The BenevolentAI Data Diversity Initiative
AI-enabled drug discovery relies on data and algorithms to develop novel medicines, but the lack of diversity in most biomedical datasets and biases in the algorithms themselves risks exacerbating the already biased and inefficient drug discovery process. BenevolentAI’s Data Diversity Initiative aims to improve the representation of the whole population in biomedical data, and contribute to developing medicines that work for everyone.
Poor data diversity impacts patient care
Drug development is failing many patients with debilitating diseases, and one key reason for this is a lack of understanding of underlying disease biology. Failure to identify significant drivers of a disease is in part driven by the whole patient population not being represented in the data. With only partial information to reason across, scientists and the AI models being trained, run the risk of making biased decisions.
However, we are now seeing clear examples of biomedical analyses or findings that would have not been possible without addressing this diversity issue (for example in hypercholesterolemia and diabetes). This gives important and positive signals that by changing and challenging our practices in data diversity, we can improve drug discovery for the benefit of everyone, not just the few.
Better representation of the whole population in biomedical data for health equity
At BenevolentAI, data and knowledge are the foundations of our AI-enabled drug discovery approach. To this end, we are committed to providing and analysing diverse data that drives the development of better medicines for all patients.
Our Data Diversity Initiative aims to better represent the whole patient population, and to understand and address biases in models. Our overarching goal is to expand the understanding of human biology to make the drug discovery process more equitable for all patients.
Proud partner of the link23 Initiative run by Genomics England
Increasing the diversity of data and developing more inclusive models is a collaborative industry effort.
We are proud to be partnering with link23, a global open source and community-driven initiative making genomic tools that work for all patients. With a diverse community of contributors and partners the goal is to curate, build, implement and scale practical solutions to ensure genomics research is as equitable as possible.
By equipping individuals with practical tools link23 aims to lessen the immediate effect of poor diversity in genomics, and create the conditions to change behaviours and processes for good.
