top of page

The Evolution of the Generative Cooperative Network: A Leap in Longevity Research

Today, we’re thrilled to share an update on the progress of our Generative Cooperative Network, a groundbreaking initiative that promises to redefine the landscape of longevity research.

The Genesis of the Generative Cooperative Network

Our journey began with the identification of four initial Generative Neural Network models, earmarked for integration into our expansive network. While these selections are preliminary and subject to refinement, they represent the cutting edge of AI-driven longevity research. The models in focus are:

  • CPGTransformer: processes methylation data to estimate physiological age and the pace of aging. Refinements are underway using a large dataset of 1300 CPG sites.

  • Progressing from the work carried out on the CPGTransformer, custom Neural Nets trained on multiple aging clocks are now being developed for incorporation into the GCN.

  • scGPT Transformer: Single Cell Generative Pretrained Transformer (scGPT) analyzes individual cells and genes using multi-omics data. Ongoing work focuses on refining gene mappings.

  • MethylNet and Beta VAE are specialized models using blood data to predict age.

Substantial progress is being made behind the scenes in the development of the ‘seed models’ that will form the basis of the GCN which enables unique generative models to self-organize and co-operate with each other to achieve the desired outcomes.

New Horizons in Age Estimation

A significant application of our network lies in the age estimator feature within the Longevity App. Preliminary tests using the VAEs have been promising, with a mean absolute error (MAE) of 3 to 3.1 years. This suggests that our age prediction algorithms can estimate age with an impressive accuracy within a 3-year range. For context, we’ve also established elastic net baselines, such as the revered Horvath and Hannum clocks, to serve as benchmarks.

CPGTransformer: Charting the Path Forward

The CPGTransformer, designed to establish causal relationships, offered insights into physiological age through use of the Horvath elastic regression technique. A standout feature is its ability to access the speed of aging from blood data, leveraging 1300 methylation sites. The model utilizes a plethora of data, including blood proteomic data and the TruDiagnostic dataset, to provide insights into real age, predicted age, and the predicted speed of aging.

Building upon the work undertaken with the CPGTransformer, the team has identified opportunities to improve on this method and are currently working on training custom neural networks on multiple aging clocks to combine their analytical power and improve age prediction accuracy.

The Marvel of scGPT: Single-Cell Generative Pretrained Transformer

Generative pre-trained models, especially transformers, have revolutionized fields like natural language processing. Drawing parallels between words in texts and genes in cells, the scGPT, a generative pre-trained transformer for over 10 million cells, is a testament to this evolution. It deciphers intricate biological insights into genes and cells and can be fine-tuned for a myriad of tasks.

Our data science and AI team is meticulously fine-tuning scGPT, diving deep into chromosome information positions, methylation expression, and more. Integrating heterogeneous datasets is vital for the network. Current efforts centre on incorporating omics data like methylation, metabolomic, and proteomic biomarkers from partners which will enhance model accuracy and create a unique and powerful tool.

Incorporating Multi-Omics Data

Our collaboration with TruDiagnostic has been instrumental in integrating omics data, encompassing methylation, metabolomic, and proteomic data, into our models. This synergy is poised to enhance the accuracy and depth of our predictions.

Foundation Models and Beyond

Beyond the aforementioned models, we’re excited to announce that our network will also have access to additional models including large language models which will be incorporated into the ensemble of agents.

Furthermore, we’re in the process of developing our own foundation models which promise to bring fresh perspectives to our network and assist with uncovering the mysteries of aging.

Infrastructure Investments

To support the demanding computations involved, the GCN has upgraded its server infrastructure gaining the parallel processing power needed as the network expands.

Ongoing Initiatives

Current initiatives aim to further develop the GCN:

  • Collaboration with partners like TruDiagnostic to access datasets.

  • Discussions with potential partners to augment capabilities.

  • Rigorous testing across dimensions like model accuracy, integration, security, and performance.

  • Exploring additional models that can be integrated into the framework.

The Road Ahead:

Our overarching plan is multi-faceted:

  1. Refine and develop the foundation models.

  2. Identify individuals who are aging faster or slower.

  3. Incorporate additional suitable models into the GCN ensemble.

  4. Integration of GCN into Longevity App

Final Thoughts:

The GCN represents significant progress in leveraging AI for biogerontology. The models create a foundation to analyse aging with increasing granularity. This can shed light on key questions like why some individuals age faster than others. The system also enables continuous improvement as new data emerges.

Ultimately, the knowledge generated by the GCN will help advance the broader mission of developing interventions to extend healthy human lifespan. The network remains a work in progress, but demonstrates the growing potential of AI in decoding and influencing the biology of aging.

In conclusion, the Generative Cooperative Network represents a monumental stride in longevity research. With your continued support, we’re confident in pushing the boundaries of what’s possible in the realm of longevity research. Stay tuned for more exciting updates!


bottom of page