jdti 0.1.0

No project description provided

JDtI – Python library for scRNAseq/RNAseq data analysis

Author: Jakub Kubiś

Institute of Bioorganic Chemistry
Polish Academy of Sciences
Laboratory of Single Cell Analyses

Description

JDtI (JDataIntegration) is a Python library for data integration and advanced post-processing of single-cell datasets.

JDtI enables basic quality control steps such as control of cells per cluster, number of genes per cell, and more advanced tasks like subclustering, integration, and wide visualization. In this approach, we do not drop the cell information during separate set analyses; instead, we use previous cluster cell lineage information for integrating data based on cluster markers and data harmonization. After integration, it is possible to visualize cell interactions and correlations in many ways, including cell distance, correlations, and more.

Despite this, it is also able to conduct DEG analysis between sets, selected cells, or grouped cells, and visualize the results on UMAP, volcano plots, and regression plots comparing pairs of cells. It is very powerful for more advanced analyses focusing on specific issues within the data that may not be discovered in basic analyses.

Additionally, JDtI offers many functions for data visualization and processing within clean visual outputs, such as volcano plots, gene expression analysis of different data types, clustering, heatmaps, and more.

It is compatible with various sequencing approaches, including scRNA-seq and bulk RNA-seq, and supports interoperability with tools such as Seurat, Scanpy, and other bioinformatics frameworks using the 10x sparse matrix format as input. More details about the available functions can be found in the Documentation and Example Usage section on GitHub.

Table of contents

Installation Documenation

Example usage: 1. Basic functions 2. Data clustering 3. Data integration 4. Data subclustering

Installation

pip install jdti

Documentation

Documentation for classes and functions is available here 👉 Documentation 📄

Example usage

1. Basic functions

1. Loading functions

from jdti import *

2. Loading data

# load sparse matrix as pd.DataFrame data with creating metadata
data, metadata = load_sparse(path = 'data/set1', name = 'set1')

#load data frame from different data type (.tsv, .txt, .tsv)
data = pd.read_csv('example_data.csv')

fl = find_features(data, features =['KIT', 'MC1', 'EDNRB', 'PAX3'])

fl2 = find_features(data, features =['KIT', 'MC1R', 'EDNRB', 'PAX3'])

nam = find_names(data, names = ['0', '1', '2','10', '1&'])

data_reduced = reduce_data(data,
                features = fl2['included'],
                names = nam['included'])

DEG = calc_DEG(data_reduced, 
             metadata_list  = None, 
             entities = compare_dict, 
             sets = None, 
             min_exp = 0, 
             min_pct = 0.1, 
             n_proc =10)

DEG2 = calc_DEG(data, 
             metadata_list = metadata['sets'], 
             entities = compare_dict, 
             sets = None, 
             min_exp = 0, 
             min_pct = 0.1, 
             n_proc = 10)

fig = volcano_plot(DEG3, 
                 p_adj = True, 
                 top = 25, 
                 p_val = 0.05, 
                 lfc = 0.25, 
                 standard_scale = False, 
                 rescale_adj = True, 
                 image_width = 12, 
                 image_high = 12)

DEG3_10 = DEG3.sort_values(['p_val', 'esm', 'log(FC)'], ascending=[True, False, False]).head(10)

data_reduced = reduce_data(data,
                features = list(set(DEG3_10['feature'])),
                names = nam['included'])

avg = average(data_reduced)
occ = occurrence(data_reduced)

fig = features_scatter(expression_data = avg, 
                     occurence_data = occ,
                     features = None, 
                     metadata_list = None, 
                     colors = 'viridis', 
                     hclust = 'complete', 
                     img_width = 8, 
                     img_high = 5, 
                     label_size = 10, 
                     size_scale = 100,
                     x_lab = 'Genes', 
                     legend_lab = 'log(CPM + 1)',
                     bbox_to_anchor_scale = 25,
                     bbox_to_anchor_perc=(0.91, 0.55),
                     bbox_to_anchor_group=(1.01, 0.4))

fig = development_clust(data = avg, 
                      method = 'ward',
                      img_width = 5,
                      img_high = 5)

