rnalysis.filtering.CountFilter.split_kmedoids

CountFilter.split_kmedoids(n_clusters: Union[PositiveInt, List[PositiveInt], Literal['gap', 'silhouette', 'calinski_harabasz', 'davies_bouldin', 'bic']], n_init: PositiveInt = 3, max_iter: PositiveInt = 300, random_seed: Optional[NonNegativeInt] = None, metric: Union[str, Literal['Euclidean', 'Cosine', 'Pearson', 'Spearman', 'Manhattan', 'L1', 'L2', 'Jackknife', 'YS1', 'YR1', 'Sharpened_Cosine', 'Hamming']] = 'Euclidean', power_transform: bool = True, plot_style: Literal['all', 'std_area', 'std_bar'] = 'all', split_plots: bool = False, max_n_clusters_estimate: Union[PositiveInt, Literal['auto']] = 'auto', parallel_backend: Literal['multiprocessing', 'loky', 'threading', 'sequential'] = 'loky', gui_mode: bool = False) → Union[Tuple[CountFilter, ...], Tuple[Tuple[CountFilter, ...], ...]]

Clusters the features in the CountFilter object using the K-medoids clustering algorithm, and then splits those features into multiple non-overlapping CountFilter objects, based on the clustering result.

Parameters

  • n_clusters (int, list of ints, 'gap', 'silhouette', 'calinski_harabasz', 'davies_bouldin', or 'bic') – The number of clusters the algorithm will seek.

  • random_seed (Union[int, None] or None (default=None)) – determines random number generation for centroid initialization. Use an int to make the randomness deterministic.

  • n_init (int (default=3)) – number of time the k-medoids algorithm will be run with different medoid seeds. The final results will be the best output of n_init consecutive runs in terms of inertia.

  • max_iter (int (default=300)) – maximum number of iterations of the k-medoids algorithm for a single run.

  • metric (str (default='Euclidean')) – the distance metric used to determine similarity between data points. For a full list of supported distance metrics see the user guide.

  • power_transform (bool (default=True)) – if True, RNAlysis will apply a power transform (Box-Cox) to the data prior to clustering.

  • plot_style ('all', 'std_area', or 'std_bar' (default='all')) – determines the visual style of the cluster expression plot.

  • split_plots (bool (default=False)) – if True, each discovered cluster will be plotted on its own. Otherwise, all clusters will be plotted in the same Figure.

  • max_n_clusters_estimate (int or 'auto' (default='auto')) – the maximum number of clusters to test if trying to automatically estimate the optimal number of clusters. If `max_n_clusters_estimate`=’default’, an appropriate value will be picked automatically.

  • parallel_backend (Literal[PARALLEL_BACKENDS] (default='loky')) – Determines the babckend used to run the analysis. if parallel_backend not ‘sequential’, will calculate the statistical tests using parallel processing. In most cases parallel processing will lead to shorter computation time, but does not affect the results of the analysis otherwise.

Returns

if n_clusters is an int, returns a tuple of n_clusters CountFilter objects, each corresponding to a discovered cluster. If n_clusters is a list, returns one tuple of CountFilter objects per value in n_clusters.

Examples
>>> from rnalysis import filtering
>>> dev_stages = filtering.CountFilter('tests/test_files/elegans_developmental_stages.tsv')
>>> dev_stages.filter_low_reads(100)
Filtered 44072 features, leaving 2326 of the original 46398 features. Filtered inplace.
>>> clusters = dev_stages.split_kmedoids(n_clusters=14, metric='spearman', power_transform=True)
Filtered 1967 features, leaving 359 of the original 2326 features. Filtering result saved to new object.
Filtered 2020 features, leaving 306 of the original 2326 features. Filtering result saved to new object.
Filtered 2071 features, leaving 255 of the original 2326 features. Filtering result saved to new object.
Filtered 2131 features, leaving 195 of the original 2326 features. Filtering result saved to new object.
Filtered 2145 features, leaving 181 of the original 2326 features. Filtering result saved to new object.
Filtered 2157 features, leaving 169 of the original 2326 features. Filtering result saved to new object.
Filtered 2159 features, leaving 167 of the original 2326 features. Filtering result saved to new object.
Filtered 2182 features, leaving 144 of the original 2326 features. Filtering result saved to new object.
Filtered 2190 features, leaving 136 of the original 2326 features. Filtering result saved to new object.
Filtered 2192 features, leaving 134 of the original 2326 features. Filtering result saved to new object.
Filtered 2229 features, leaving 97 of the original 2326 features. Filtering result saved to new object.
Filtered 2252 features, leaving 74 of the original 2326 features. Filtering result saved to new object.
Filtered 2268 features, leaving 58 of the original 2326 features. Filtering result saved to new object.
Filtered 2275 features, leaving 51 of the original 2326 features. Filtering result saved to new object.
_images/kmedoids_all.png

Example plot of split_kmedoids(plot_style=’all’)

_images/clustering_PCA_kmedoids.png

Example plot of split_kmedoids()