AIKYATAN

AIKYATAN annotates distal regulatory elements in the non-coding genome by classifying enhancers and other distal epigenomic regulatory sites from histone modification patterns using convolutional neural networks and other machine-learning models.

Key Features:

• Binary Classification System: Employs a binary classifier to distinguish distal regulatory regions from non-regulatory regions using combinatorial histone modification signatures.
• Machine Learning Models Suite: Implements Support Vector Machines (SVM), random forest variants, deep neural networks (DNN), and convolutional neural networks (CNN) for comparative modeling.
• Convolutional Neural Network Performance: The CNN reached 97.9% accuracy on the human embryonic cell line H1 dataset, and GPU training yielded 21× acceleration for DNNs and 30× acceleration for CNNs compared to CPUs.
• Comparative Performance Metrics: AIKYATAN-CNN achieved a 40% higher validation rate than CSIANN and matched the accuracy of RFECS.
• Spatial Feature Pooling: Represents epigenomic inputs with image-like properties and applies spatial pooling of features to capture salient epigenomic characteristics.
• Scalability and Efficiency: Scales efficiently to large and diverse epigenomic datasets suitable for large-scale genomic studies.

Scientific Applications:

• Functional Annotation: Identifies enhancers and insulators to annotate non-coding regulatory elements.
• Epigenomics Research: Deciphers complex epigenomic landscapes associated with distal regulatory elements and functional genomic variants.
• Genetic Studies: Provides regulatory-region annotations to support studies of gene regulation and its implications in health and disease.

Methodology:

AIKYATAN uses convolutional learning on GPUs to process integrated epigenomic datasets, optimizes activation and pooling functions within CNNs, and conducts empirical validation experiments.