Comparing t-SNE Hyperparameters in the Hydra Atlas

In [1]:

import random
import scanpy as sc
import numpy as np
import urllib.request
import tempfile
random.seed(20250409)

temp_file = tempfile.mktemp(suffix=".h5ad")
urllib.request.urlretrieve("https://uwmadison.box.com/shared/static/gp1d9akvyq5r7a0rzxw27n86zlqe0d7f.h5ad", temp_file)
adata = sc.read_h5ad(temp_file)
adata.X = adata.X.todense()

sc.pp.normalize_total(adata)
sc.pp.log1p(adata)
sc.pp.highly_variable_genes(adata, n_top_genes=1000)

n_neighbors = 50
ix = np.random.choice(len(adata), 2000, replace=False)
adata = adata[ix, adata.var.highly_variable]
sc.pp.neighbors(adata, n_neighbors=n_neighbors)
sc.tl.tsne(adata, n_pcs=30, perplexity=500)

import random import scanpy as sc import numpy as np import urllib.request import tempfile random.seed(20250409) temp_file = tempfile.mktemp(suffix=".h5ad") urllib.request.urlretrieve("https://uwmadison.box.com/shared/static/gp1d9akvyq5r7a0rzxw27n86zlqe0d7f.h5ad", temp_file) adata = sc.read_h5ad(temp_file) adata.X = adata.X.todense() sc.pp.normalize_total(adata) sc.pp.log1p(adata) sc.pp.highly_variable_genes(adata, n_top_genes=1000) n_neighbors = 50 ix = np.random.choice(len(adata), 2000, replace=False) adata = adata[ix, adata.var.highly_variable] sc.pp.neighbors(adata, n_neighbors=n_neighbors) sc.tl.tsne(adata, n_pcs=30, perplexity=500)

In [2]:

from distortions.geometry import Geometry, bind_metric, local_distortions

embedding = adata.obsm["X_tsne"].copy()
radius = 3 * np.mean(adata.obsp["distances"].data)
geom = Geometry("brute", laplacian_method="geometric", affinity_kwds={"radius": radius}, adjacency_kwds={"n_neighbors": n_neighbors}, laplacian_kwds={"scaling_epps": 5})
H, Hvv, Hs = local_distortions(embedding, adata.X, geom)
embedding = bind_metric(embedding, Hvv, Hs)
embedding["ident"] = adata.obs["ident"].str.split("-").str[1].values

from distortions.geometry import Geometry, bind_metric, local_distortions embedding = adata.obsm["X_tsne"].copy() radius = 3 * np.mean(adata.obsp["distances"].data) geom = Geometry("brute", laplacian_method="geometric", affinity_kwds={"radius": radius}, adjacency_kwds={"n_neighbors": n_neighbors}, laplacian_kwds={"scaling_epps": 5}) H, Hvv, Hs = local_distortions(embedding, adata.X, geom) embedding = bind_metric(embedding, Hvv, Hs) embedding["ident"] = adata.obs["ident"].str.split("-").str[1].values

In [3]:

import altair as alt
alt.data_transformers.enable("vegafusion")

alt.Chart(embedding).mark_circle(opacity=1).encode(
    x=alt.X("embedding_0"),
    y=alt.Y("embedding_1")
).properties(width=400, height=400)

import altair as alt alt.data_transformers.enable("vegafusion") alt.Chart(embedding).mark_circle(opacity=1).encode( x=alt.X("embedding_0"), y=alt.Y("embedding_1") ).properties(width=400, height=400)

Out[3]:

