The field of biogerontology has established itself through significant lines of research in recent decades. However, despite early breakthroughs, progress in understanding the aging process has been slow. To push the field forward, new methodologies and technologies are likely needed to unravel the complexity of aging. This meeting brought together leading scientists and innovators to explore some emerging approaches, presenting groundbreaking advancements in four key sessions, culminating in a panel discussion.
The first session focused on the use of artificial intelligence to advance our understanding of complex biological phenomena. Jackie Han from Peking University presented her latest work using their 3D facial map technology to predict age in two distinct populations: one from Ghana and another from China. Her findings revealed that while both populations age similarly along the depth axis of the 3D face, the African population exhibited slower aging on their ethnic-shared 3D facial aging clock. Although Ghana's average life expectancy is lower than that of China, her group found African-likeness immune-associated processes in Asians may alter facial appearance in a youthful-like way, for example the face looks tighter and more lifted. Finally, they found that genes associated with African-like features were linked to immune processes, such as neutrophil degranulation. Her group is now using this technology to study phenotypic biomarkers of aging to a broader understanding of human aging and its variability.
With regard to the hallmarks of aging, a problem that has been present in the field is the lack of specific biomarkers for senescent cells. Indra Heckenbach, from the University of Copenhagen, addressed this challenge through the use of deep learning to discriminate senescent cells based on nuclear morphology. A surprising application of this method was used on breast tissue to predict cancer. The group found that two of their trained models significantly correlated with the post-diagnosis of breast cancer, and when one of these models was combined with the Gail-score they were able to find a stronger odds ratio. The group as well found that through their model it was also possible to predict risk of development of cancer from benign breast disease.
Finally, with the rise of tools such as large language models (LLMs), Georg Fuellen from the Rostock University Medical Center in collaboration with Brian Kennedy from the National University of Singapore, presented novel insights on the use of these tools for creating AI-based longevity recommendations. The model needs to be effective at evaluating the interventions with an understanding of the patient (or the healthy subject asking for recommendations) in mind. Thus, the subject needs to provide a biomarker profile, which in turn must be considered appropriately by the LLM. In addition, the use of LLMs for designing effective longevity studies was discussed. In particular, asking GPT4o for longevity study designs, it returns standard designs centering around diet and exercise, as these best reflect the "short head" of the training data distribution. However, with some nudging, designs for, e.g., testing epigenetic rejuvenation interventions can be obtained. It may be worthwhile to start a "Leaderboard" of good designs, contributed and evaluated by humans and LLMs alike. This work emphasizes the importance of tailoring interventions based on patient profiles, as well as considering key factors such as study population and feasibility when designing longevity studies.