General view
- 1: Add a new dataset
- 2: For a given dataset: details on method run on the dataset (run status, run id…)
- 3: For a given dataset: View leaderboard of computational methods apply on similar datasets
- 4: For a given dataset: View the results of the methods applied on the dataset
- 5: For a given dataset: Delete the dataset
Submit a dataset
- 1: Name your dataset
- 2: Specify your type of query: do you want to quantify all cell type or only immune cell types?
- 3: Give information of your dataset and upload it
View the leaderboard and run method on your dataset
- 1: In columns: benchmark dataset similar to your clinical dataset according to the descriptors you entered while submitting your dataset. You can retreive more detailed information on each datase by clicking on the html link
- 2: Computational methods (MT) that have been applied to quantify heterogeneity of these benchmark dataset on codabench. You can retreive more detailed information on each datase by clicking on the “i” button (including the R code of the algorithm)
- 3: Accurary score (mean, standart deviation and computation time) given by the codabench platform. You can sort the scores by clicking on the column names
- 4: You can choose which method you want to run on your dataset by clicking on this button
View results
- 1: Get more information on the method
- 2: Download all results, including algorithm code, estimated proportion matrix and estimated cell-type specific profiles. In this folder, you have the following files:
- the estimated proportion matrix in .csv and .rds format :
results_1.rds and results_A_1.csv
- the estimated reference profiles matrix in .csv and .rds format :
ref_1.rds and results_T_1.csv
- the input dataset in .rds format and associated metadata :
test_data.rds (the clinician dataset), cancer_type.rds (the cancer type according to the TCGA nomenclature), input_k_value.rds (the number of cell type to consider when running unsupervised methods, this file in empty when running semi-supervised or supervised methods)
- some log and metadata files generated by the codabench platform (not of use for results interpretation):
metadata, output_program.txt, Rprof.out, Rprof.rds
- 3: Dowload only the estimated proportions in a .csv format (only the
results_A_1.csv file)
Check status of methods run
- 1: Click on the arrow to view the Methoods run agains the dataset
- 2: Date of the run
- 3: Status of the run
Dataset contained in the benchmark
DT1: Pancreatic adenocarcinoma
- omic type: transcriptome RNAseq (pseudo-log)
- cohort size : 30
- Ground Truth : in silico simulations of all cell types
- More informations at dataset_factsheet_DT1.html
DT2: Pancreatic adenocarcinoma
- omic type: transcriptome RNAseq (linear scale)
- cohort size : 30
- Ground Truth : in silico simulations of all cell types
- More informations at dataset_factsheet_DT2.html
DT4: Colorectal adenocarcinoma
- omic type: transcriptome RNAseq (pseudolog)
- cohort size : 12
- Ground Truth : in silico simulations of different immune composition
- More informations at dataset_factsheet_DT4.html
DT5: Colorectal adenocarcinoma
- omic type: transcriptome RNAseq (linear scale)
- cohort size : 12
- Ground Truth : in silico simulations of different immune composition
- More informations at dataset_factsheet_DT5.html
DT8: Breast invasive cancer
- omic type: transcriptome RNAseq (pseudolog)
- cohort size : 32
- Ground Truth : in vitro mixture of different cell types
- More informations at dataset_factsheet_DT8.html
DT9: Breast invasive cancer
- omic type: transcriptome RNAseq (linear scale)
- cohort size : 32
- Ground Truth : in vitro mixture of different cell types
- More informations at dataset_factsheet_DT9.html
DT10: Breast invasive cancer
- omic type: transcriptome RNAseq (pseudolog)
- cohort size : 30
- Ground Truth : in silico simulations of all cell types
- More informations at dataset_factsheet_DT10.html
DT11: Breast invasive cancer
- omic type: transcriptome RNAseq (linear scale)
- cohort size : 30
- Ground Truth : in silico simulations of all cell types
- More informations at dataset_factsheet_DT11.html
DT12: Breast invasive cancer
- omic type: methylome targeted bisulfite sequencing (beta_value)
- cohort size : 30
- Ground Truth : in silico simulations of all cell types
- More informations at dataset_factsheet_DT12.html
Methods contained in the benchmark
Unsupervised
Unsupervised methods do not use biological references.
MT1_ICA_fs: ICA with ICA-based feature selection
MT2_NMF_fs: NMF with ICA-based feature selection
MT3_edec: Edec method
MT14_ICA: ICA without feature selection
MT19_NMF: NMF without feature selection
Semi-supervised
Semi-supervised methods use a gene marker list as input of the deconvolution.
MT16_cellmix_ssKL: CellMix semi-supervised NMF using KL divergence
MT17_cellmix_DSA: CellMix digital sorting algorithm
MT18_cellmix_ssFrobenius: CellMix semi-supervised NMF using eucideann distance
Supervised
Supervised method use a cell-type specific expression matrix as input of the deconvolution.
MT8_cibersort: Cibersort method
MT9_epic: EPIC method
MT11_quantiseq: Quantiseq method
