Building Machine Learning Models on Limited Transcriptomic RNA-Seq Data

Researchers can use transcriptomic RNA-seq data to acquire gene expression information for specialized domains like tumor diagnosis and radiation biomarker identification. The collection costs of transcriptomic RNA-seq data are high, and the available sampling sources are limited because of numerous in-fluencing factors such as age, gender, and population. The limited number of data samples will significantly amplify the influence of unqualified data resulting from system and experimental errors. Furthermore, the limited size of the dataset will hinder the use of machine learning modeling techniques, thereby obstructing the research. This paper proposes a processing framework to solve this problem. We employ a set of techniques to remove unqualified samples and expand the existing data volume in a reasonable manner, given the limited number of samples. We tested the proposed method on bulk transcriptomic RNA-seq data, and the experiment successfully constructed machine learning models with a limited data volume, most of which yielded impressive outcomes.