Current Projects!

Fragment-Based Drug Discovery (FBDD) is a novel and promising approach to drug design, exemplified by its success in identifying inhibitors for the cancer-related protein Bcl-2. FBDD is computationally less expensive and more cost- effective than standard screening, making it a valuable tool for drug discovery. However, FBDD is often limited by the difficulty of identifying optimal fragments, due to the many criteria a fragment needs to meet, such as suitable molecular weight, solubility, and binding affinity. Thus, there is an unmet need for a tool that predicts the best fragment for a target molecule, especially in medicinal chemistry.

This project explores the use of clozapine, an antipsychotic medication primarily prescribed for treatment-resistant schizophrenia. Our objective is to examine the interactions between clozapine, its metabolites, and specific Human Leukocyte Antigen (HLA) proteins. HLA proteins play a crucial role in the immune system, helping to regulate immune responses. Abnormal interactions with medications like clozapine can lead to severe side effects, such as agranulocytosis, a condition characterized by a dangerously low white blood cell count. Clozapine, known for its efficacy in managing symptoms of schizophrenia that don't respond to other treatments, has a unique pharmacological profile, which makes understanding its interaction with HLA proteins critical. Schizophrenia affects about 1% of the population worldwide, and patients with treatment-resistant forms of this disorder often rely on clozapine as a last resort. This study aims to shed light on the molecular mechanisms of clozapine's side effects, potentially improving treatment safety for this challenging psychiatric condition

Bi-Specific T-Cell Engager (BiTE) therapy is an immunotherapy platform that redirects cytotoxic T cells toward tumor cells by dual binding to CD3 on T cells and to tumor-associated antigens. While the FDA-approved BiTE blinatumomab has shown success in B-cell acute lymphoblastic leukemia, conventional BiTEs face limitations due to off-tumor effects when targeting surface antigens shared with normal cells. To address this, we designed two novel TCR-BiTE structures composed of a single-chain variable fragment (scFv) fused with a T-cell receptor (TCR) engineered to recognize the KRAS G12V neoantigen, a mutation lacking any FDA-approved targeted therapies. This approach aims to combine the precision of TCR recognition with the potent cytotoxic engagement of BiTEs, offering a potential therapeutic strategy against KRAS-driven cancers.

This project introduces DeepBERTa, a ChemBERTa-based model trained on DeepSMILES, a simplified molecular representation that reduces parsing errors compared to SMILES. Using ~34,000 molecule–fragment pairs from FDA-approved drugs and the BBBP benchmark, we show that a hybrid approach combining SMILES and DeepSMILES predictions improves mean Tanimoto similarity from 0.290 to 0.347, demonstrating the value of alternative encodings for fragment based drug discovery.

This project, in collaboration with the Cunha Lab, explores the role of bacterial metabolites in modulating T-cell activation and their potential links to colorectal cancer. Focusing on key metabolic pathways like histidine metabolism and lipopolysaccharide biosynthesis, we aim to unravel how specific bacterial products influence immune responses and tumor development. Additionally, we leverage docking, molecular dynamics, and other computational chemistry approaches to deepen our understanding of these interactions.

The KRAS gene, originally identified as an oncogene in the Kirsten Rat Sarcoma virus, directs the production of the K-Ras protein. This protein plays a crucial role in transmitting growth signals from outside the cell to its nucleus. Mutations in KRAS can lead to its constant activation, resulting in uncontrolled cell growth and signaling, and thus contributing to the development of cancer. KRAS mutations are present in about 25% of all human cancers, including over 90% of pancreatic cancers and around 25% of lung cancers. Despite being a prevalent oncogene, the lack of distinct binding sites on K-Ras has made it one of the most challenging targets in cancer therapy. Our team is focusing on developing novel compounds that can act as sensors for KRAS activity, potentially opening new avenues for therapeutic intervention.

This project investigates how routine ICU data can be used to identify patients with Alzheimer’s disease. Using the MIMIC IV v3.1 database, we study adult ICU patients and incorporate demographics, insurance information, vital signs, laboratory results such as creatinine and complete blood count, and medication exposures including cognitive enhancers, anticholinergics, antipsychotics, statins, and insulin. Logistic Regression and XGBoost models are trained and compared, with interpretability provided through regression coefficients and SHAP values. A central focus is iron homeostasis, where iron-related laboratory features and blood indices are evaluated for their contribution to distinguishing patients with Alzheimer’s disease. This approach highlights how medications and routine clinical data can provide insights into Alzheimer’s detection in critically ill populations.

This project examines potential safety signals of GLP-1 receptor agonists (GLP-1 RAs) in two areas: neovascular age-related macular degeneration (nAMD) and papillary thyroid cancer (PTC). Building on recent evidence of increased nAMD risk and prior reports suggesting a thyroid link, we will use computational docking, molecular dynamics, and pathway modeling to test whether GLP-1 RAs interact with ocular angiogenic or thyroid signaling pathways. Alongside a structured review of published studies, this approach aims to clarify whether observed associations reflect true drug effects, off-target interactions, or bias, providing mechanistic insight to guide clinical interpretation.