APPLICATION of DEEP LEARNING ALGORITHMS for NEXT GENERATION SEQUENCING DATA
Project no. 21725 · Status: Kapanmış · 2020
Summary
The era, which started with Mendel's discovery of the rules of inheritance, gained a different dimension with the Human Genome Project (IGP). With the widespread use of biological data processing technology, statistics, mathematics and computer sciences have become complementary to each other in biological data. Due to the decrease in the prices of technologies in this field, the direction of studies in genetics and medicine has changed. Next-generation sequencing (NGS) technology has become an indispensable part of biological research in many fields. By using NGS technologies, billions of nucleotides (building blocks that make up DNA) in the whole genome, transcriptome or smaller target regions can be sequenced. After these operations, very large data matrices appear. In the process of determining meaningful information from large-scale data matrices, the need for computer technologies and deep learning-based approaches has increased. In this study, the predictive power of deep learning algorithms for the bistable (0/1; patient-healthy…) classroom was investigated in real whole human genome and simulated data. In the research, the effect of hyperparameters of the deep learning algorithm has been demonstrated. The performance of the DL models is reported when the Epoch 500, 1000, 2000 and LR 0.01 and 0.001 are taken. The highest classification estimates in real and simulated data were obtained at 2000 epochs and 0.001 LR values. In addition, the success of estimating the Recall and Precision values of deep learning algorithms very close to each other in cases of class imbalance shows the researchers that the results of the method can be trusted in population-based studies. Obtained comparison results will benefit researchers in terms of time and cost in their studies. Researchers who know the effect of changes in parameters such as epoch, number of layers, number of iterations on model performance will be able to report the analysis results of genetic data quickly and with high accuracy by using the outputs of this study.