The Kenny/Lengyel laboratory is part of the Department of Obstetrics and Gynecology/ Section of Gynecologic Oncology, studying the biology of ovarian cancer. The laboratory has about 15 members investigating the role of metabolism and methyltransferases in ovarian cancer metastasis. We use a variety of cutting-edge methods, including spatial proteomics, spatial metabolomics, 3D organotypic cultures of human tissue, spatiotemporal characterization of the immune system, and stable-isotype tracing in patients. Bioinformatic support and access to all Core facilities at the University of Chicago are available in the laboratory. We develop novel inhibitors to treat cancer and study biomarkers to diagnose ovarian cancer and detect ovarian cancer early. Our translational research laboratory is in the Center for Integrated Science, a research building on campus that houses 40 independent research groups. This at-will position is wholly or partially funded by contractual grant funding, which is renewed under provisions set by the grantor of the contract. Employment will be contingent upon the continued receipt of these grant funds and satisfactory job performance. The job facilitates and promotes a research project or contributes to the scientific direction of a research resource. Receives a moderate level of guidance and direction. The focus of the job is to oversee a subgroup in the laboratory dedicated to projects using machine learning, data analysis, and artificial intelligence. The overall goal is to conduct multiple projects: Textual Prediction of Survival (LLM classification & Attention Modelling) This project develops a model to predict patient survival by analyzing heterogeneous clinical documents. Unlike traditional methods that rely on predefined predictive factors, it is designed to operate where risk markers are poorly understood, as is the case in ovarian cancer. The model integrates predictive accuracy with interpretability, offering individualized survival forecasts and treatment response predictions. Importantly, it highlights the textual features driving predictions, enabling discovery of novel risk associations for advancing precision oncology. Multimodal Prediction of Response to Platinum-Based Chemotherapy (Computational Pathology & MIL) This work combines histopathological hematoxylin and eosin (H&E) stained slides with clinical and demographic patient data to predict chemotherapy response. Using data from patients at the University of Chicago, the model will be validated in the Finnish DECIDER cohort of patients. Ongoing efforts aim to validate performance in the Australian AOCS cohort. The project demonstrates the potential of integrating multimodal data for improved clinical decision support. Scanner Calibration (Computational Pathology & Omics normalization) This project introduces a method to calibrate latent feature spaces of pathology foundation models to correct for scanner-induced variability. Without calibration, performance drops by ~15% across different scanners. Results will be demonstrated across 16 scanner pairs, 5 foundation model spaces, and 3 cancer/tissue subtypes. Pathological Phenotypes Show Survival Groups (Clustering & Computational Pathology) This study leverages advanced clustering methods to identify pathological phenotypes associated with survival outcomes in ovarian cancer. Findings show that stronger predictive models concentrate attention on phenotypes most linked to survival differences. Adding clinical data in a multimodal approach further sharpens these predictions by reducing noise and enhancing focus on biologically relevant regions. Deep Visual Proteomics (Cell Segmentation) This project advances the deep visual proteomics pipeline by integrating modern vision AI techniques to detect, segment, and extract specific tissue regions such as cell populations. Pathological slides (H&E or immunofluorescence) are processed, and selected cells or regions are laser-extracted into 384 wells for protein analysis. The pipeline will be fully integrated with QuPath, enabling visual verification by pathologists at every step. This ensures transparency and control, bridging computational methods with clinical workflows. CODEX (Cell Segmentation, Cell Classification, Community Detection) We analyze CODEX multiplexed imaging data using established, state-of-the-art computational pipelines to detect immunological differences between sarcomous and carcinomous tumors. It integrates methods for segmentation (InstanSeg), cell typing (MAPS), and neighborhood analysis (CytoCommunity) to process large-scale whole-slide images. The workflow generates millions of cell-level features, enabling deep biological insights while minimizing the need for custom tool development. Each step remains visually verifiable, ensuring confidence in results and flexibility for refinement with pathologist feedback. The applicant will participate in grant writing, represent the laboratory at conferences, and provide project management for translational projects.