ahbarnett / toybife

simple experiments in Bayesian inference of free energy and path selection for cryo-EM

toyBIFE

simple proposals and experiments in Bayesian inference of free energy and path selection for cryo-EM.

Alex Barnett, Pilar Cossio, et al... June 2021.

See notes and utils for Alex's stuff.

Pilar's group added a new folder 2DHsp90 with 3 subfolders:

FES: Data for vizualizing the 2D distribution or free energy surface.

Files in folder:

• Num_Images_grid: number of images at each x,y point (in matrix form 20 by 20). This is the 'scatter plot'.
• 2D-FES_plot.py: python code to graph FES using Num_Images_grid and the two reference paths (black and orange). Run: python3.6 2D-FES_plot.py Output: 3-wells-paths.png

LikeLihood: Data for calculating the log-likelihood given a predetermined path and G that maximizes it.

Files in folder:

• likelihood_calc.py: calculates the log-L (total and for each image) given a path and Gmax. Run: python3.6 likelihood_calc.py [FILE_PATH] [FILE_GMAX] > Log-L-image e.g., FILE_PATH=Path-black and FILE_GMAX=Gmax-black Output: log-L of individual images and final line total log-L.
• Path-black and Path-orange: Node indexes (following Julian's notation) of the black and orange paths
• Gmax-black and Gmax-orange: Gs that maximize cryoBIFE likelihood.
• all_images: LogP BioEM for all images (rows) and the 225 models (columns) within the green square (slide 8 alexdiscussion.pdf; typo there not 200 but 225)

cryoBIFE: Data for running cryoBIFE given a predetermined path.

Files in folder:

• cryoBIFE.py: MCMC from paper using Alex's logsumexp function. Run: python3.6 cryoBIFE.py [FILE_PATH] > LogL-[FILE_PATH] Output: LogL-[FILE_PATH] logL at each MCMC step, an additional output file "Gs" has the Gs at each MCMC step.
• get-max.sh: bash script to get Gs that maximize Log-L. Run: ./get-max.sh LogL-[FILE_PATH]
• Check-logsumexp: Checking the results from using Alex's routine and the old (correct) one that we had (see commented-out line 148 - 160)
• all_images: same as above.
• Path-black and Path-orange: same as above.

toybife's Issues

confused why interface to logweisumexp Py version differs from my matlab

toyBIFE/2DHsp90/cryoBIFE/cryoBIFE.py

Line 20 in 284dc4d

def w_logsumexp(x, weight, maxalpha): ##To calculate the weighted logsumexp (Alex's idea). In our case, x is the

I had deliberately made the inputs simply x and w (in my notation; see utils/logweisumexp.m).
Nothing else is needed.
So, I don't understand what the input maxalpha is - this is confusing and potentially a source of error.
There is no speedup to be gained by avoiding the max over x, since the exp is always slower.

Also, could you add Py docstring so we know exactly what it does, and add a separate test routine?

Cheers, Alex