Day Two

• Leverage OASIS construction and liver-focused cell painting data to link complex in vitro phenotypes with known human safety outcomes
• Improve confidence in translation by comparing mechanistic in vitro signatures against clinical liver toxicity patterns rather than relying on single endpoints
• Strengthen early safety decisions by positioning high-content liver NAMs as a bridge between discovery screening and human risk understanding

• Appreciate the history of approved Bispecifics and the clinical landscape
• Understand regulatory and industry guidance as well as challenges that come with selecting starting dose for bispecifics in oncology indications
• Learn how integrated approaches can enable a safe starting dose while potentially accelerating dose escalation

• Review the ADC landscape to understand translatability of ADC platform toxicities
• Integrate off-target and on-target findings
• Therapeutic window prediction by selecting appropriate preclinical efficacy models and dosing schedules

• What does “validated enough” mean in your organisation?
• How do validation expectations differ between discovery screening, candidate selection, and IND-enabling decisions?
• How are you selecting positive and negative control compounds, and where do gaps remain?
• What trade-offs exist between scientific relevance, historical data availability, and program timelines?
• How do you demonstrate human relevance when clinical anchoring data are limited or absent?
• What surrogate endpoints (e.g. biomarkers, exposure-response, phenotypic signatures) are most persuasive internally and externally?
• How do regulatory expectations influence your internal validation bar?
• Are FDA/EMA interactions raising standards, or creating flexibility around context-ofuse?
• Where do cross-industry frameworks (e.g. HESI, IQ MPS, consortia efforts) genuinely help, and where do they fall short?

• Understand how model performance is limited by the training dataset and why incomplete or narrow data subsets lead to reduced predictivity and continued reliance on in vivo studies
• Learn practical approaches from pharma using integrated in vitro and in silico data for screening while ensuring IND-enabling decisions remain grounded in translatable evidence
• Explore collaboration strategies between toxicologists and AI scientists to generate scalable datasets and build models that support future first-in-human decision making

• Overview of VICT3R and Virtual Control Groups (VCGs): rationale in the context of current global regulatory changes, and how using statistical methods and AI on historical control data can reduce animal use by up to 25%, supporting ethical innovation in toxicology
• Qualification strategy to demonstrate scientific validity, robustness, and reproducibility of studies including VCGs – examples from reanalysed legacy studies and ongoing prospective studies with industry partners and CROs
• Regulatory engagement and leveraging global partnerships to translate VCG qualification results into scientific opinions and internationally harmonised guidance documents – overview of our interactions with EMA, FDA, and OECD, and the role of the Scientific and Regulatory Advisory Board

• Understand how immunogenicity, stickiness, and other liabilities frequently cannot be solved in isolation without worsening other properties
• Explore how In silico predictors can be tuned to not only guide efforts to verify predicted risks but also unify potential solutions in sequence space
• Discover a unified framework for evaluating and allocating effort to solving developmental liabilities, including toxicity, can significantly accelerate development efforts

AI is rapidly advancing toxicology research, but its true value lies in augmenting human expertise, improving data usage, and accelerating learning curves for new modalities. This facilitated discussion will explore where AI produces real gains today and what still needs validation to earn regulatory and industry trust.

• Closing information gaps for novel modalities. Strategies to make AI useful when clinical and preclinical data for LNPs, cell therapies, and gene therapies are still limited.
• Combining AI + TSA + expert judgment. How preliminary target-associated literature insights and mechanistic toxicology guide model development where in vivo datasets don’t yet exist.
• From animal prediction to human relevance. Debating current limits of LLMs and ML tools in forecasting outcomes across species and modalities.
• Scaling confidence through validation & collaboration. What data standards, regulatory engagement, and partnerships between AI developers and toxicologists are needed for broader adoption.