Proteomics, Post-translational Modifications, andIntegrative Analyses Reveal MolecularHeterogeneity within Medulloblastoma Subgroups¶
SIMILARITY NETWORK FUSION ANALYSIS¶
Abstract¶
There is a pressing need to identify therapeutic targets in tumors with low mutation rates such as the malignant pediatric brain tumor medulloblastoma. To address this challenge, we quantitatively profiled global proteomes and phospho-proteomes of 45 medulloblastoma samples. Integrated analyses revealed that tumors with similar RNA expression vary extensively at the post-transcriptional and post-translational levels. We identified distinct pathways associated with two subsets of SHH tumors, and found post-translational modifications of MYC that are associated with poor outcomes in Group 3 tumors. We found kinases associated with subtypes and showed that inhibiting PRKDC sensitizes MYC-driven cells to radiation. Our study shows that proteomics enables a more comprehensive, functional readout, providing a foundation for future therapeutic strategies.
|experimental\_design| Figure 1: Summary of data types included in this study, depth of proteomic data types, and cohort composition.
Consensus clustering, principal component analysis (PCA), and t-distributed stochastic neighbor embedding (t-SNE) carried out separately on each data type predominantly revealed the known subgroups: Group 3, Group 4, and SHH (Figures (Figures2A,2A, S1-S3). The proteomic data also identified very stable subsets of two known subgroups. Here, we refer to Group 3 clusters as Group 3a (G3a) and Group 3b (G3b), and SHH clusters as SHHa and SHHb.
Figure 2: Comparison of clustering results.
Objectives¶
Use CKG’s analytics core to:
Re-analyze the proteomics dataset:
Preprocessing
Stratification looking at covariates: Age, Gender, Metastatic status and histology
Differential regulation including covariates
Functional enrichment
Identify patients subgroups
Integration of all omics datasets using Similarity Network Fusion (REF: https://www.nature.com/articles/nmeth.2810).
Idenfity clinical and molecular differences between the identifierd subgroups
Investigate correlating PTMs on relevant protein complexes
Loading the Data¶
In this analysis we use:
Clinical metadata
RNA-sequencing
Proteomics
Phosphoproteomics
Phosphotyrosine proteomics
Acetylomics
This datasets have been previously formated (samples x proteins) and standardized to UniProt identifiers (data available in /assets) (see notebook Meduloblastoma Data Processing.ipynb).
[5]:
import pandas as pd
import numpy as np
from ckg.analytics_core.analytics import analytics
from ckg.analytics_core.viz import viz
from ckg.graphdb_builder import mapping
from ckg.graphdb_connector import connector
from plotly.offline import init_notebook_mode, iplot
%matplotlib inline
init_notebook_mode(connected=True)
[6]:
clinical_data = pd.read_excel('../../assets/samples.xlsx', header=0)
rnaseq_data = pd.read_csv("../../assets/rnaseq.tsv", sep='\t', header=0)
proteomics_data = pd.read_csv("../../assets/proteomics.tsv", sep='\t', header=0)
phosphoSTY_data = pd.read_csv("../../assets/phosphoSTY.tsv", sep='\t', header=0)
phosphoY_data = pd.read_csv("../../assets/phosphoY.tsv", sep='\t', header=0)
acetylomics_data = pd.read_csv("../../assets/acetylomics.tsv", sep='\t', header=0)
2. Identify patients subgroups¶
Preprocessing:¶
Remove WNT samples (low number of samples)
