TWO-SIGMA

"TWO-SIGMA" (TWO-component SInGle cell Model-based Association) is an R package for differential expression (DE) analyses in single-cell RNA sequencing (scRNA-seq) data. It employs a two-component modeling approach to address the complexities and unique challenges of scRNA-seq data, including drop-outs, overdispersion, and zero inflation.

Key Features:

1. Drop-out Modeling: The first component of TWO-SIGMA uses a mixed-effects logistic regression model to account for the probability of "drop-out," a common issue in scRNA-seq data where transcripts are not detected even though they are present.

2. Expression Modeling: The second component employs a mixed-effects negative binomial regression model to model the mean expression of genes, accommodating the overdispersed and zero-inflated nature of scRNA-seq data.

3. Flexibility and Versatility: TWO-SIGMA is designed to be extremely flexible, offering several advantageous features:
- No requirement for log-transformation of outcome data.
- Ability to handle overdispersed and zero-inflated counts.
- Incorporation of a correlation structure between cells from the same individual through random effect terms.
- Suitability for unbalanced designs, where the number of cells may vary across samples.
- Controls for additional sample-level and cell-level covariates, including batch effects.
- Provision of interpretable effect size estimates.
- It is possible to conduct general differential expression tests beyond simple two-group comparisons.