increase the impact of your scientific work - share the data

[zivy]

How we helped create an AI foundation model without knowing it

Publicly shared datasets generated as part of the development of the IBEX method (doi: 10.5281/zenodo.5244550) and research into multi-omic profiling of follicular lymphoma (doi: 10.17867/10000199) were recently used by research groups, unrelated to those involved in the original work, to create an AI foundation model for spatial proteomics. These datasets accounted for 38% of the training data. This new work is described in a pre-print aptly named, "A Foundation Model for Spatial Proteomics" (arxiv doi: 10.48550/arXiv.2506.03373).

By sharing imaging data acquired as part of your current research you too can enable others to develop tools that in turn will enable you to answer your scientific questions more quickly, accelerating the rate of scientific discovery - share the data!

Q&A

1. What is an AI foundation model?
   An AI foundation model is a machine learning model trained on extremely large datasets, usually for a task that can be formulated in an unsupervised manner. For example, learning to predict missing portions of an image by masking them in the original image. The model is trained so that the similarity between the predicted and known masked image patches is maximized. Once training is completed, the model is used as a foundation for specialized downstream tasks for which there are fewer datasets. The model described in the pre-print was specialized for cell phenotyping, artifact classification and more. For additional information on foundation models see wikipedia.

2. Where can I share my images?
   There are many options, both domain specific and generalist repositories. We recommend using domain specific when possible and have previously utilized:

   • The Image Data Resource.
   • The BioImage Archive.
   • The Zenodo generalist repository.

3. Where can I share my reagent information?
   You are already in the right place, The IBEX Imaging Knowledge-Base. All you need to do is follow these instructions.

4. Where can I share my custom 3D printed lab hardware?
   The NIH 3D model repository.

[radtkea]

Thanks, @zivy, for your great post and for encouraging others to share their imaging data. It's really rewarding to see the data I've generated helping to advance science! We are happy to answer questions and support your data sharing journey!