Deep Learning models for interpreting biological data with prior knowledge

Cancer subtype classification and modeling by pathway attention and propagation

• Designing a pathway-based explainable deep learning model by graph convolutional network and attention mechanism
• Predicting cancer subtypes using gene expression data and pathway information
• Paper accepted by Bioinformatics journal

Learning Cell-Type-Specific Gene Regulation Mechanisms by Multi-Attention Based Deep Learning with Regulatory Latent Space

• Designing multi-modal deep learning model for learning gene regulation mechanism
• Elucidating epigenetic gene regulation mechanism by various attention layers
• Paper accepted by Frontiers in Genetics journal

Multi-layered Knowledge Graph Neural Network Reveals Pathway-level Agreement of Three Breast Cancer Multi-gene Assays

• Designing an explainable deep learning model by integrating intra- and inter- pathway level attention
• Identification of shared regulatory mechanisms of three breast cancer multi-gene assays
• Under review in SCIE journal

AI with graph structured data in biomedical domains

Sparse Structure Learning via Graph Neural Networks for Inductive Document Classification

• A novel GNN-based sparse structure learning model for inductive document classification
• Employing structure learning to sparsely select edges between words by considering dynamic contextual dependencies
• Paper accepted by AAAI 2022

Biomedical knowledge graph learning for drug repurposing by extending guilt-by-association to multiple layers

• A semantic multi-layer guilt-by-association approach that leverages the principle of guilt-by-association - “similar genes share similar functions” at the drug-gene-disease level
• Designing a semantic information-guided random walk to generate embeddings of drugs and disease in a unified embedding space
• Paper accepted by Nature Communications journal

Cheminformatics and AI in Drug discovery

Multi-Task Informed Learnable Prototypes on Few Shot learning for Molecular Property Prediction

• Leveraging shared knowledge across multiple molecular properties in few-shot learning
• Incorporating a stochastic attention mechanisms to reflect information from multiple assays
• Under review in International Conference

Dual Representation Learning for Predicting Drug-side Effect Frequency using Protein Target Information

• Utilizing heterogenous features of drugs and side effects to get improved latent representations
• Compensating for the drugs without clear target proteins using the Adaboost method
• Paper accepted by IEEE Journal of Biomedical and Health Informatics journal

Improved drug response prediction by drug target data integration via network-based profiling

• A novel model-agnostic framework to enhance performances of existing deep learning models for drug response prediction
• Network-based estimation of gene perturbation by drug treatment
• Paper accepted by Briefings in Bioinformatics journal