Humanity Inc. Enhances Biomedical Datasets with Generative AI

Humanity Inc., an application which supports individuals to extend their health span, partnered with Neurons Lab to improve their biomedical datasets using generative AI technology.

Partner Overview

Humanity Inc., an application which supports individuals to extend their health span, partnered with Neurons Lab to improve their biomedical datasets using generative AI technology.

What if there were a way to know which actions to take to slow down ageing and increase your health span?

UK-based Humanity’s mission is to extend the health span of humanity. The consumer subscription tracking service enables users to find out what actions they can take to slow down their aging process as well as to reduce the risk of disease by means of tracking key digital, blood and genetic biomarkers.

Business Challenges

Humanity uses ML /AI to determine the biological age of users and generate lifestyle recommendations based on the user’s historical biomedical data. For training ML / AI models, Humanity needed massive amounts of data which represented real users.

However, due to significant restrictions on the distribution and use of personal datasets, Humanity needed to truly anonymize their biomedical data. The quality and suitability for training models also needed to be augmented for optimum performance, improving the quality and filling missing data.

To overcome this challenge, the solution would need to involve the generation of synthetic data sets with high similarity to the original data.

Since there are various generative models which could be a potential fit, the team needed to investigate, identify and implement the best approach or combination of approaches in order to produce a high-quality dataset that is not burdened by regulatory restrictions. This data set would also lead to significant improvement in development and business processes.

Solution

The team split the scope of work into three sections: Discovery, Feasibility study and MVP development. Using a managed capacity model and scrum framework, the team, made up of Principal Applied Scientist, ML Delivery Manager, ML Engineer and Cloud Architect followed the three step process to produce a working MVP in four sprints in a timeline of 8 weeks.

Implementing the following Generative AI models: Variational Auto-Encoder (VAE), Generative adversarial network (GAN), VAE-GAN hybrid, and their other modifications, they were able to generate synthetic data sets that closely resemble original user data. In this way, they successfully leveraged synthetic datasets to enhance the performance of AI models while fully complying with data privacy regulations.

They also employed Amazon SageMaker’s comprehensive suite of machine learning services for easier deployment and management of generative models – Neurons Lab Container Accelerator integrated with Humanity data lake and Amazon Sagemaker. Incorporating the Neurons Lab Container Accelerator into our infrastructure has significantly enhanced the ability to deploy applications seamlessly to our Amazon ECS cluster. Leveraging this technology has brought about several key advantages for our team and partners.

• Firstly, the true reusability facilitated by AWS CDK has streamlined our development process, allowing us to build upon existing components and accelerate application development.
• Secondly, with well-established DevOps practices in place, our speed to market has significantly improved, enabling us to quickly respond to evolving market demands. Moreover, the cost efficiency achieved through cloud-native workloads and the AWS Gravitom ARM Chip has not only optimized our resource utilization but also translated into significant cost savings.
• Lastly, the accelerated containerization capabilities have drastically reduced the time spent on infrastructure setup and configuration, enabling our development teams to focus on what matters most – delivering high-quality applications efficiently.

Results

By the end of the 8 week project, the team successfully created a fully anonymized synthetic dataset, eliminating data privacy concerns.

In addition, by fixing errors and outliers in existing data and augmenting missing data, the team also enhanced dataset quality,  leading to improved model training performance.

Neurons Lab were a great team to work with. They helped us put together the proposal for this challenging innovative project, and always acted with good grace and insight as we encountered and overcame difficulties. All the problems are not solved, but we have a much better handle on the issues and potential solutions. Thanks Lilia, Artem and team.

Peter Joshi Chief Scientific Officer | Humanity Inc.