Skip to content

Conformational State Refinement (refine3D_states)

Problem

Given particles with a common pose scaffold from consensus refinement, refine several same-lineage conformational maps and their particle assignments. The shared estimator, bandwidth schedule, and final reconstruction are described in the heterogeneity-analysis overview.

The input may contain an existing label set with project state maps, the flex_pca initializer (the default for state-0/1 input), the stochastic state initializer (flex=no), or an abinitio3D docked checkpoint. Supplied reference volumes are not accepted; that input belongs to classify3D_refs.

Algorithm

pose_policy controls how much of the pose may move while states are decided:

  • fixed freezes the projection direction and optimizes only the in-plane angle, shift, and state. This is classification given the consensus geometry.
  • local permits the direction to move within the coarse Voronoi cell of its previous value. A bound alpha in degrees becomes nspace_sub = min(5000, max(2, 2/(1 - cos alpha))), matching the number of cells to the solid angle of that cone.
  • global (the default) performs a full coarse peak search across all directions and states in every iteration.

The workflow samples 10,000 particles per state, capped at 100,000 overall, and balances the sample over projection-direction bins so every view contributes to the state decision. State and pose assignments then follow the common likelihood-ratio and probabilistic-assignment rules before each state is reconstructed from its members.

Rationale

A shared pose and map lineage makes direct Euclidean state competition valid. The selectable pose policy separates nearly fixed-pose classification from local or global conformational refinement, while balanced view sampling keeps an initially larger state from winning solely through better angular coverage.

Implementation

  • Workflow: src/main/commanders/simple/simple_commanders_refine3D.f90.
  • Frequency schedule: src/main/simple_refine3D_stage_plan.f90.
  • State assignment: src/main/strategies/search/probabilistic/simple_eul_prob_tab.f90 and src/main/strategies/search/probabilistic/simple_strategy3D_prob.f90.
  • Policy: doc/policies/heterogeneity/refine3D_states_policy.md.