Transformer-based aging clock provides insights into aging

One clock to rule them all – Insilico’s new aging clock hopes to generate a slew of precious therapeutic targets.

Clinical stage generative AI-driven drug discovery company Insilico Medicine has today published a paper on a new multimodal transformer-based aging clock; the new clock is capable of processing diverse data sets and providing insights into biomarkers for aging, mapping them to genes relevant to both aging and disease, and discovering new therapeutic targets to slow or reverse both aging and aging-related diseases.

Insilico calls the aging clock Precious1GPT, in a nod to the powerful “One Ring” in Tolkien’s Lord of the Rings; the findings have been published in the journal Aging.

Longevity.Technology: Insilico has been at the forefront of both generative AI and aging research, and has been publishing studies on biomarkers of aging using advanced bioinformatics since 2014. Later, the company trained deep neural networks (DNNs) on human “multi-omics” longitudinal data and retrained them on diseases to develop its end-to-end Pharma.AI platform for target discovery, drug design, and clinical trial prediction.

Transformer-based aging clock provides insights into aging

“We have long used DNNs to better understand human disease and aging biology,” says Alex Zhavoronkov, PhD, founder and CEO of Insilico Medicine and the study’s corresponding author.

“Now, with great advancements in generative AI capabilities, including AI-based transformers, we are able to further accelerate this process to make an aging clock that can not only identify where aging and disease intersect, but connect that information to actionable therapeutic targets.”

Transformer-based neural networks have only recently become available – one does not simply create a new network. First, scientists pretrain algorithms on unlabeled data, and then they further refine those algorithms with smaller sets of labeled data. Multimodal transformer models can process diverse data types, including genomic, proteomic, microscopy, computational chemistry, and clinical imaging data.

In the study, a team of researchers used a method called Precious1GPT that involves a multimodal transformer-based regressor trained on diverse data – including RNA sequencing and epigenetics methylation – for age prediction, and identified genes most relevant to both aging and diseases. The multimodal transformer was able to predict biological age and distinguish between disease and control samples. Scientists then fed the gene lists into Insilico’s PandaOmics target identification platform, and discovered targets highly associated with both aging and four age-related diseases – idiopathic pulmonary fibrosis, chronic obstructive pulmonary disease, Parkinson’s disease, and heart failure. Both APLNR and IL23R emerged as potential targets for the delay and treatment of age-related diseases.

“Applying generative biology to aging and disease using a transformer-based approach gives us new insight into how these complex biological processes interact, and clues as to how we can slow or reverse their progress through new therapeutic approaches,” says Frank Pun, PhD, Head of Insilico’s Hong Kong Office and co-author of the study.

Insilico scientists plan to further develop this approach by applying it to larger, proprietary disease-specific datasets and validating their findings through lab experiments. Ultimately, they hope to use this technique to better understand the molecular mechanisms of aging and develop new interventions for both aging and aging-related diseases.