Separate structurally heterogenous data sets into homogeneous classes by a multi-reference 3D-angular refinement using a maximum-likelihood(ML) target function.

Parameters

$: Metadata or stack with input images $`--nref <int`1> $`--ref <reference_file`> $`--oroot <rootname`mlf2d> $: Also check mirror image of each reference $--thr <N1> $--iem <blocks1> $--no_ctf <pixel_size1> $--ang <float10>

ML additional options: $--eps <float5e-5> $--iter <int25> $--psi_step <float5.> $--noise <float1> $--offset <float3.> $--frac <docfile> $-C <double1e-12> $: Kick-start the fast algorithm from all-zero offsets $: Do not re-estimate the standard deviation in the pixel noise $: Do not re-estimate the standard deviation in the origin offsets $: Do not re-estimate the model fractions $--student <df6> =: Use t-distributed instead of Gaussian model for the noise df Degrees of freedom for the t-distribution: Refined normalization parameters for each particle $: Save memory A(deprecated) $: Save memory B(deprecated) $--search_shift <int3> $--reduce_snr <factor1> $: Use this if the experimental images have not been phase flipped $: Use this if the references (-ref) are not CTF-deconvoluted $--limit_resolution <first_high0> <high=0> <low=999> $--fix_high <float-1> $``:

Additional options: $--recons <recons_typewlsART> where <recons_type> can be:

• • $: Start wlsART reconstructions from all-zero volumes $`--sym <symfile`c1> $`--low_pass <freq`-1> $`--sym_mask <maskfile`> $`--tilt <min`-91.> <max=91.> $: Randomly perturb reference projection directions