2. Data clustering

from jdti import Clustering, load_sparse

data, metadata = load_sparse(path = 'data/set2', name = 'set2')
clusters = Clustering.add_data_frame(data, metadata)

clusters.clustering_data
clusters.clustering_metadata

clusters.perform_PCA(pc_num=100, width=8, height=6)

clusters.knee_plot_PCA(width=8, height=6)

clusters.harmonize_sets(harmonize_type='harmony')

clusters.find_clusters_PCA(pc_num=0, eps=0.5, min_samples=10, width=8, height=6, harmonized=False)

clusters.perform_UMAP(factorize=False, umap_num=0, pc_num=5, harmonized=False)


clusters.knee_plot_umap(eps=0.5, min_samples=10)

clusters.find_clusters_UMAP(umap_n=5, eps=0.5, min_samples=10, width=8, height=6)


clusters.UMAP_vis(names_slot='cell_names', set_sep=True, point_size=0.6)

clusters.UMAP_feature(feature_name = 'KIT', features_data=None, point_size=0.6)

clusters.get_umap_data()

clusters.get_pca_data()

clusters.return_clusters(clusters='umap')

3. Data integration

from jdti import COMPsc, volcano_plot

jseq_object = COMPsc.project_dir('data', ['set1', 'set2'])

jseq_object.load_sparse_from_projects(normalized_data=True)

dt = jseq_object.get_partial_data(names=['10'], features=['KIT', 'PAX3', 'MITF'], name_slot='cell_names')

jseq_object.gene_histograme(bins=100)

jseq_object.gene_threshold(min_n = 50, max_n = 3000)

jseq_object.gene_histograme(bins=100)

jseq_object.reduce(reg = '5', inc_set = False)

jseq_object.gene_histograme(bins=100)

jseq_object.cell_histograme(name_slot = 'cell_names')

jseq_object.cluster_threshold(min_n = 20, name_slot = 'cell_names')

jseq_object.cell_histograme(name_slot = 'cell_names')

# returny

met = jseq_object.input_metadata

data = jseq_object.get_data(set_info=True) 

metadata = jseq_object.get_metadata()

jseq_object.calculate_difference_markers(min_exp = 0, 
                                         min_pct = 0.25, 
                                         n_proc=10, 
                                         force = False)



jseq_object.estimating_similarity(method = 'pearson', 
                                  p_val = 0.05,
                                  top_n = 10)
    

pl = jseq_object.similarity_plot(split_sets = True, 
                                 set_info = True,
                                 cmap='seismic', 
                                 width = 16, height = 14)

   
# pl.savefig(f'sim_plot_top_{top}.svg', dpi=300, bbox_inches='tight')

pl2 = jseq_object.spatial_similarity(set_info= True, bandwidth = 1, n_neighbors = 5,
min_dist = 0.1, legend_split = 2, point_size = 20, spread=1.0,
set_op_mix_ratio=1.0,
local_connectivity=1,
repulsion_strength=1.0,
negative_sample_rate=5,
width = 12, height = 10)

pl2.savefig(f'sim_plot_map_top_{top}.svg', dpi=300, bbox_inches='tight')

sim_data = jseq_object.similarity sim_data = sim_data[sim_data['set1'] != sim_data['set2']]

jseq_object.cell_regression( cell_x = '2', cell_y = '6', set_x = 'set1', set_y = 'set2', threshold = 6, image_width = 12, image_high = 7, color = 'black')

jseq_object.clustering_features(name_slot = 'cell_names', features_list = None, p_val = 0.05, top_n = 10, adj_mean = False, beta = 0.2)

jseq_object.perform_PCA(pc_num = 50)

jseq_object.knee_plot_PCA()

jseq_object.harmonize_sets(harmonize_type = 'harmony')

# jseq_object.find_clusters_PCA(pc_num = 100, eps = 0.5, min_samples = 10)

jseq_object.perform_UMAP(factorize=False, umap_num = 2, pc_num = 10, harmonized = True)


# jseq_object.knee_plot_umap(eps = 0.5, min_samples = 10)


# jseq_object.find_clusters_UMAP(umap_n = 6, eps = 1, min_samples = 20)


plu = jseq_object.UMAP_vis( 
             names_slot = 'cell_names', 
             set_sep = True,
             point_size = 1,
             font_size = 6,
             legend_split_col = 2,
             width = 8,
             height = 6,
             inc_num = True)

