Welcome to sumo’s documentation!¶
Contents:
sumo is a command-line tool to identify molecular subtypes in multi-omics datasets. It implements a novel nonnegative matrix factorization (NMF) algorithm to identify groups of samples that share molecular signatures, and provides additional modules to evaluate such assignments and identify features that drive the classification.
To see how sumo performs the joint factorization of patient-similarity networks and through integration of multi-omic data identifies significantly different molecular subtypes of LGG read our publication in CellReportsMethods:
Sienkiewicz, K., Chen, J., Chatrath, A., Lawson, J. T., Sheffield, N. C., Zhang, L., & Ratan, A. (2022). Detecting molecular subtypes from multi-omics datasets using SUMO. In Cell Reports Methods (Vol. 2, Issue 1, p. 100152). Elsevier BV. https://doi.org/10.1016/j.crmeth.2021.100152
For practical details about sumo analysis pipeline, examples of downstream analysis and troubleshooting please refer to our published STAR Protocol and/or package documentation available at https://python-sumo.readthedocs.io.
Installation¶
You can install sumo from PyPI, by executing command below. Please note that we require python 3.6+.
pip install python-sumo
(March 2021): We have noted an installation issue with the llvmlite package (required for one of sumo dependencies). To avoid errors with installation, upgrade pip to a >19.0 version.
Usage¶
sumo consists of four subroutines. A typical workflow includes running prepare mode for preparation of similarity matrices from feature matrices, followed by factorization of produced multiplex network (mode run). Third mode evaluate can be used for comparison of created cluster labels against biologically significant labels. A fourth mode interpret can be used to detect the importance of each feature in driving the classification.
(February 2022) As of SUMO v0.3, a semi-supervised classification of samples is now supported. This allows the inclusion of “a priori” knowledge about labels of fraction of samples to improve the factorization results. The supervised version of solver is automatically enabled in sumo run, when the ‘-labels’ parameter is used.
prepare¶
Generates similarity matrices for samples based on biological data and saves them into multiplex network files.
usage: sumo prepare [-h] [-method METHOD] [-k K] [-alpha ALPHA] [-missing MISSING]
[-atol ATOL] [-sn SN] [-fn FN] [-df DF] [-ds DS] [-logfile LOGFILE]
[-log {DEBUG,INFO,WARNING}] [-plot PLOT]
infile1,infile2,... outfile.npz
positional arguments:
infile1,infile2,... comma-delimited list of paths to input files, containing
standardized feature matrices, with samples in columns and
features in rows (supported types of files: ['.txt', '.txt.gz',
'.txt.bz2', '.tsv', '.tsv.gz', '.tsv.bz2'])
outfile.npz path to output .npz file
optional arguments:
-h, --help show this help message and exit
-method METHOD either one method of sample-sample similarity calculation, or
comma-separated list of methods for every layer (available
methods: ['euclidean', 'cosine', 'pearson', 'spearman'], default
of euclidean)
-k K fraction of nearest neighbours to use for sample similarity
calculation using Euclidean distance similarity (default of 0.1)
-alpha ALPHA hypherparameter of RBF similarity kernel, for Euclidean distance
similarity (default of 0.5)
-missing MISSING acceptable fraction of available values for assessment of
distance/similarity between pairs of samples - either one value
or comma-delimited list for every layer (default of [0.1])
-atol ATOL if input files have continuous values, sumo checks if data is
standardized feature-wise, meaning all features should have mean
close to zero, with standard deviation around one; use this
parameter to set tolerance of standardization checks (default of
0.01)
-sn SN index of row with sample names for input files (default of 0)
-fn FN index of column with feature names for input files (default of 0)
-df DF if percentage of missing values for feature exceeds this value,
remove feature (default of 0.1)
-ds DS if percentage of missing values for sample (that remains after
feature dropping) exceeds this value, remove sample (default of
0.1)
-logfile LOGFILE path to save log file, by default stdout is used
-log {DEBUG,INFO,WARNING}
sets the logging level (default of INFO)
-plot PLOT path to save adjacency matrix heatmap(s), by default plots are
displayed on screen
Example
sumo prepare -plot plot.png methylation.txt,expression.txt prepared.data.npz
run¶
Cluster multiplex network using non-negative matrix tri-factorization to identify molecular subtypes.
