Postdoctoral Fellow, Multimodal Modeling

Biohub is the first large-scale initiative bringing frontier AI models, massive compute, and frontier experimental capabilities under one roof. We're building a general-purpose system to accelerate scientific discovery, integrating frontier AI models, biological foundation models, and lab capabilities, with the ultimate goal of curing disease. Our technology powers scientists around the world, translating AI capabilities into tools that accelerate research everywhere. ## The Team Our decoding inflammation team builds tools to enable precise molecular-level measurements of inflammation within human tissues in real time, and develop proactive, early interventions that can be deployed when inflammation — which underlies the most significant causes of death worldwide — first flares in the body. You can learn more about our work [here](https://biohub.org/inflammation/). Our team collaborates with three powerhouse universities - Northwestern University, the University of Chicago, and the University of Illinois Urbana-Champaign - to develop first-in-class technologies and make breakthroughs. **Our Vision** - Pursue large scientific challenges that cannot be pursued in conventional environments - Enable individual investigators to pursue their riskiest and most innovative ideas - Facilitate research by scientists and clinicians at our home institutions and beyond We are a team of passionate individuals powered by technology, guided by scientific research, and driven by collaboration, working toward a mission to cure or prevent all disease. ## The Opportunity The Chan Zuckerberg Biohub Chicago is seeking outstanding early-career scientists to join and participate in the launch of the Proteoform Spatial Biology Group by continuing their training as a Postdoctoral Fellow in Multimodal Modeling. The Proteoform Spatial Biology Group aims to uncover the spatiotemporal regulation of proteins and their unique molecular forms, proteoforms, in inflammation and autoimmunity. For this position, the ideal candidate is expected to have experience in working with multimodal and multiscale modeling of diverse datatypes across confocal microscopy images, mass spectrometry-based proteomics, phosphoproteomics, and/or interactomics. ## What You'll Do - Design and train self-supervised multimodal models that fuse confocal protein imaging, single-cell protein proximity networks, and mass spectrometry-based phosphoproteomics into shared representations, using objectives such as reconstruction and contrastive alignment (e.g., CLIP) - Work with graph-structured proximity-network data, collaborating on graph- and topology-aware modeling approaches, and - Leverage existing high-performing imaging models for feature extraction and inference, adapting them for co-embedding and building new image models where needed - Develop cross-modal alignment strategies relating surface organization to signaling and localization, and use the learned representations to model continuous cell-state structure and the features driving state transitions - Present findings internally and externally, and co-author publications ## What You'll Bring - Essential: - PhD in machine learning, computational biology, bioengineering, biophysics, or a related field - Experience applying deep learning to images, including use of pretrained vision models - Demonstrated experience building both supervised and unsupervised models - Experience with multimodal modeling or data fusion across heterogeneous data types - Experience with graph neural networks or other graph/network representation learning - Proficiency in Python and modern deep learning frameworks (e.g., PyTorch, JAX, or TensorFlow) - Nice to Have: - Experience with contrastive or self-supervised learning (e.g., CLIP) for multimodal data - Familiarity with topology-aware or higher-order modeling (e.g., simplicial or motif-based methods) - Background in proteomics or mass spectrometry data analysis - Experience with sequencing-based or s…