Comparing in-memory real-space map computed by rs.find_difference_peaks to gemmi sf2map

[maggie-klureza]

As discussed a bit on the discourse forum, I'm hoping to compare the results of running rs.find_difference_peaks vs ccp4's fft and peakmax. This would be made easier by gaining access to the real-space map that rs.find_difference_peaks computes along the way. As rs calculates the map using only gemmi internal tools, it then raises the question of whether that process is fully identical/equal to gemmi's sf2map program. If so, I could then use the latter program as a proxy for rs.find_difference_peaks in-memory map and proceed with my analysis using that.

[JBGreisman]

Internally, it looks like this method is using gemmi to generate the map:

rs-booster/rsbooster/realspace/find_peaks.py

Lines 239 to 243 in 9586eed

	mtz = mtz.to_gemmi()
	grid = mtz.transform_f_phi_to_map(
	parser.structure_factor_key, parser.phase_key, sample_rate=parser.sample_rate
	)

It seems plausible that it would be equivalent to sf2map (although the --weight-key option might not have a perfect parallel in the gemmi methods), but would require some testing to confirm one way or another. For what it's worth, the real-space map can always be written to a ccp4 map format: docs

[JBGreisman]

Based on the discourse thread it seems that @kmdalton might be looking into this test, so I'll leave that to him. This short snippet is the sort of thing that could be used to interactively generate a map file (which just represents the 3D real-space map) in case that's useful to you:

import reciprocalspaceship as rs

ds = rs.read_mtz("filename.mtz")
mtz = ds.to_gemmi()
grid = mtz.transform_f_phi_to_map("FWT", "PHWT") # Or whatever your columns are
                                                 # grid object contains the real-space voxel values
rs.io.write_ccp4_map(grid.array, "test.map", grid.unit_cell, grid.spacegroup)

[kmdalton]

Thanks, @JBGreisman. This is helpful info.