Spatial Autocorrelation#

This notebook benchmarks the spatial autocorrelation (Moran’s I and Geary’s C) for squidpy and rapids-singlecell.

import rapids_singlecell as rsc
import squidpy as sq
import cupy as cp

import rmm
from rmm.allocators.cupy import rmm_cupy_allocator

rmm.reinitialize(
    managed_memory=False,  # Allows oversubscription
    pool_allocator=False,  # default is False
    devices=0,  # GPU device IDs to register. By default registers only GPU 0.
)
cp.cuda.set_allocator(rmm_cupy_allocator)

adata = sq.datasets.visium_hne_adata()

We first need to compute a spatial graph with squidpy.gr.spatial_neighbors().

%%time
sq.gr.spatial_neighbors(adata)

CPU times: user 21.8 ms, sys: 642 µs, total: 22.5 ms
Wall time: 21.7 ms

genes = adata.var_names

Moran’s I#

GPU#

%%timeit
rsc.gr.spatial_autocorr(adata, mode="moran", genes=genes, n_perms=100, use_sparse=False)

420 ms ± 21 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

adata.uns["moranI"]

	I	pval_norm	var_norm	pval_z_sim	pval_sim	var_sim	pval_norm_fdr_bh	pval_z_sim_fdr_bh	pval_sim_fdr_bh
Nrgn	0.874753	0.000000	0.000131	0.000000	0.009901	0.000364	0.000000	0.000000	0.016677
Mbp	0.868723	0.000000	0.000131	0.000000	0.009901	0.000328	0.000000	0.000000	0.016677
Camk2n1	0.866545	0.000000	0.000131	0.000000	0.009901	0.000334	0.000000	0.000000	0.016677
Slc17a7	0.861761	0.000000	0.000131	0.000000	0.009901	0.000337	0.000000	0.000000	0.016677
Ttr	0.841988	0.000000	0.000131	0.000000	0.009901	0.000314	0.000000	0.000000	0.016677
...	...	...	...	...	...	...	...	...	...
Slc9a8	-0.026960	0.010027	0.000131	0.000190	0.009901	0.000060	0.017979	0.000362	0.016677
Pramef8	-0.028659	0.006681	0.000131	0.000217	0.009901	0.000064	0.012348	0.000411	0.016677
Klf12	-0.028764	0.006512	0.000131	0.000103	0.009901	0.000055	0.012052	0.000201	0.016677
Gart	-0.028859	0.006361	0.000131	0.000192	0.009901	0.000068	0.011784	0.000366	0.016677
Zbed4	-0.030420	0.004295	0.000131	0.000217	0.009901	0.000075	0.008154	0.000412	0.016677

18078 rows × 9 columns

CPU#

%%time
sq.gr.spatial_autocorr(adata, mode="moran", genes=genes, n_perms=100, n_jobs=32)

CPU times: user 4min 53s, sys: 2.32 s, total: 4min 55s
Wall time: 41.3 s

adata.uns["moranI"]

	I	pval_norm	var_norm	pval_z_sim	pval_sim	var_sim	pval_norm_fdr_bh	pval_z_sim_fdr_bh	pval_sim_fdr_bh
Nrgn	0.874753	0.000000	0.000131	0.000000	0.009901	0.000300	0.000000	0.000000	0.016675
Mbp	0.868723	0.000000	0.000131	0.000000	0.009901	0.000289	0.000000	0.000000	0.016675
Camk2n1	0.866542	0.000000	0.000131	0.000000	0.009901	0.000376	0.000000	0.000000	0.016675
Slc17a7	0.861761	0.000000	0.000131	0.000000	0.009901	0.000295	0.000000	0.000000	0.016675
Ttr	0.841986	0.000000	0.000131	0.000000	0.009901	0.000302	0.000000	0.000000	0.016675
...	...	...	...	...	...	...	...	...	...
Slc9a8	-0.026960	0.010026	0.000131	0.000061	0.009901	0.000052	0.017977	0.000121	0.016675
Pramef8	-0.028656	0.006687	0.000131	0.000135	0.009901	0.000057	0.012356	0.000261	0.016675
Klf12	-0.028760	0.006518	0.000131	0.000090	0.009901	0.000057	0.012061	0.000176	0.016675
Gart	-0.028857	0.006364	0.000131	0.000130	0.009901	0.000063	0.011789	0.000251	0.016675
Zbed4	-0.030415	0.004300	0.000131	0.000064	0.009901	0.000064	0.008163	0.000127	0.016675