Filter proteins/PTMs/transcipts with more than 80% missing values in all groups
Normalize data using median-centered normalization
Impute missing values using a mixed model: K-Nearest Neighbors algorithm (KNN) if at least 60% valid values, otherwise Minimum Probability (MinProb) drawing random values from a down-shifted Gaussian distribution
Include Protein name in columns using CKG’s mapping functionality
[7]:
proteomics_data = proteomics_data[proteomics_data['group'] != 'WNT']
phosphoSTY_data = phosphoSTY_data[phosphoSTY_data['group'] != 'WNT']
phosphoY_data = phosphoY_data[phosphoY_data['group'] != 'WNT']
acetylomics_data = acetylomics_data[acetylomics_data['group'] != 'WNT']
rnaseq_data = rnaseq_data[rnaseq_data['group'] != 'WNT']
[8]:
valid_proteins = analytics.extract_percentage_missing(proteomics_data, missing_max=0.2, drop_cols=['index'], group='group', how='all')
proteomics_data = proteomics_data[valid_proteins + ['index', 'group']]
valid_sites = analytics.extract_percentage_missing(phosphoSTY_data, missing_max=0.2, drop_cols=['index'], group='group', how='all')
phosphoSTY_data = phosphoSTY_data[valid_sites + ['index', 'group']]
valid_sites = analytics.extract_percentage_missing(phosphoY_data, missing_max=0.2, drop_cols=['index'], group='group', how='all')
phosphoY_data = phosphoY_data[valid_sites + ['index', 'group']]
valid_sites = analytics.extract_percentage_missing(acetylomics_data, missing_max=0.2, drop_cols=['index'], group='group', how='all')
acetylomics_data = acetylomics_data[valid_sites + ['index', 'group']]
valid_transcripts = analytics.extract_percentage_missing(rnaseq_data, missing_max=0.2, drop_cols=['index'], group='group', how='all')
rnaseq_data = rnaseq_data[valid_transcripts + ['index', 'group']]
[9]:
proteomics_data = analytics.normalize_data(proteomics_data, method='median', normalize='samples')
phosphoSTY_data = analytics.normalize_data(phosphoSTY_data, method='median', normalize='samples')
phosphoY_data = analytics.normalize_data(phosphoY_data, method='median', normalize='samples')
acetylomics_data = analytics.normalize_data(acetylomics_data, method='median', normalize='samples')
rnaseq_data = analytics.normalize_data(rnaseq_data, method='median', normalize='samples')
[10]:
proteomics_data = analytics.imputation_mixed_norm_KNN(proteomics_data, index_cols=['index','group']).reset_index()
proteomics_data = proteomics_data.rename(columns={'index':'sample'})
phosphoSTY_data = analytics.imputation_mixed_norm_KNN(phosphoSTY_data, index_cols=['index','group']).reset_index()
phosphoSTY_data = phosphoSTY_data.rename(columns={'index':'sample'})
phosphoY_data = analytics.imputation_mixed_norm_KNN(phosphoY_data, index_cols=['index','group']).reset_index()
phosphoY_data = phosphoY_data.rename(columns={'index':'sample'})
acetylomics_data = analytics.imputation_mixed_norm_KNN(acetylomics_data, index_cols=['index','group']).reset_index()
acetylomics_data = acetylomics_data.rename(columns={'index':'sample'})
rnaseq_data = analytics.imputation_mixed_norm_KNN(rnaseq_data, index_cols=['index','group']).reset_index()
rnaseq_data = rnaseq_data.rename(columns={'index':'sample'})
[11]:
m = mapping.getMappingFromDatabase(proteomics_data.columns, node="Protein", attribute_from='id', attribute_to='name')
columns = [m[c]+"~"+c if c in m and m[c] is not None else c for c in proteomics_data.columns]
