�COVID-Datathon: Biomarker identification for�COVID-19 severity based on BALF scRNA-seq data��Seyednami Niyakan Xiaoning Qian ��IEEE COVID-19 Hackathon�

• Outline:

• Dataset:

• Single-cell RNA sequencing (scRNA-seq) data of BALF cells
• 23189 cells with three imbalanced classes:

3292 Mild cells, 7919 Severe cells, 11978 Normal cells

• 1999 genes (features)
• Batch normalization have been done
• Each gene expression across data can be approximated

with Normal distribution

• Biomarker Identification:

• Monocle R package for scRNA-seq data with choice of Normal distribution for data
• Three sets of Gene Differential Expression (DE) analysis between one label vs the rest

1- Normal cells vs Rest 2- Severe cells vs Rest 3- Mild cells vs Rest

• Ranking genes based on their adjusted p-values
• Intersection of top 100 DE genes in three DE analyzes as Covid-19 severity biomarkers

• Biomarker Identification:

• Out of 12 detected Biomarkers, 6 of them have been recently studied as genetic COVID-19 severity biomarkers in BALF cells: (RSAD2, CXCL10, CXCL11, CCL3, IFIT1, CCL2)
• The other half can be new potential biomarkers of COVID-19 severity suggested by our proposed bioinformatic pipeline: (IDO1, GCH1, CRYBA4, LGMN, CTSB, GBP1)

​

​

• Cell Classification :

• We propose three different gene sets, all constructed from the top 100 DE genes in three initial DE analyzes:

G1: Intersection of top 100 DE genes in all three analyzes ( 12 genes)

​

G2: Intersection of top 100 DE genes in Normal vs rest and Severe vs rest DE analyzes(88 genes)

​

G3: Union of top 100 DE genes in all three analyzes (194 genes)

​

• We run each classification algorithm on each of these proposed gene sets

​

• Cell Classification :

• Randomly diving Data to 75/25 split to make Training and Test set data

• Cell Classification :

• Questions :

• All the Codes are available in our Github page:

https://github.com/namini94/scBALF_Hackathon

​

• Questions!