Constructing interpretable parameter estimates in complex models using inverse regression and Approximate Bayesian Computation