proteomics_data.columns = columns
phospho_prots = [c.split('_')[0] for c in phosphoSTY_data.columns]
m = mapping.getMappingFromDatabase(phospho_prots, node="Protein", attribute_from='id', attribute_to='name')
columns = [m[c.split('_')[0]]+"_"+c.split('_')[1] if c.split('_')[0] in m and m[c.split('_')[0]] is not None else c for c in phosphoSTY_data.columns]
phosphoSTY_data.columns = columns
phospho_prots = [c.split('_')[0] for c in phosphoY_data.columns]
m = mapping.getMappingFromDatabase(phospho_prots, node="Protein", attribute_from='id', attribute_to='name')
columns = [m[c.split('_')[0]]+"_"+c.split('_')[1] if c.split('_')[0] in m and m[c.split('_')[0]] is not None else c for c in phosphoY_data.columns]
phosphoY_data.columns = columns
ace_prots = [c.split('_')[0] for c in acetylomics_data.columns]
m = mapping.getMappingFromDatabase(ace_prots, node="Protein", attribute_from='id', attribute_to='name')
columns = [m[c.split('_')[0]]+"_"+c.split('_')[1] if c.split('_')[0] in m and m[c.split('_')[0]] is not None else c for c in acetylomics_data.columns]
acetylomics_data.columns = columns
[12]:
proteomics_data.head()
[12]:
| sample | group | FLJ10292~A0A023T6R1 | RBM8~A0A023T787 | A0A024QYR3 | TM9SF2~A0A024QYR8 | A0A024QYS2 | AP1S1~A0A024QYT6 | EIF3S8~A0A024QYU9 | RABL5~A0A024QYV3 | ... | PMF1-BGLAP~U3KQ54 | CSNK1G1~U3KQB3 | LSM4~U3KQK1 | HEL-76~V9HW21 | MPND~W4VSR2 | YP_003024026 | YP_003024028 | YP_003024029 | YP_003024030 | YP_003024036 | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 0 | MB018 | GR3 | 0.305789 | 0.121693 | -0.306807 | -1.420907 | -1.992240 | 0.080430 | -0.586125 | -0.170322 | ... | 1.010434 | -0.398855 | 0.362923 | 2.222930 | 0.454971 | 0.169304 | -3.687196 | -1.779577 | 0.673982 | 1.286578 |
| 1 | MB095 | GR3 | 0.161486 | 0.004686 | -0.117270 | 0.385487 | 0.106731 | -0.577716 | 0.938022 | 0.397931 | ... | 0.440243 | 1.082378 | 0.149042 | -0.338782 | 1.171978 | 1.405934 | 0.320776 | 1.109756 | -0.313893 | 1.475623 |
| 2 | MB106 | GR3 | 0.246131 | 0.722363 | 0.252263 | 0.248175 | 0.037652 | 1.270131 | 0.865436 | 0.188902 | ... | 1.186331 | -0.450842 | -0.465150 | -2.932151 | -0.575521 | 0.571113 | -0.064543 | 0.368766 | -1.517764 | 1.302834 |
| 3 | MB170 | GR3 | 0.158128 | 0.141028 | -1.245953 | -1.109155 | -1.055955 | -1.152854 | 0.448824 | -0.016670 | ... | 0.942817 | -0.839358 | 0.255027 | 1.818705 | 0.072629 | -0.073669 | -0.537262 | -1.215553 | 0.564722 | -0.225667 |
| 4 | MB226 | GR3 | 0.829814 | 0.617831 | -0.223676 | -0.799670 | -0.551286 | -0.803952 | 0.647809 | -0.435659 | ... | 1.281616 | 0.088945 | -0.294337 | 0.016143 | -0.225818 | -0.465636 | -0.119610 | -1.510562 | 0.589995 | -0.559851 |
5 rows × 11137 columns
Smilarity Network Fusion¶
Integration of all omics data generated using Similarity Network Fusion (SNF):
SNF builds sample similarity networks using each omics data (RNAseq, Proteomics, PhosphoSTY, PhosphoY, Acetylomics) and fuses those networks into a single network that uses to identify clusters
We use these clusters to identify subgroups of patients
[14]:
df_dict = {'rnaseq': rnaseq_data, 'proteomics': proteomics_data, 'phosphoSTY': phosphoSTY_data, 'phosphoY': phosphoY_data, 'acetylomics': acetylomics_data}
feature_df, fussed_aff, fused_labels, silhouette_score = analytics.run_snf(df_dict, index=['sample', 'group'], num_clusters=None, distance_metric='euclidean', k_affinity=5, mu_affinity=0.5)