18078 rows × 9 columns

Geary’s C#

GPU#

%%timeit
rsc.gr.spatial_autocorr(adata, mode="geary", genes=genes, n_perms=100, use_sparse=False)

411 ms ± 512 µs per loop (mean ± std. dev. of 7 runs, 1 loop each)

adata.uns["gearyC"]

	C	pval_norm	var_norm	pval_z_sim	pval_sim	var_sim	pval_norm_fdr_bh	pval_z_sim_fdr_bh	pval_sim_fdr_bh
Nrgn	0.126481	0.000000e+00	0.000131	0.000000	0.009901	0.000393	0.000000e+00	0.000000	0.016691
Mbp	0.132854	0.000000e+00	0.000131	0.000000	0.009901	0.000341	0.000000e+00	0.000000	0.016691
Camk2n1	0.135106	0.000000e+00	0.000131	0.000000	0.009901	0.000347	0.000000e+00	0.000000	0.016691
Slc17a7	0.138819	0.000000e+00	0.000131	0.000000	0.009901	0.000399	0.000000e+00	0.000000	0.016691
Ttr	0.156155	0.000000e+00	0.000131	0.000000	0.009901	0.000250	0.000000e+00	0.000000	0.016691
...	...	...	...	...	...	...	...	...	...
AC166344.1	1.079271	2.059686e-12	0.000131	0.403457	0.099010	0.000047	7.064124e-12	0.425065	0.133336
Gm14091	1.083541	1.372236e-13	0.000131	0.380466	0.188119	0.000083	4.932845e-13	0.411113	0.231301
Slc6a5	1.087126	1.265654e-14	0.000131	0.368844	0.376238	0.000058	4.722497e-14	0.403410	0.404281
Soat2	1.093646	1.110223e-16	0.000131	0.438600	0.297030	0.000129	4.432556e-16	0.449287	0.335355
Bmp8b	1.125921	0.000000e+00	0.000131	0.421862	0.059406	0.000086	0.000000e+00	0.436469	0.084689

18078 rows × 9 columns

CPU#

%%time
sq.gr.spatial_autocorr(adata, mode="geary", genes=genes, n_perms=100, n_jobs=32)

CPU times: user 5min, sys: 273 ms, total: 5min
Wall time: 37.5 s

adata.uns["gearyC"]

	C	pval_norm	var_norm	pval_z_sim	pval_sim	var_sim	pval_norm_fdr_bh	pval_z_sim_fdr_bh	pval_sim_fdr_bh
Nrgn	0.126481	0.000000e+00	0.000131	0.000000	0.009901	0.000282	0.000000e+00	0.000000	0.016764
Mbp	0.132855	0.000000e+00	0.000131	0.000000	0.009901	0.000315	0.000000e+00	0.000000	0.016764
Camk2n1	0.135106	0.000000e+00	0.000131	0.000000	0.009901	0.000269	0.000000e+00	0.000000	0.016764
Slc17a7	0.138820	0.000000e+00	0.000131	0.000000	0.009901	0.000317	0.000000e+00	0.000000	0.016764
Ttr	0.156156	0.000000e+00	0.000131	0.000000	0.009901	0.000312	0.000000e+00	0.000000	0.016764
...	...	...	...	...	...	...	...	...	...
AC166344.1	1.079269	2.062794e-12	0.000131	0.432510	0.504950	0.000027	7.074786e-12	0.443551	0.504950
Gm14091	1.083544	1.368905e-13	0.000131	0.408545	0.475248	0.000173	4.920872e-13	0.427605	0.492916
Slc6a5	1.087121	1.276756e-14	0.000131	0.379592	0.495050	0.000051	4.762939e-14	0.409957	0.498969
Soat2	1.093644	1.110223e-16	0.000131	0.425606	0.475248	0.000307	4.432626e-16	0.438888	0.492916
Bmp8b	1.125922	0.000000e+00	0.000131	0.412736	0.465347	0.000500	0.000000e+00	0.429831	0.489158

18078 rows × 9 columns

Spatial Autocorrelation

Contents

Spatial Autocorrelation#

Moran’s I#

GPU#

CPU#

Geary’s C#

GPU#

CPU#