Nested sampling’s fitting results in better-fitting simulations than simulations using randomly selected parameter values.

Left, Data distribution, prior predictive distribution, and posterior predictive distribution for each dataset and all candidate models. Data is represented by a Beta distribution, bounded by zero and one, and used in the likelihood function input for Multinest (see Methods). Prior predictive distribution represents model simulations using parameters randomly drawn from the prior. Posterior predictive is generated by model simulations using best-fitting parameters returned by Multinest. Right, prior predictive distribution, simulation steady-state proportions using independently-sampled posterior marginal parameters (“subtype independent post. predictive”), and simulation steady-state proportions using parameters sampled from the joint posterior distribution (“subtype joint posterior predictive”, same as posterior predictive distribution on the left). See S3 Text for more detail and discussion related to these results. Data and predictive distributions for each dataset shown. (A) TKO, (B) RPM, (C) SCLC-A cell lines. (TIFF)