Fig 1.
Benchmarking of VBI (A) Selection of optimal ensembles sizes using model evidence. Histogram shows maximum model evidence for all possible ensembles of 2–4 members for a synthetic ensemble consisting of 3 members selected from a library of 10 conformers. (B) Effect of noise on ensemble inference. The magnitude of noise is scaled in relation to the noise in ΔmC2 experimental data (σ = 1). Inferred number of models (orange), number of recovered models from synthetic ensemble and Ng(blue) as function of simulated noise (σ). Synthetic ensemble was generated using 5 models with arbitrary weights and structural library consisted of 100 conformers. (C) Error in weights inference as function of noise. Root mean square deviation (rmsd) between simulated and inferred weights as function of simulated noise. Magnitude of noise defined as in (A). (D) Inference with and without the Rosetta energies. Simulated ensemble with 5 members that are assigned equal population weights. Recovery of number of models in the synthetic ensembles from a library of 100 conformers in the presence (orange, conformers energies: -135.2, -140.0, -126.7, -125.5, -124.0) and absence (blue) of energy prior as a function of noise. Magnitude of noise defined as in (B).
Fig 2.
Bayesian inference of CaM conformational ensembles from (A) SAXS data only, (B) SAXS data with Rosetta energies, (C) SAXS and chemical shift data only and (D) SAXS and chemical shift data with Rosetta energies. (E-H) Population weight distributions for inferred ensembles from all four inference scenarios. (I-L) Ensemble model fit to SAXS data from the point estimate of population weights from VBI. (M-P) Error weighted intensity difference plots for each ensemble model fit to the SAXS data. Structural models were aligned on N-terminal domain (cyan). Different C-terminal orientations (various colors and numbers 1–13) correspond to different conformers.
Fig 3.
Bayesian inference of ΔmC2 conformational ensemble from (A) SAXS data only, (B) SAXS data + Rosetta energies, (C) SAXS data + CS and d) SAXS data + CS + Rosetta energies. (E-H) Population weight distributions for inferred ensembles in all four inference scenarios in A. (I-L) Ensemble model fits to SAXS data based on the point estimate of population weights from VBI. (M-P) Error weighted intensity difference plots for each ensemble. Structural models were aligned on N-terminal domain (cyan). Different C-terminal orientations (various colors and numbers 1–9) correspond to different conformers.