Recent Landscape of Deep Learning Intervention and Consecutive Clustering on Biomedical Diagnosis

Abstract

Background: Consecutive Clustering is one type of learning method that is built on neural network. It is frequently used in different domains including biomedical research. It is very useful for consecutive clustering (adjacent clustering). Adjacent clustering is highly used where there are various specific locations or addresses denoting each individual features in the data that need to be grouped consecutively. One of the useful consecutive clustering in the field of biomedical research is differentially methylated region (DMR) finding analysis on various CpG sites (features).Method: So far, many researches have been carried out on deep learn-ing and consecutive clustering in biomedical domain. But for epigenetics study, very limited survey papers have been published till now where con-secutive clustering has been demonstrated together. Hence, in this study, we contributed a comprehensive survey on several fundamental categories of consecutive clustering, e.g., Convolutional Neural Network(CNN) Auto-Encoder (AE), Restricted Boltzmann Machines (RBM) and Deep Belief Net-work (DBN), Recurrent Neural Network (RNN), Deep Stacking Networks (DSN), Long Short Term Memory (LSTM) / Gated Recurrent Unit (GRU) Network etc., along with their applications, advantages and disadvantages. Different forms of consecutive clustering algorithms are covered in the second section (viz., supervised and unsupervised DMR finding methods) used for DNA methylation data have been described here along with their advantages, shortcomings and overall performance estimation (power, time). Conclusion:Our survey paper provides a latest research work that have been done for consecutive clustering algorithms for healthcare purposes. All the usages, benefits and shortcomings along with their performance evaluation of each algorithm has been elaborated in our manuscript by which new biomedical researchers can understand and use those tools and algorithms for their research prospective.