usage: sumo run [-h] [-sparsity SPARSITY] [-labels labels.tsv] [-n N]
[-method {max_value,spectral}] [-max_iter MAX_ITER] [-tol TOL]
[-subsample SUBSAMPLE] [-calc_cost CALC_COST] [-logfile LOGFILE]
[-log {DEBUG,INFO,WARNING}] [-h_init H_INIT] [-t T] [-rep REP]
[-seed SEED]
infile.npz k outdir
positional arguments:
infile.npz input .npz file containing adjacency matrices for every network
layer and sample names (file created by running program with mode
"run") - consecutive adjacency arrays in file are indexed in
following way: "0", "1" ... and index of sample name vector is
"samples"
k either one value describing number of clusters or coma-delimited
range of values to check (sumo will suggest cluster structure
based on cophenetic correlation coefficient)
outdir path to save output files
optional arguments:
-h, --help show this help message and exit
-sparsity SPARSITY either one value or coma-delimited list of sparsity penalty
values for H matrix (sumo will try different values and select
the best results; default of [0.1])
-labels labels.tsv optional path to .tsv file containing some of known sample labels
to be included as prior knowledge during the factorization
(inclusion of this parameter enables the 'supervised' mode of
sumo), the file should contain sample names in 'sample' and labels
in 'label' column
-n N number of repetitions (default of 60)
-method {max_value,spectral}
method of cluster extraction (default of "max_value")
-max_iter MAX_ITER maximum number of iterations for factorization (default of 500)
-tol TOL if objective cost function value fluctuation (|Δℒ|) is smaller
than this value, stop iterations before reaching max_iter
(default of 1e-05)
-subsample SUBSAMPLE fraction of samples randomly removed from each run, cannot be
greater then 0.5 (default of 0.05)
-calc_cost CALC_COST number of steps between every calculation of objective cost
function (default of 20)
-logfile LOGFILE path to save log file (by default printed to stdout)
-log {DEBUG,INFO,WARNING}
set the logging level (default of INFO)
-h_init H_INIT index of adjacency matrix to use for H matrix initialization (by
default using average adjacency), only for unsupervised
classification (when no "-labels" are set)
-t T number of threads (default of 1)
-rep REP number of times consensus matrix is created for the purpose of
assessing clustering quality (default of 5)
-seed SEED random state (none by default)
Example
sumo run -t 8 prepared.data.npz 2,5 results_dir
evaluate¶
Evaluate clustering results, given set of labels.
usage: sumo evaluate [-h] [-metric {NMI,purity,ARI}] [-logfile LOGFILE]
[-log {DEBUG,INFO,WARNING}]
infile.tsv labels
positional arguments:
infile.tsv input .tsv file containing sample names in 'sample' and
clustering labels in 'label' column (clusters.tsv file created by
running sumo with mode 'run')
labels .tsv of the same structure as input file
optional arguments:
-h, --help show this help message and exit
-metric {NMI,purity,ARI}
metric for accuracy evaluation (by default all metrics are
calculated)
-logfile LOGFILE path to save log file (by default printed to stdout)
-log {DEBUG,INFO,WARNING}
sets the logging level (default of INFO)
Example
sumo evaluate results_dir/k3/clusters.tsv labels.tsv
interpret¶
Find features that support clusters separation.
usage: sumo interpret [-h] [-logfile LOGFILE] [-log {DEBUG,INFO,WARNING}] [-hits HITS]
[-max_iter MAX_ITER] [-n_folds N_FOLDS] [-t T] [-seed SEED]
[-sn SN] [-fn FN] [-df DF] [-ds DS]
sumo_results.npz infile1,infile2,... output_prefix
positional arguments:
sumo_results.npz path to sumo_results.npz (created by running program with mode
"run")
infile1,infile2,... comma-delimited list of paths to input files, containing
standardized feature matrices, with samples in columns and
features in rows(supported types of files: ['.txt', '.txt.gz',
'.txt.bz2', '.tsv', '.tsv.gz', '.tsv.bz2'])
output_prefix prefix of output files - sumo will create two output files (1)
.tsv file containing matrix (features x clusters), where the
value in each cell is the importance of the feature in that
cluster; (2) .hits.tsv file containing features of most
importance
optional arguments:
-h, --help show this help message and exit
-logfile LOGFILE path to save log file (by default printed to stdout)
-log {DEBUG,INFO,WARNING}
sets the logging level (default of INFO)
-hits HITS sets number of most important features for every cluster, that
are logged in .hits.tsv file
-max_iter MAX_ITER maximum number of iterations, while searching through
hyperparameter space
-n_folds N_FOLDS number of folds for model cross validation (default of 5)
-t T number of threads (default of 1)
-seed SEED random state (default of 1)
-sn SN index of row with sample names for input files (default of 0)
-fn FN index of column with feature names for input files (default of 0)
-df DF if percentage of missing values for feature exceeds this value,
remove feature (default of 0.1)
-ds DS if percentage of missing values for sample (that remains after
feature dropping) exceeds this value, remove sample (default of
0.1)
Example
sumo interpret -t 8 results_dir/k3/sumo_results.npz methylation.txt,expression.txt interpret_results