Table 3. Maximum likelihood inference of spatially resolved anomalous scattering factors for the ferredoxin simulation.

Root-mean-squared agreement between the model and the ground truth anomalous scattering parameters were calculated over the 7105–7136 eV range. The number of crystal lattices used for parameter modeling was always less than the number of diffraction patterns selected for analysis due to the rejection of those images with two or fewer indexed shoeboxes in the Fig. 4 ▸ region of interest. Anomalous scattering factor refinements (24 LBFGS iterations per macrocycle, except the negative control, which used 12) were performed using the crystal rotation model from Method 4, while the spot background level and image scale factors were refined once per macrocycle.

	Starting valence state model for the two metal sites				R.m.s. agreement between model and ground truth scattering factors, including both metal sites m (e−)
					Starting model		Refined model
Comment	Fe1	Fe2	Number of macrocycles	Number of diffraction patterns (lattice models)	Δf′m(λ)	Δf′′m(λ)	Δf′m(λ)	Δf′′m(λ)
Negative control (H,K,L + 1)	+3	+2	1	50000 (33923)	0.0	0.0	7.786	5.673
Ground truth	+3	+2	3	50000 (33923)	0.0	0.0	0.180†	0.165†
Differing valence models as starting guess	+2	+2	3	50000 (33923)	0.265	0.278	0.182	0.167
	+2	+3	3	50000 (33923)	0.374	0.393	0.183	0.168
	+3	+3	3	50000 (33923)	0.265	0.278	0.184	0.164
	0	0	3	50000 (33923)	0.898†	1.235†	0.179†	0.198†
Fewer macrocycles	0	0	2	50000 (33923)			0.199†	0.281†
	0	0	1	50000 (33923)			0.332†	0.514†
Fewer images	0	0	3	1500 (1028)			0.327	0.306
	0	0	3	3000 (2051)			0.297	0.252
	0	0	3	6000 (4097)			0.291	0.246
	0	0	3	12000 (8181)			0.237	0.224
	0	0	3	25000 (17004)			0.191	0.210
Alternate (disjointed) data cohort	0	0	3	50000 (34012)			0.176	0.184

^†Values correspond to the data shown in Fig. 9 ▸.