Current Bio3 Seminar Series

Abstract: The immune-oncology trial with delayed treatment effect violates the PH assumption and presents unique challenges in using the standard log-rank test for trial design and data analysis. To overcome the difficulties and better understand the magnitude of treatment effect under delayed treatment effect model, we developed a data-adaptive design for cancer immunotherapy trials. Using the proposed data-adaptive approach could gain remarkably study power compared to the conventional method when the delayed treatment effect exists. Another main advantage of proposed data-adaptive approach allows for inference on a number of interpretable and useful measures of treatment efficacy.

Location: Online via Zoom

Abstract: Digital health technologies (DHT), strengthened by artificial intelligence (AI) driven analytics, are becoming central to advancing mobility, healthy aging, and sleep health. Human in the loop (HITL) approaches that leverage participants’ own smart devices—such as smartphones, wearables, and connected sensors, offer a practical and patient‑friendly way to capture real‑world data. Using familiar smart devices can enhance engagement, reduce burden, and generate more naturalistic insights into daily functioning, movement patterns, and sleep behaviors.

However, integrating personal smart devices into research introduces technical, operational, regulatory, and ethical challenges. This paper provides guidance for researchers, clinicians, and digital health innovators designing human‑centric studies that combine participant‑owned smart devices with AI‑enabled data streams to assess mobility, aging‑related outcomes, and sleep health. Key considerations include stakeholder engagement, inclusive and accessible study design, selection of meaningful outcomes and validated technologies, robust data privacy and monitoring strategies, and statistical and regulatory requirements for heterogeneous real‑world data. Use-case examples are provided. 

Together, these considerations offer a roadmap for incorporating human‑centricity via smart‑device, and AI‑enabled digital health methods. Such work may help enhance quality of life (QoL), improve mobility, support healthy aging, and advance sleep research.

Location: Online via Zoom

Abstract: Single-cell omics technologies are now widely used to study cellular heterogeneity and dynamics in biomedical research. When cells sampled from a population represent different states along a continuous biological process, pseudotime and trajectory inference methods are commonly applied to reconstruct the underlying progression from transcriptomic variation. However, most existing trajectory methods are designed for single-sample analyses and do not naturally extend to comparisons across multiple samples or to the analysis of longitudinal data. In addition, many approaches rely on heuristic assumptions and have limited ability to infer directionality or quantify uncertainty in dynamic processes. In this talk, I will present our recent work addressing these limitations. First, I will introduce a statistical machine learning method for inferring and comparing cellular trajectories across multiple samples, enabling principled cross-condition and cross-subject comparisons. I will then describe a model-based approach for analyzing dynamic cellular processes using longitudinal single-cell immune profiling data, which explicitly captures state transitions, directionality, and temporal evolution. Together, these tools provide a rigorous statistical framework for studying dynamic cellular processes in complex single-cell datasets.

Location: Building D, Warwick Evans Conference Room

Abstract: The rapid growth of publicly available, large-scale omics datasets across diverse diseases and species is creating unprecedented opportunities for data-driven biomedical discovery and precision medicine. Concurrently, continued breakthroughs in AI are transforming scientific data analysis and knowledge discovery. The convergence of large-scale omics data and AI holds great potential to automate and accelerate biomedical discovery. However, AI for omics-driven biomedical research remains in its early stages, presenting fundamental theoretical and practical challenges. This talk presents a novel multi-omics data representation unifying textual biomedical knowledge, multi-omics data, and signaling networks, along with novel LLM and graph AI models to identify key disease signaling targets and pathways. Additionally, it showcases multi-agent AI systems that automate omics-data analysis and hypothesis generation to accelerate biomedical discovery, concluding with challenges and future directions in the field.