HATSA Workflow Examples — hatsa-workflows • hatsa

This documentation provides comprehensive examples of common HATSA workflows, from basic alignment to advanced task-informed analyses.

Basic Workflow


# Load your data (list of subject matrices)
data <- load_fmri_data()  # Returns list of time × voxel matrices

# Get parameter suggestions
params <- hatsa_suggest(data)

# Run basic HATSA
result <- hatsa(data, components = params$components)

# Extract aligned data
aligned_data <- get_aligned_data(result)
template <- get_template(result)

# Check quality
hatsa_summary(result)
plot_hatsa(result, type = "eigenvalues")

Task-Informed Workflow


# Load task design matrices
task_data <- load_task_designs()  # List of time × condition matrices

# Automatic method selection
result <- hatsa_task(data, task_data, method = "auto")

# Or choose specific method
result <- hatsa_task(data, task_data,
                     method = "blend",
                     lambda_blend = 0.2)

# Analyze task-specific alignment
task_metrics <- get_task_alignment_metrics(result)

Advanced Anchor Selection


# Manual anchor selection based on ROI
roi_indices <- get_roi_voxels("visual_cortex")
result <- hatsa(data, anchors = roi_indices)

# Multi-resolution anchor selection
anchors <- select_anchors_mra(
  U_list = preliminary_decomposition,
  n_anchors = 100,
  n_resolutions = 5
)
result <- hatsa(data, anchors = anchors)

Preprocessing Integration


# HATSA works best with preprocessed data
data_clean <- lapply(data, function(X) {
  X <- scale(X)  # Z-score time series
  X[is.na(X)] <- 0  # Handle missing data
  X
})

# Run with custom preprocessing
result <- hatsa(data_clean, preset = "accurate")

Group Analysis


# Align multiple groups separately
result_controls <- hatsa(data[control_idx])
result_patients <- hatsa(data[patient_idx])

# Compare alignment quality
compare_alignments(result_controls, result_patients)

# Project new subjects to existing space
new_aligned <- predict(result_controls, newdata_list = new_subjects)

Performance Optimization


# For large datasets, use fast preset
result <- hatsa(big_data, preset = "fast")

# For parallel processing (if available)
options(hatsa.parallel = TRUE)
options(hatsa.cores = 4)
result <- hatsa(data)

# For very high-dimensional data
# First reduce dimensions
data_reduced <- lapply(data, function(X) {
  svd_X <- svd(X, nu = 100, nv = 100)
  svd_X$u 
})
result <- hatsa(data_reduced)