ML Algorithms to Solve Neuronal Cell Type Matching

Group Members: Beverly Peng, Daniel Carrillo, Huy Le, and Janelle Uy

Advisors: Richard Scheuermann, Brian Aevermann, and Renee Zhang

Problem Statement

Single-cell RNA sequencing is rising in popularity compared to the traditional bulk analysis due to its ability to account for the heterogeneous nature of tissues and to discover novel cell types. While scRNA-seq can allow for powerful computational classification of cells, there is a problem of cross referencing cell types between different anatomic regions, species, and platforms. This project aims to define cell type matching as a classification problem and utilize various machine learning techniques, such as decision trees and deep learning neural networks, to cross reference cell types from different samples. The data used in this analysis include transcriptomic, anatomical, physiological, and synaptomic data of neuronal cells from the six cortical layers from four previously published papers focusing on the classification of these cells via clustering techniques. A metric must be developed to test the performance of the chosen machine learning algorithms, allowing them to be compared with each other for better results and run-time. The results must be evaluated with varying explainability and prediction accuracy in mind, so the students will develop a strong understanding of how and why each method works with various datasets.

Objectives

1. Develop a series of metrics for measuring the performance and reliability of each machine learning method.

2. Compare the performance of different machine learning classification methods for cell type classification.

3. Assess the impact of hyper-parameter adjustment and feature selection on model performance.

Made by Janelle Uy