As AI and machine learning become core to biomedical research, the challenge is no longer whether models can be built—it’s whether organizations can operate them reliably, securely, and at a global scale. We will explore how cloud-native compute, scalable model-serving architectures, multimodal data platforms, and rigorous security and compliance frameworks come together to form an integrated, end-to-end capability for AI in biomedicine.

In this session, we will discuss a cloud-native AI platform that democratizes access to AI for business teams, scientists, and analysts across the organization. We will go deeper into how we unified data, domain knowledge, and secure model orchestration to create a platform where teams can rapidly prototype, validate, and operationalize AI applications—from natural-language analytics and workflow automation to decision-support copilots and RAG-powered knowledge systems. The talk will highlight the architectural and governance principles that ensure reliability, safety, and compliance in a regulated environment, while harnessing AI without deep machine learning expertise resulting in accelerated innovation, improvement in operational efficiency, and unlocking of tangible business value by making AI a practical, self-service capability across the organization.

Agentic AI can transform how enterprises operate—but only with the right foundation. This session explores what it means to be agentic AI-ready, covering control tower and responsible AI, agent identity/access and security, data ontology, and FinOps. We will introduce an agentic orchestration framework and clarify platform, data, application, and business agents, giving you practical guidance to design a scalable, secure, and value-driven agentic AI ecosystem.

AI has transformed protein structure prediction; yet many R&D teams still struggle with fragmented tools, inconsistent infrastructure, and steep learning curves. We present a unified platform that combines the best public, commercial, and proprietary AI models within a seamless hybrid-cloud environment. With intuitive visualization, robust API integration, and integrated access to diverse prediction methods, the platform enables scientists to move from raw sequence to actionable insights for therapeutic design and optimization, delivering results faster, more reliably, and at scale.

Large foundation model training demands access to diverse—and geographically distributed—computation and data resources. Coordinating access to these resources requires building container images, managing training jobs, monitoring job progress, and managing data and provenance across multiple storage systems and computing clusters. We will describe how cloud services on the Globus platform are being used to orchestrate this process and highlight use cases from computational biology and other scientific domains.

Scientific instruments generate vast datasets, yet making it usable for scientists and machine learning remains a challenge. This talk presents a cloud-native architecture for high-throughput ingestion of laboratory data, enabling timely access for analysis and model training. Drawing on real-world experience, we highlight design patterns that reduce latency, improve reliability, and make experimental data a first-class input to scientific discovery and AI development.

We present a systems-focused pipeline that converts population-scale biosensor-based actigraphy and PPG data (i.e., biosensor data) into reproducible, high-throughput training datasets for multimodal digital health foundation models. We implement a two-stage design: preprocessing and chunking the time series into binned window sizes using scalable EKS clusters, followed by distributed multi-GPU transformer training with a GPU-aware DataLoader and a time-series database to reduce I/O contention, and enable fast metadata-driven retrieval across multi-terabyte storage. In brief, this work emphasizes the infrastructure and data-engineering trade-offs, as well as the operational patterns required to convert siloed time series into reproducible, high-throughput training pipelines. 

Cloud platforms offer unprecedented opportunities to unify fragmented lab, clinical, and real-world data—but the challenge is turning this data into actionable AI insights. This panel brings together leaders from pharma, biotech, and digital health to discuss how cloud-native architectures, high-throughput data pipelines, and scalable storage solutions can enable AI/ML at scale, while maintaining governance, security, and regulatory compliance.

For 20 years, Trends from the Trenches has been Bio-IT World’s unscripted pulse check, offering candid, insider perspectives on what works, what fails, and what’s pure hype in scientific computing. As the field evolved, the session broadened to reflect the real operational challenges and breakthroughs shaping R&D. For 2026, the format evolves again: a focused, credibility-driven keynote paired with a community-powered unconference built from attendee input. The result is a forum for late-breaking insights, grounded realities, forward-looking perspectives, and practical solutions you won’t find in vendor decks, marketing summaries, or any LLM. It leaves participants energized by the collective intelligence in the room and inspired by the emerging possibilities shaping the future of life-science computing.

This presentation frames the industry’s next chapter by tracing how Trends from the Trenches has shaped digital R&D for two decades and by spotlighting the forces redefining scientific computing today: AI–HPC convergence, modality-driven compute, multimodal data, and rising expectations for speed, interoperability, and trust. Remarks set the foundation for a forward-looking exploration of where digital biology and computational innovation are heading next.

This session features a structured, participatory unconference built around topics sourced from Bio-IT event attendees during the conference week. A working group synthesizes all submissions Wednesday evening into a small set of high-value discussion themes. The facilitator guides the room through rapid-fire exchanges, micro-debates, and collaborative problem-solving focused on operational realities in AI, computing, data engineering, and scientific software. The goal is to surface patterns, stress-test ideas, and extract practical solutions emerging across the community—creating an annual, crowd-generated state-of-the-field snapshot that only this session can produce.