In [4]:

from distortions.visualization import dplot
from distortions.geometry import neighborhoods

plots = {}
N = neighborhoods(adata, threshold=.2, outlier_factor=3, embed_key="X_tsne", frame=[100, 100], method="window")
plots["hydra_link_80"] = dplot(embedding, width=440, height=440)\
   .mapping(x="embedding_0", y="embedding_1")\
   .inter_edge_link(N=N, threshold=3, stroke="#F25E7A", highlightColor="#F25E7A", backgroundOpacity=0.2, strokeWidth=0.2, opacity=0.6)\
   .geom_ellipse(radiusMax=10, radiusMin=.8)\
   .labs(x="t-SNE 1", y="t-SNE 2")
plots["hydra_link_80"]

from distortions.visualization import dplot from distortions.geometry import neighborhoods plots = {} N = neighborhoods(adata, threshold=.2, outlier_factor=3, embed_key="X_tsne", frame=[100, 100], method="window") plots["hydra_link_80"] = dplot(embedding, width=440, height=440)\ .mapping(x="embedding_0", y="embedding_1")\ .inter_edge_link(N=N, threshold=3, stroke="#F25E7A", highlightColor="#F25E7A", backgroundOpacity=0.2, strokeWidth=0.2, opacity=0.6)\ .geom_ellipse(radiusMax=10, radiusMin=.8)\ .labs(x="t-SNE 1", y="t-SNE 2") plots["hydra_link_80"]

Out[4]:

dplot(dataset=[{'embedding_0': -5.127078056335449, 'embedding_1': -6.697624206542969, 'x0': -0.833874391295024…

In [5]:

metrics = {k: H[k] / H.mean() for k in range(len(H))}
plots["hydra_isometry"] = dplot(embedding, width=440, height=440)\
    .mapping(x="embedding_0", y="embedding_1")\
    .inter_isometry(metrics=metrics, metrics_bw=.05, transformation_bw=.1, stroke="#dcdcdc")\
    .geom_ellipse(radiusMax=10, radiusMin=.8)\
    .labs(x="t-SNE 1", y="t-SNE 2")
plots["hydra_isometry"]

metrics = {k: H[k] / H.mean() for k in range(len(H))} plots["hydra_isometry"] = dplot(embedding, width=440, height=440)\ .mapping(x="embedding_0", y="embedding_1")\ .inter_isometry(metrics=metrics, metrics_bw=.05, transformation_bw=.1, stroke="#dcdcdc")\ .geom_ellipse(radiusMax=10, radiusMin=.8)\ .labs(x="t-SNE 1", y="t-SNE 2") plots["hydra_isometry"]

Out[5]:

dplot(dataset=[{'embedding_0': -5.127078056335449, 'embedding_1': -6.697624206542969, 'x0': -0.833874391295024…

In [6]:

from distortions.geometry import neighborhood_distances

dists = neighborhood_distances(adata, embed_key="X_tsne")
plots["hydra_boxplot"] = dplot(embedding, width=550, height=440)\
    .mapping(x="embedding_0", y="embedding_1")\
    .geom_ellipse(radiusMax=8, radiusMin=.5)\
    .inter_boxplot(dists=dists, strokeWidth=0.2)\
    .labs(x = "t-SNE 1", y = "t-SNE 2")
plots["hydra_boxplot"]

from distortions.geometry import neighborhood_distances dists = neighborhood_distances(adata, embed_key="X_tsne") plots["hydra_boxplot"] = dplot(embedding, width=550, height=440)\ .mapping(x="embedding_0", y="embedding_1")\ .geom_ellipse(radiusMax=8, radiusMin=.5)\ .inter_boxplot(dists=dists, strokeWidth=0.2)\ .labs(x = "t-SNE 1", y = "t-SNE 2") plots["hydra_boxplot"]

Out[6]:

dplot(dataset=[{'embedding_0': -5.127078056335449, 'embedding_1': -6.697624206542969, 'x0': -0.833874391295024…

In [7]:

#[p.save(f"../paper/figures/{k}.svg") for k, p in plots.items()]

#[p.save(f"../paper/figures/{k}.svg") for k, p in plots.items()]

In [8]:

[display(p) for p in plots.values()]

[display(p) for p in plots.values()]

dplot(dataset=[{'embedding_0': -5.127078056335449, 'embedding_1': -6.697624206542969, 'x0': -0.833874391295024…

dplot(dataset=[{'embedding_0': -5.127078056335449, 'embedding_1': -6.697624206542969, 'x0': -0.833874391295024…

dplot(dataset=[{'embedding_0': -5.127078056335449, 'embedding_1': -6.697624206542969, 'x0': -0.833874391295024…

Out[8]:

[None, None, None]

In [ ]: