Application of Deep Learning in Biological Big Data Analysis

Introduction

The generation of massive amount of data in this era is a good process in the biological systems and contributes for big data. The big data can be of any type like epigenome, genome, proteome, transcriptome, and metabolome, etc. The big data is defined by its four key characteristics first is volume and others are variety, velocity, and variability. The structure of big data is defined by three types: first is structured, next is unstructured and the last is semi-structured (Mirza et al., 2019). The biological data is very complex as compared to other data because the regulation of one gene or protein depends on the behavior of other genes or regulatory element proteins. The one entity in biological data can regulate many entities and vice versa. In recent years the vast amount of biological data is generated due to the technology advancement towards the high throughput sequencing, medical image processing, genome-wide association studies, gene expression analysis, protein binding motifs and expression studies, pathway and network level analyses, and structural investigations of biological entities etc. These types of data need a complete workflow for the analysis. As earlier described, the biological systems are very complex and not regulated by one entity. Hence in the case of genome-wide association studies, scientists focus on the genetic variants which are associated with the measured phenotypes while only one phenotype is not involved in the disease. It is a very complex process and several elements participate in the disease cascade so by the analysis of one gene or protein or a single type of data, we cannot analyze all the disease spreading factors (C. Xu & Jackson, 2019). Therefore the analysis of all the factors simultaneously can give a better measurement of the disease causing and spreading factors (Zitnik et al., 2019). The other and major challenge regarding the big data is its dimensionality. The big data have high dimensions described by high-resolution data, while in the case of biological data the samples which are collected from the different patients are limited and much less than the number of variables due to its high costs or limited resources like Alzheimer disease patients or replicates of sequencing so they lead to data sparsity, multicollinearity, multiple testing, and overfitting (Altman & Krzywinski, 2018).