This repository contains raw data and analysis notebook for manuscript (in progress):

"Single cell mRNA signals reveal a distinct developmental state of KMT2A-rearranged infant B-lymphoblastic leukaemia"

"data" folder contains following:

• single_cell_data (10X single cell leukemia data):

  • raw_counts = output of cellranger, i.e content of filtered_gene_bc_matrices (ALL_AML_1) or filtered_feature_bc_matrix (ALL_AML_2 and ALL_AML_3)
  • SoupX_processed_counts = "decontaminated" raw data, i.e. after running SoupX
  • P1/P3/P5_iALL_clusters.tsv = output file from souporcell software for multiplexed samples.
  • final_h5ad = h5ad object for each sample with raw counts in adata.X slot and sample ID and cancer/normal cell classification in adata.obs.patient_cancer *Please note that I use slightly extended IDs, which correspond to the case ID in the paper as following:

  case ID	extended ID
  1	P1_iALLM
  2	P2_iALLF
  3	P3_iALL
  4	P4_iALL
  5	P5_iALL
  6	P6_iALL
  7	P7_iALL7_NUTM1
  8	P8_iALL8_ETV6
  9	P9_iAML_Dx
  10	P10_iAML

• bulk_data (bulk data run through Salmon):

  • StJude = matrix of counts with gene length in second column (stJudes_bulk_txiConvGlen.csv) together with metadata
  • TARGET = matrix of counts (TARGET_raw_countsALLwGlen.csv) together with metadata

"analysis_results" folder contains following:

• deconvolution = results of running of bulk rna-seq deconvolution (cellular signal analysis, github.com/constantAmateur/cellSignalAnalysis), includes StJudes*, TARGET* and InHouse* (lineage switch case, as on Figure 3).

• logreg = results fo logistic regression (single cell cancer to normal comparison, github.com/constantAmateur/scKidneyTumors) and includes FetalBM_logreg_trained_model.rds (logistic regression model trained on the fetal bone marrow reference) and InfALL_allSAMPLES_LogregFBM.csv (probability of similarity to each of the normal cell type per leukemia single cell)

Leukemia_data_analysis_final.ipynb Jupiter notebook shows the preprocessing steps (demultiplexing, normalising) and filtering (mitochondrial content < 20%), as well as code to reproduce Figure 1 (deconvolution results) and Figure 2 (UMAPs, result of logistic regression and siluette plot)