among other lessons, that under a reasonable set of assumptions a fixed budget is better spent on collecting many short time series data than on a few long time series data.