This week our speaker will be our very own Applied Mathematics Graduate Student Ali Heydari.
When: Wednesday, April 20 at 11am
Where: Zoom (https://ucmerced.zoom.us/j/98050375649?pwd=bFFYaFlJUG5TTVFNeStzaHBCS1grUT09) and in-person (whichever Applied Math Conference Room is available!
Title: N-ACT: An Interpretable Deep Learning Model for Automatic Marker Gene and Cell Type Identification
Abstract:
Single-cell RNA sequencing (scRNAseq) is rapidly advancing our understanding of the cellular composition within complex tissues and organisms. A major limitation in most single-cell RNA sequencing (scRNAseq) analysis pipelines is the reliance on manual annotations to determine cell identities, which are time-consuming and subjective. Given the growth in number of sequenced cells, supervised methods--specially Deep Learning (DL) models--have been developed for automatic cell type identification (ACTI), achieving high accuracy and providing scalability. However, all existing DL frameworks for ACTI lack interpretability and are used as ``black-box” models. We present N-ACT (Neural-Attention for Cell Type identification): the first interpretable deep neural network for ACTI that utilizes a novel attentive mechanism for detecting landmark genes used to identify cell-types. On all tested datasets, our results demonstrate that N-ACT accurately identifies landmark genes and cell types in an unsupervised (or semi-supervised) manner, while performing comparable to the current state-of-the-art ACTI on traditional supervised classification tasks.