# plu.savefig(f'sim_umap_top.svg', dpi=300, bbox_inches='tight')


plu = jseq_object.UMAP_vis( 
             names_slot = 'sets', 
             set_sep = True,
             point_size = 1,
             font_size = 6,
             legend_split_col = 1,
             width = 8,
             height = 6,
             inc_num = False)

# plu.savefig(f'sim_umap_sets_top_.svg', dpi=300, bbox_inches='tight')

vis = jseq_object.UMAP_feature( 
             features_data = jseq_object.get_data(set_info = False) ,
             feature_name = 'MAP1B',
             point_size = 0.6,
             font_size = 6,
             width = 8,
             height = 6,
             palette = 'light')

# vis.savefig(f'sim_umap_sets_top_vis.svg', dpi=300, bbox_inches='tight')

jseq_object.var_data


# jseq_object.save_project(name = 'topola')

stats = jseq_object.statistic(cells=None, sets='All', min_exp=0, min_pct=0.025, n_proc=10)
stats_5 = stats.sort_values(['valid_group', 'esm', 'log(FC)'], ascending=[True, False, False]).groupby('valid_group').head(5)



fig = volcano_plot(stats)

jseq_object.scatter_plot(
                 names = None,
                 features = list(set(stats_5['feature'])),
                 name_slot = 'cell_names',
                 scale = False,
                 colors = 'viridis', 
                 hclust = 'complete', 
                 img_width  = 15, 
                 img_high  = 3, 
                 label_size = 10, 
                 size_scale = 200,
                 x_lab = 'Genes', 
                 legend_lab = 'log(CPM + 1)',
                 set_box_size = 5,
                 set_box_high = 0.1,
                 bbox_to_anchor_scale = 25,
                 bbox_to_anchor_perc=(0.90, 0.5),
                 bbox_to_anchor_group=(0.9, 0.3))

import re

jseq_object.data_composition( 
                     features_count = list(set([re.sub(r' .*$', '',x) for x in list(set(jseq_object.input_metadata['cell_names']))])),
                     name_slot = 'cell_names',
                     set_sep = True
                     )


jseq_object.composition_pie( 
                    width = 6, 
                    height = 6, 
                    font_size = 15,
                    cmap  = "tab20",
                    legend_split_col = 1,
                    offset_labels = 0.5,
                    legend_bbox = (1.15, 0.95))


jseq_object.bar_composition( 
                    cmap = 'tab20b', 
                    width = 2, 
                    height = 6, 
                    font_size = 15,
                    legend_split_col = 1,
                    legend_bbox = (1.3, 1))

4. Data subclustering

from jdti import COMPsc

jseq_object = COMPsc.project_dir('data', ['set2'])

jseq_object.load_sparse_from_projects(normalized_data=True)

jseq_object.subcluster_prepare(features = ['HMGCS1', 'MAP1B', 'SOX4'], 
                               cluster='10')

jseq_object.define_subclusters( 
                          umap_num = 5,
                          eps = 1, 
                          min_samples = 5,
                          n_neighbors = 5,  
                          min_dist = 0.1, 
                          spread = 1.0,              
                          set_op_mix_ratio = 1.0,    
                          local_connectivity = 1,    
                          repulsion_strength = 1.0,  
                          negative_sample_rate = 5,  
                          width = 8, 
                          height = 6)
  

jseq_object.subcluster_features_scatter(
                                        colors = 'viridis', 
                                        hclust = 'complete', 
                                        img_width = 3, 
                                        img_high = 5, 
                                        label_size = 6, 
                                        size_scale = 70,
                                        x_lab = 'Genes', 
                                        legend_lab = 'normalized')
    

mapping = {
    "old_name": ["-1", "1", "4"],
    "new_name": ["1", "1", "1"]
}

jseq_object.rename_subclusters(mapping)

jseq_object.subcluster_DEG_scatter(
                                    top_n = 3,
                                    min_exp = 0, 
                                    min_pct = 0.1, 
                                    p_val = 0.05,
                                    colors = 'viridis', 
                                    hclust = 'complete', 
                                    img_width = 3, 
                                    img_high = 5, 
                                    label_size = 6, 
                                    size_scale = 70,
                                    x_lab = 'Genes', 
                                    legend_lab = 'normalized',
                                    n_proc=10)
    
 

jseq_object.accept_subclusters()

l = set(jseq_object.input_metadata['cell_names'])
        

Have fun JBS