Authors:

Kyle Mundy

Date Created:

Not specified

Course Title:

Professor:

Not specified

About Paper:

New genome sequencing technologies have greatly increased the number of transcriptomic datasets available for research. With this, machine learning (ML) models have been used to predict clinical information, including age, smoking/drinking history, and BMI. However, a notable limitation is that the majority of these samples lack proper annotation, so large datasets cannot be easily constructed to increase the accuracy of ML models. This project addresses these issues by attempting to build an ML model to predict clinical information from large-scale sequencing data across multiple databases. There are 2 major components: 1. Collect all human RNA-seq data from public databases to establish a comprehensive and large-scale dataset, 2. train ML models for predicting clinical traits. The human RNA- seq samples with annotated traits are collected from public datasets, and the matching processed data is taken from existing transcriptomic project portals, providing a potential dataset of > 1.7 million samples. Then, separate ML models are trained, tested, and optimized for predictions from information across all matched RNA-seq samples. The resulting dataset is larger than any previous studies, with hundreds of thousands of samples collected. Additionally, the Receiver Operating Characteristic (ROC) curve and related results matrix are compared against current standards for transcriptomic ML models, with increased accuracy showing the benefits of growing datasets and clear annotations by researchers. These findings will serve as a valuable asset for researchers, enabling more simple and accurate extraction of pertinent clinical information from transcriptomic data.

Source:

Purdue University / 2023

Topics:

No topics listed

Co-authors:

Kyle Mundy