Skip to content

Latest commit

 

History

History
24 lines (20 loc) · 2.29 KB

README.md

File metadata and controls

24 lines (20 loc) · 2.29 KB
Raw

Myungin Baek et al., Molecular Logic of Spinocerebellar Tract Neuron Diversity and Connectivity, Cell Reports, 2019.

SBC-analysis

This repository contains codes which are used in analyzing the data from DISIST. Note that

  • We can't provide the used data (merged_count_matrix_3) and its metadata (condtion_3)
  • During analyzing, we figure out that the used dataset has problems (in preprocessing-level). So DIGIST has been resolving this problem. We will re-analyze the corrected data and report results.
  • The code 92-113 might not be able to work. I will deal with this part after getting the data.

Background

Suppose we have a n_gene by n_cell matrix. (it is ok to understand a gene as a feature and a cell as one datapoint if you want). Each cell follows from a known population. In this situation, which genes represent a populations? how much represent?

Method

Statistical models give us good answers. "How much" is arranged by p-value and "Which" is determined by genes having lower p-value. But, if there is covariate effect, we should adjust the original data or establish a corrected data matrix. Our data contains two different sequencing data. One is from DropSeq and the other is from Smart-seq2. They are much different. One way to adress this issue is to introduce a surrogate variable, which will be used at modeling step with the data. Please refer to Draft_1_4.pptx in this repository if you want to know more backgorunds, methods, and results. (NB: the defintion of our surrogate variable is still private.)

Results (partial)

This figure shows that 40% genes are filtered out (that is, they are not tested whether each of them is meaningful or not)

The same result of the previous figure, but different aspect. Indeed, the more meaningless (higher p-value) it was, the less tested it was.

A final output of DESeq2 with some genes

A gene clustering result. More results can be fonund in Draft_1_4.pptx.