Practical Example

The following outlines potential applications for template matching using match_template.py.

• Picking Ribosomes
  • Template and Mask Generation
  • Template Matching
  • Parameter Comparison
  • Remarks on Tomogram Preprocessing
  • References
• Fitting Atomic Structures
• Alignment of Densities

You will learn how to analyze the output of these experiments in the Postprocessing section.

Configuration

The following provides a background on some of the parameters of match_template.py, classified based on their area of influence. An overview of all available parameters can be acquired using

match_template.py -h

Input

• target/template: Specifies the paths to your target and template files, which are essential inputs for match_template.

• target_mask/template_mask: Define masks for the target and template, allowing the tool to focus only on regions of interest and potentially improve accuracy.

• invert_target_contrast: Whether to invert the target’s contrast and perform linear rescaling between zero and one. This option is intended for targets, where the object to-be-matched has negative values, i.e. tomograms with black white contrast.

Performance

• cutoff_target/cutoff_template: By defining a minimal enclosing box containing all significant density values, the execution time can be significantly improved.

• pad_edges: Pad each dimension of the target by the template shape. This helps to avoid erroneous scores at the boundaries. pytme will automatically set this flag if the target has to be split to fit into memory, as otherwise grid-like artifacts will be present in the scores. Setting this option will typically yield a longer runtime.

• pad_fourier: Zero pad the target to the full convolution shape, which is defined as cumulative box size of template and target minus one. Setting this option improves numerical stability, but yields a longer runtime. When working with large data such as tomograms this flag does not need to be set.

• no_centering: Omit moving the template’s center of mass to the center of a new box that is sufficiently sized to represent all rotations.

• interpolation_order: Defines the spline interpolation order for rotations. While higher values ensure more accuracy reducing it can lead to performance improvements, especially on CPU, at a slight accuracy trade-off.

• use_mixed_precision: By utilizing float16 for GPU operations, memory consumption is lowered, and certain hardware might observe a performance boost.

• use_memmap: When working with large inputs, this option lets the tool offload objects to the disk, making computations feasible even with limited RAM, though this comes with a slight IO overhead.

Accuracy

• score: The choice of the scoring function plays a pivotal role in the accuracy of results. Note, that if your mask is not symmetric nor encompasses all possible rotations of the template you have to use a scoring method that also rotates the mask, such FLC or MCC.

• angular_sampling: Granularity of rotations. A lower value will sample more rotations, potentially yielding more accurate results but at a computational cost.

• scramble_phases: This option scrambles the phases of the templates to simulate a noise background, aiding in refining the score space.

Computation

• memory_scaling: Fraction of available memory that will be used. Corresponds to total amount of RAM for CPU backends, and sum of available GPU memory for GPU backends.

• ram: Memory in bytes that should be used for computations. If set will overwrite the value determined by memory_scaling.

• use_gpu: Wheter to perform computations using a GPU backend.

• gpu_indices: GPU device indices to use for computation, e.g. 0,1. If this parameter is provided, the shell environment CUDA_VISIBLE_DEVICES that otherwise specifies the GPU device indices will be ignored.