A limitation in using mathematical models to describe biological data is that they can be subject to error caused by a lack of knowledge or misspecification of the underlying processes leading to the observable data. Conversely, machine learning and data science methods have limited interpretability, yet can still be used to make accurate predictions from data without a mechanistic model. The use of data science methods can be hindered by the high-dimensionality of the system, the required amount of data needed to make accurate predictions, and, may not generalize well to data outside of the observed data distribution. The increasingly rapid collection and availability of time series data in biology provide an opportunity to bridge the gap between mechanistic model based approaches and data science techniques. In this talk, I will describe how the complementary strengths of these different paradigms can be leveraged to enable more robust forecasting. In particular, I will describe how methods from data science can help address model error and how fusing model-free methods into the parameter estimation process can result in more accurate parameter estimates and improved predictions.