Multi-dimensional Blood Testing and A.I.

I suggested long ago that sufficiently-comprehensive blood tests could effectively predict a person’s risk of developing a broad array of different diseases. We would use artificial intelligence to find patterns of varying concentrations of blood proteins to predict and/or diagnose disease. Someone much better funded than me has a newly developed platform called SomaScan which can scan five thousand individual proteins from a single blood sample.

In a new study testing the efficacy of predicting 11 different health indicators using these protein expression patterns some models were much more effective than others, such as the protein expression model predicting percentage body fat. The cardiovascular risk model was cited as only modestly predictive, however, the researchers do suggest the protein-pattern-based system is generally more convenient, and cheaper, than many traditional tests currently available for evaluating health conditions.

The study in Nature Medicine was funded by SomaLogic which owns SomaScan, so grain of salt, people. But it’s exciting to see that someone is actually looking into what I feel will be the method of the future for maximizing health…also, the study used ~85 million protein measurements in 16,894 participants, which is a pretty damn good sample size.  Plenty of data there for an A.I. to examine for hidden relationships.

Homework:

Plasma protein patterns as comprehensive indicators of health, Nature Medicine, Stephen A. Williams, Mika Kivimaki, Claudia Langenberg, Aroon D. Hingorani, J. P. Casas, Claude Bouchard, Christian Jonasson, Mark A. Sarzynski, Martin J. Shipley, Leigh Alexander, Jessica Ash, Tim Bauer, Jessica Chadwick, Gargi Datta, Robert Kirk DeLisle, Yolanda Hagar, Michael Hinterberg, Rachel Ostroff, Sophie Weiss, Peter Ganz & Nicholas J. Wareham