• Karen Sachs

This workshop on causal discovery in biomolecular networks is designed to introduce participants to the principles and methodologies for uncovering causal relationships in biological regulatory networks, from molecular datasets. As the field increasingly relies on large-scale data from genomic, transcriptomic, and proteomic studies, the ability to decipher causal, mechanistic information from these datasets is essential for expanding our understanding of complex biological systems, and for maximally extracting information from multimodal high dimensional datasets.

This workshop will cover foundational concepts of causal modeling, with a focus on causal discovery: inference of the mechanism or structure of the regulatory networks which gave rise to the omics data observed. Participants will gain a clear and firm grasp of the fundamental principles of causal inference, learn about a variety of methods being used in the field, become aware of potential pitfalls and concerns with the techniques, and gain hands-on experience with real datasets. We will discuss practical methods to incorporate literature derived databases such as INDRA and OMNIPATH, to integrate information from these prior knowledge networks into analyses of molecular datasets. In the practice session, we will apply causal discovery methods to elucidate mechanisms underlying biological processes.

• Danielle Stover

This tutorial will focus on the rapidly advancing field of multi-modal modeling for precision medicine, a critical area where artificial intelligence is reshaping the landscape of personalized healthcare. The session will delve into the integration of diverse data types—including genomics, imaging, electronic health records (EHRs), and patient-reported outcomes—to improve diagnosis, treatment, and prognosis in precision medicine.

The research theme centers around how to effectively use AI to combine these disparate data sources into a unified framework, providing deeper insights into patient-specific treatment responses and disease mechanisms. Attendees will learn about the latest advancements in machine learning models (e.g., transformers, graph neural networks) tailored for multimodal data fusion, as well as practical strategies for implementing these models in real-world medical settings.

The session is designed for participants including graduate students, postdoctoral researchers, bioinformatics professionals, clinical data scientists, and industry practitioners. It is especially suitable for those with a foundational understanding of machine learning and a keen interest in advancing their knowledge of AI applications in healthcare.

The tutorial’s timeliness lies in the ongoing explosion of healthcare data and the need for sophisticated AI approaches to manage and leverage this data effectively. The bioinformatics community stands at the forefront of this effort, and the tutorial aims to bridge the gap between cutting-edge AI techniques and their real-world application in precision medicine. This session will empower participants to develop models that can drive more personalized, effective, and efficient patient care.

• Sumaiya Iqbal

This tutorial introduces the Genomics 2 Proteins (G2P) portal, a bioinformatics tool that connects genetic variants with protein sequences and structures, enabling researchers to interpret genomic data in the context of protein function.

Participants will learn to integrate multi-omics datasets – genomics, proteomics, and structural biology – to investigate the structural implications of genetic mutations.

Hands-on sessions will guide participants in navigating G2P modules, mapping genetic variants to structural changes in proteins, visualizing protein features, and using APIs for advanced applications.

Case studies will showcase real-world applications, such as mapping clinical mutations and interpreting base-edited variants.

• Quincy Gu

Artificial Intelligence (AI) has profoundly impacted the field of digital pathology. One innovative AI-driven technology, virtual staining, can convert unstained tissues into stained ones or change one type of stain to another. In clinical diagnostics, hematoxylin and eosin (H&E) stained whole slide images (WSIs) serve as the gold standard for cancer diagnosis, often followed by special staining WSIs, such as immunohistochemistry (IHC) or multiplexed WSIs, to assess patients' genetic mutation status and aid in selecting personalized treatment options. However, the special staining process is time-consuming and requires adjacent tissue samples, making it challenging to directly apply AI models developed for cell segmentation and classification on H&E-stained WSIs to special stained WSIs. Virtual staining, which can digitally convert H&E stained WSIs into other specially stained formats (like IHC or multiplexing), provides an ideal solution for cell segmentation and classification tasks without needing physically stained samples, enhancing single-cell analysis. Due to time constraints and resource limitations, this tutorial will focus solely on deep learning-based methods for cell segmentation and single-cell analysis on multiplexed WSIs.