Our research specializes in the AI-driven design and optimization of novel therapeutic molecules, including small-molecule drugs and antibodies. By employing cutting-edge deep generative models, we pioneer de novo molecular design, creating and optimizing new chemical entities with specific, desired properties from scratch. This approach allows for the exploration of vast chemical spaces beyond human intuition to discover promising drug candidates and engineer antibodies with enhanced efficacy and stability.
Our lab employs sophisticated AI and graph-based techniques to advance drug repurposing and predict drug-drug interactions. We develop novel methods that leverage heterogeneous biological data to identify new therapeutic applications for existing drugs, significantly accelerating their potential for clinical use. Furthermore, we utilize graph neural networks to systematically predict and analyze complex drug-drug interactions, aiming to enhance patient safety and optimize combination therapies by uncovering latent pharmacological effects.
A critical focus of our research is the AI-powered prediction of polypharmacy side effects to ensure the safety of multi-drug regimens. We develop novel computational methods, including Positive-Unlabeled Learning with Graph Neural Networks and multi-source data integration, to identify adverse drug interactions that may not be evident in clinical trials. This work is vital for proactively assessing drug combination risks, ultimately guiding safer prescription practices and improving patient outcomes in complex therapeutic scenarios.
We also leverage AI-based models to advance peptide therapeutics through two main approaches: the accurate classification of peptide types and the generative design of new candidates. Our work specifically targets critical classes such as Antimicrobial, Anti-cancer, Anti-inflammatory, Antiviral, and Antifungal peptides.
A central research thrust in our lab involves the identification of cancer driver genes using advanced AI and graph representation learning. We develop interpretable, transformer-based models that integrate multi-omics data (such as mutations, expression, and copy number alterations) with the complex topologies of biological networks. This approach not only achieves state-of-the-art accuracy in pinpointing genes responsible for cancer initiation and progression but also provides crucial biological insights by revealing the influential omics features and higher-order network pathways—such as those involving transcription factors and miRNAs—that underpin these predictions. Our work facilitates a deeper understanding of carcinogenesis and aims to reliably discover novel cancer gene candidates for therapeutic targeting.
A key research direction in our lab focuses on the AI-driven prediction of drug synergy to accelerate the discovery of effective combination therapies, particularly in oncology. We develop advanced deep learning models, such as our novel BT-Synergy framework, which integrates hybrid BiLSTM-Transformer architectures to capture complex patterns from molecular drug representations (SELFIES). By simultaneously incorporating contextual protein embeddings from pre-trained language models to characterize cell line-specific biology, our models accurately identify synergistic drug pairs. This approach consistently outperforms existing state-of-the-art methods, providing a powerful and generalizable computational tool to efficiently screen the vast combinatorial drug space and guide experimental validation.
Our research also focuses on applying AI to clinical diagnosis. This includes tumor detection in medical scans, early diagnosis of autism and Alzheimer’s, and retinal image analysis for diabetic retinopathy and vascular segmentation. These tools aim to support clinicians in delivering earlier and more accurate care.
Our research also covers some distinct AI-based applications such as virtual try-on, voice conversion, driver drowsiness detection, sentiment analysis, image enhancement, and deepfake detection.