Intelli-NGS

Intelli-NGS applies deep neural networks to improve variant calling accuracy in IonTorrent sequencing data by distinguishing true variant calls from erroneous calls.

Key Features:

• Deep Neural Network Architecture: A deep learning model employing fully connected dense layers processes input features to classify variants.
• Implementation: The model is implemented in Python 3 using TensorFlow.
• Input Data Handling: Accepts Variant Call Format (VCF) files containing variant information from IonTorrent sequencing runs.
• Comprehensive Parameter Utilization: Analyzes variants based on thirty-five parameters provided by the IonTorrent platform, including flow-space information.
• Performance Metrics: Validated against Genome in a Bottle (GIAB) data with reported accuracy of 93.08% and ROC-AUC of 0.95.
• Variant Annotation: Annotates variants using online databases such as dbSNP and ClinVar.
• Probability Scoring: Assigns a probability score to each variant indicating the likelihood of being a true positive or false positive.

Scientific Applications:

• IonTorrent variant refinement: Reduces false positive and false negative rates in IonTorrent sequencing analyses while maintaining high recall.
• Clinical and research interpretation: Supports clinical diagnostics and personalized medicine by improving the reliability of variant interpretation through annotated and scored variant calls.

Methodology:

The model was created from scratch and implemented in Python 3 with TensorFlow using fully connected dense layers to process input features, and it was trained and extensively validated using established datasets including Genome in a Bottle (GIAB).