MSc Thesis Presentation - Runtong Cai

Name: Runtong Cai

Date: Dec 8th

Time:  1-3 pm

Location: ICICS 204

Supervisor's name: Jiarui Ding
Thesis title: Marker Gene Guided Additive Models can Decompose and Extrapolate Single-cell Expression Aligned with Known Gene Programs

Abstract: Gene expression, viewed through scRNA-seq results in a high-dimensional matrix that is difficult to interpret due to high information content that does not reflect underlying biological structure. This thesis presents an extension of the VAE architecture in the form of a Supervised Additive decoder Model (SAM) which parallels the model of gene expression as a composition of largely independent subsystems. SAM was applied to signal isolation, noise removal, and perturbation prediction where it exhibited overall superior performance when benchmarked against a panel of state-of-the-art models on tasks such as resolving tissue zonation, activities analysis, and cell type transfer, among others. SAM provides a method for interpreting and making predictions on gene expression that is aligned with prior knowledge.