Design and Development of Intelligent Gene Patterns Discovery Mechanisms to Predict Human Diseases

Abstract

Understanding and predicting Human Genome Patterns are one of the challenging issues regarding human health. To achieve the highest Classification Accuracy, a large amount of Genome Data Sets need to analyze. It is noted that a single Gene is not responsible for many Human Diseases and instead, diseases occur by different or group of genomes interacting together and causes diseases. Hence it needs to analyze and associate the complete genome sequences with understanding or predicting various possible human diseases. This research work identified three recently proposed popular Genome Cluster-Classifiers, namely i. Hierarchical-Random Forest based Clustering (HRF-Cluster), ii. Genetic Algorithm-Gene Association Classifier and iii. Weighted Common Neighbor Classifier (wCN). These Classifiers were implemented and thoroughly studied in terms of Prediction Accuracy, Memory Utilization, Memory Usage and Processing Time. From our experimental results, it is noted that the performances of these three classifiers purely depend on the patterns of genomes. newlineThis Research work noticed that identifying Gene Signatures in association with Gene Expression will help to predict diseases patterns with higher Classification Accuracy. This research work is proposed Gene Signature based HRF Cluster (G-HR) which is based on the idea mentioned above.vi newlineThe results of the proposed G-HR Classifier-Cluster was outperforming as compared with the identified existing three Classifiers. To improve the performances of the G-HR further in term of Processing Time, the G-HR was implemented under Parallel Framework with Two, Four, Eight and Sixteen Parallel Processors and evaluated. Results show that the Processing Time decreases considerably. newlineThis Work revealed that the G-HR unable to support for High Dimensional Data Sets. To address this issue, Gene Signature based Hierarchical Weighted Random Forest Clustering Technique (G-HWRF) was proposed. Results show that G-HWRF is performing well as compared with that of our previous work G-HR Cluster. As a few Genome Patterns are causing a few diseases, it is needed to associate genome patterns to predict possible human diseases. To address this demand, the Deep Learning based Intelligent Human Diseases-Gene Association Prediction Technique (IHDGAP) was proposed and this employs Convolution Neural Network (CNN) algorithm. To improve Prediction Accuracy of IHDGAP further, an Enhanced Convolution Neural Network (ECNN) algorithm is employed and the Classifier was named as Human Diseases Pattern Prediction Technique (ECNN-HDPT). Results show that the Classification Accuracy and FScore are higher, but it consumes relatively more Memory and Processing Time as we employ SVM-ECNN to train and test Patterns. newline newline newline newline