Data provenance and model complexity are recurring challenges when simulating neural function, particularly signaling, at the level of detail needed to address disease conditions. Here we report a data-driven modeling and model abstraction framework applied to ~40 signaling pathways involved in translation and FXS. FXS is an X-linked neurodevelopmental disorder caused by the loss of FMRP which is an essential regulator for local translation at synapses. Bowling et al., 2019 had suggested that the de-novo translation profiles tend to differ in FXS mouse at steady-state and in the presence of mGluR5 agonist. Thus, we wanted to study how the mGluR5 signaling cascade is affected in FXS using an in-silico model of rodent dendritic spine. We used detailed biochemical (Mass-action+ODE) modeling to study synaptic translation in FXS neurons. We drew on ~320 published experiments, spanning measurements from receptors such as mGluR1/5, TrKB, EGFR and b2AR, to intervening kinases, to protein synthesis. The resultant model has ~250 reactions and 350 protein pools, making it difficult to parameterize. Therefore, we developed a pipeline to hierarchically optimize the model to fit experiments and score subsections of the model, based on how closely the model outcome matches experiments. To anchor this enormous parameter fitting process, we built abstract models with all major nodes of the detailed model using the HillTau formulation (Bhalla, 2020 bioRxiv) but with a much reduced parameter space of ~180 parameters. These abstract models were used to synthesize experiments to provide a scaffold of input-output properties that the detailed model subsets must fit. The resulting models will be used to understand how different signalling cascades are affected in a FXS neuron. In summary, we have developed a principled, hierarchical methodology to use experiments to generate both detailed and abstract models of complex cellular signaling, which are all valuable resources for the field.