Run PCA on compression simulations

Notebooks contains steps for plotting combined Cytosim and Readdy compression velocity simulations and applying Principal Component Analysis (PCA) on individual fibers. By default, fibers coordinates are aligned in the yz-plane to the positive y axis, keeping x axis coordinates unchanged, before running PCA.

• Define simulation conditions

• Load merged data

• Save aligned fibers

• Plot aligned fibers

• Run PCA

• Save PCA results

• Save PCA trajectories

• Save PCA transforms

• Plot PCA feature scatter

• Plot PCA inverse transform

if __name__ != "__main__":
    raise ImportError("This module is a notebook and is not meant to be imported")

import pandas as pd
from io_collection.save.save_pickle import save_pickle

from subcell_pipeline.analysis.dimensionality_reduction.fiber_data import (
    get_merged_data,
    plot_fibers_by_key_and_seed,
    save_aligned_fibers,
)
from subcell_pipeline.analysis.dimensionality_reduction.pca_dim_reduction import (
    plot_pca_feature_scatter,
    plot_pca_inverse_transform,
    run_pca,
    save_pca_results,
    save_pca_trajectories,
    save_pca_transforms,
)

Define simulation conditions

Defines the ACTIN_COMPRESSION_VELOCITY simulation series, which compresses a single 500 nm actin fiber at four different velocities (4.7, 15, 47, and 150 μm/s) with five replicates each (random seeds 1, 2, 3, 4, and 5).

# Name of the simulation series
series_name: str = "ACTIN_COMPRESSION_VELOCITY"

# S3 bucket Cytosim for input and output files
cytosim_bucket: str = "s3://cytosim-working-bucket"

# S3 bucket ReaDDy for input and output files
readdy_bucket: str = "s3://readdy-working-bucket"

# Random seeds for simulations
random_seeds: list[int] = [1, 2, 3, 4, 5]

# List of condition file keys for each velocity
condition_keys: list[str] = ["0047", "0150", "0470", "1500"]

# Location to save analysis results (S3 bucket or local path)
save_location: str = "s3://subcell-working-bucket"

Load merged data

Load merged simulation data from Cytosim and ReaDDy. Data is aligned in the yz-plane to the positive y axis, keeping x axis coordinates unchanged. Set align=False to load un-aligned data instead.

readdy_data = get_merged_data(readdy_bucket, series_name, condition_keys, random_seeds)
readdy_data["simulator"] = "readdy"

cytosim_data = get_merged_data(
    cytosim_bucket, series_name, condition_keys, random_seeds
)
cytosim_data["simulator"] = "cytosim"

data = pd.concat([cytosim_data, readdy_data])
data["repeat"] = data["seed"] - 1
data["velocity"] = data["key"].astype("int") / 10

Save aligned fibers

time_map = {
    ("cytosim", "0047"): 0.031685,
    ("cytosim", "0150"): 0.01,
    ("cytosim", "0470"): 0.00316,
    ("cytosim", "1500"): 0.001,
    ("readdy", "0047"): 100,
    ("readdy", "0150"): 100,
    ("readdy", "0470"): 100,
    ("readdy", "1500"): 100,
}

save_aligned_fibers(
    data,
    time_map,
    save_location,
    "dimensionality_reduction/actin_compression_aligned_fibers.json",
)

Plot aligned fibers

plot_fibers_by_key_and_seed(
    data, save_location, "dimensionality_reduction/actin_compression_aligned_fibers.png"
)

Run PCA

pca_results, pca = run_pca(data)

Save PCA object

save_pickle(save_location, "dimensionality_reduction/actin_compression_pca.pkl", pca)

Save PCA results

The PCA results are saved with resampled rows, which shuffles the order of the entries. Pre-shuffled data is useful for scatter plots showing each individual

save_pca_results(
    pca_results,
    save_location,
    "dimensionality_reduction/actin_compression_pca_results.csv",
    resample=True,
)

Save PCA trajectories

save_pca_trajectories(
    pca_results,
    save_location,
    "dimensionality_reduction/actin_compression_pca_trajectories.json",
)

Save PCA transforms

points: list[list[float]] = [
    [-900, -600, -300, 0, 300, 600],
    [-600, -400, -200, 0, 200],
]

save_pca_transforms(
    pca,
    points,
    save_location,
    "dimensionality_reduction/actin_compression_pca_transforms.json",
)

Plot PCA feature scatter

features = {
    "SIMULATOR": {"READDY": "red", "CYTOSIM": "blue"},
    "TIME": "magma_r",
    "VELOCITY": (
        {
            4.7: 0,
            15: 1,
            47: 2,
            150: 3,
        },
        "magma_r",
    ),
    "REPEAT": "viridis",
}

plot_pca_feature_scatter(
    pca_results,
    features,
    pca,
    save_location,
    "dimensionality_reduction/actin_compression_pca_feature_scatter.png",
)

Plot PCA inverse transform

plot_pca_inverse_transform(
    pca,
    pca_results,
    save_location,
    "dimensionality_reduction/actin_compression_pca_inverse_transform.png",
)