calc_nmis

Normalized mutual information scores can be used to indirectly measure how
important a feature may have been in producing a cluster solution. This
function will calculate the normalized mutual information between cluster
solutions in a solutions data frame as well as cluster solutions created
by including only a single feature from a provided data list, but otherwise
using all the same hyperparameters as specified in the original SNF config.
Note that NMIs can be calculated between two cluster solutions regardless
of what features were actually used to create those cluster solutions. For
example, a feature that was not involved in producing a particular cluster
solution may still have a high NMI with that cluster solution (typically
because it was highly correlated with a different feature that was used).

Framework to facilitate patient subtyping with similarity network fusion and meta clustering. The similarity network fusion (SNF) algorithm was introduced by Wang et al. (2014) in <doi:10.1038/nmeth.2810>. SNF is a data integration approach that can transform high-dimensional and diverse data types into a single similarity network suitable for clustering with minimal loss of information from each initial data source. The meta clustering approach was introduced by Caruana et al. (2006) in <doi:10.1109/ICDM.2006.103>. Meta clustering involves generating a wide range of cluster solutions by adjusting clustering hyperparameters, then clustering the solutions themselves into a manageable number of qualitatively similar solutions, and finally characterizing representative solutions to find ones that are best for the user's specific context. This package provides a framework to easily transform multi-modal data into a wide range of similarity network fusion-derived cluster solutions as well as to visualize, characterize, and validate those solutions. Core package functionality includes easy customization of distance metrics, clustering algorithms, and SNF hyperparameters to generate diverse clustering solutions; calculation and plotting of associations between features, between patients, and between cluster solutions; and standard cluster validation approaches including resampled measures of cluster stability, standard metrics of cluster quality, and label propagation to evaluate generalizability in unseen data. Associated vignettes guide the user through using the package to identify patient subtypes while adhering to best practices for unsupervised learning.

Prashanth S Velayudhan

metasnf

Meta Clustering with Similarity Network Fusion

Xiaoqiao Xu

Prajkta Kallurkar

Ana Patricia Balbon

Maria T Secara

Adam Taback

Denise Sabac

Nicholas Chan

Shihao Ma

Bo Wang

Daniel Felsky

Stephanie H Ameis

Brian Cox

Colin Hawco

Lauren Erdman

Anne L Wheeler

calc_nmis function

<dl><dt>dl</dt>
<dd>A nested list of input data from <code>data_list()</code>.</dd>
<dt>sol_df</dt>
<dd>Result of <code>batch_snf</code> storing cluster solutions and
the settings that were used to generate them. Use the same value as was used
in the original call to <code>batch_snf()</code>.</dd>
<dt>transpose</dt>
<dd>If TRUE, will transpose the output data frame.</dd>
<dt>ignore_inclusions</dt>
<dd>If TRUE, will ignore the inclusion columns in the
solutions data frame and calculate NMIs for all features. If FALSE, will
give NAs for features that were dropped on a given settings_df row.</dd>
<dt>processes</dt>
<dd>Specify number of processes used to complete SNF iterations<ul>
<li><code>1</code> (default) Sequential processing: function will iterate through the
<code>settings_df</code> one row at a time with a for loop. This option will
not make use of multiple CPU cores, but will show a progress bar.</li>
<li><code>2</code> or higher: Parallel processing will use the
<code><a href="https://future.apply.futureverse.org/reference/future_apply.html">future.apply::future_apply</a></code> to distribute the SNF iterations across
the specified number of CPU cores. If higher than the number of
available cores, a warning will be raised and the maximum number of
cores will be used.</li>
<li><code>max</code>: All available cores will be used.</li>
</ul></dd></dl>

Arguments

Calculate feature NMIs for a data list and a solutions data frame — calc_nmis

<dl>

<dt>dl</dt>
<dd>A nested list of input data from <code>data_list()</code>.</dd>


<dt>sol_df</dt>
<dd>Result of <code>batch_snf</code> storing cluster solutions and
the settings that were used to generate them. Use the same value as was used
in the original call to <code>batch_snf()</code>.</dd>


<dt>transpose</dt>
<dd>If TRUE, will transpose the output data frame.</dd>


<dt>ignore_inclusions</dt>
<dd>If TRUE, will ignore the inclusion columns in the
solutions data frame and calculate NMIs for all features. If FALSE, will
give NAs for features that were dropped on a given settings_df row.</dd>


<dt>processes</dt>
<dd>Specify number of processes used to complete SNF iterations<ul>
<li><code>1</code> (default) Sequential processing: function will iterate through the
<code>settings_df</code> one row at a time with a for loop. This option will
not make use of multiple CPU cores, but will show a progress bar.</li>
<li><code>2</code> or higher: Parallel processing will use the
<code><a href='https://future.apply.futureverse.org/reference/future_apply.html'>future.apply::future_apply</a></code> to distribute the SNF iterations across
the specified number of CPU cores. If higher than the number of
available cores, a warning will be raised and the maximum number of
cores will be used.</li>
<li><code>max</code>: All available cores will be used.</li>
</ul></dd>

</dl>

Calculate feature NMIs for a data list and a solutions data frame

calc_nmis: Calculate feature NMIs for a data list and a solutions data frame

Description

Usage

Value

Arguments

Examples