Maximize ROI and accelerate time-to-market by choosing the right Generative AI tools. This session offers a versatile evaluation formula for professionals, especially in pharma, to navigate complexities. We'll cover key factors like data compatibility, human-AI interaction, regulatory compliance, ROI, and technical debt mitigation. Join me to create a practical roadmap for GenAI tool selection and unlock unparalleled efficiency and success.

This talk will present how patient timelines constructed using healthcare-specific Large Language Models (LLMs) significantly enhance drug discovery efforts. Our case study demonstrates how applying LLMs to Electronic Health Records (EHRs) enables the creation of detailed oncology patient timelines. This involves extracting and synthesizing chemotherapy treatment data from diverse clinical documentation—ranging from primary care and oncology notes to discharge summaries, emergency department records, and pathology and radiology reports. These timelines offer a comprehensive, longitudinal view of patient treatments, providing critical insights into treatment patterns, response rates, and progression timelines that can inform drug discovery and development efforts.

This talk will highlight how generative AI can create dynamic and mathematical models for precision medicine, enabling the simulation of patient-specific responses. We will explore a platform that generates and scales these models efficiently, optimizing therapeutic predictions. The application of these AI-driven models promises to transform precision and quantitative medicine, improving personalized treatment and clinical decision-making.

AION Labs, a venture studio founded by leading pharma companies, AWS, and VCs, has launched nine generative AI start-ups focused on drug discovery. This talk will share lessons learned from building GenAI companies that integrate into pharma workflows, including real-world insights from startups specializing in antibody discovery, antibody optimization, molecular glue design, and small molecule generation. We’ll discuss the major challenges and strategies for overcoming barriers in developing and validating transformative AI technologies.

Scientific instruments now produce enormous volumes of data that exceed local processing capacity. Online analysis combined with AI/ML methods presents one way of dealing with such massive data streams, intelligently analyzing only interesting subsets or directing instruments to relevant areas of experimental space. We will explore how ML techniques enabled by the Globus platform can help life sciences researchers support distributed data management and computing pipelines to enable efficient data analysis and self-driving experiments. We will illustrate these capabilities using examples of antimicrobial peptide identification, x-ray ptychography, and small molecule screening.

This primer introduces biomedical researchers to Hugging Face tools, Large Language Models (LLMs), and Generative AI. It covers the basics of LLMs, their architecture, and applications in areas such as data extraction, literature analysis, and predictive modeling. Attendees will learn practical techniques like prompt engineering, model fine-tuning, and retrieval-augmented generation (RAG). The talk provides hands-on insights into leveraging Hugging Face’s tools for applying generative AI to transform biomedical research workflows.

Effective prompting is becoming a "must-have" skill that is needed to unlock the power of generative AI tools. This session offers a strategic approach for developing prompt upskilling programs, assessing their effectiveness, and preparing teams for the rapidly advancing landscape of generative AI technologies. Attendees will gain practical insights and engage with content and examples to empower their teams to better leverage these transformative tools.

Validating statements in publications with appropriate evidence remains a critical challenge across regulated industries and scientific publishing. Traditional manual verification methods are time-consuming, error-prone, and create bottlenecks in documentation workflows. This talk examines how generative AI technologies are transforming evidence verification and validation processes. We present a framework that leverages natural language processing and large language models to automatically extract statements of fact (claims), evaluate supporting references, and assess evidential strength. The system categorizes evidence from reference documents and evaluates how it supports the statements while also searching external databases for additional relevant literature when necessary. Testing across multiple document types demonstrates significant efficiency gains, with verification time reduced by 70-85% compared to manual methods. Beyond operational benefits, the approach improves consistency, enhances audit readiness, and enables the construction of organizational knowledge graphs connecting claims to their supporting evidence that results in an expanding knowledge base with ongoing use. This talk discusses implementation considerations, validation metrics, and ongoing challenges in applying these technologies. We conclude by examining future directions, including cross-domain inference capabilities and enhanced evidence quality assessment, while maintaining appropriate human oversight in high-stakes regulatory environments.