It's been about three years since NBT began using supervised machine learning to predict the results of more expensive or unattainable biomedical tests. With our bloodsmart.ai software, we can forecast infections and inflammation, xenobiotic and heavy metal toxicity, and metabolic health indicators like fatty liver and elevated insulin - all without directly testing these markers. As a result, we've dramatically shifted our clinical work away from direct testing, instead focusing on basic blood chemistry and supervised machine learning to guide decision making. It's one of the things I'm proudest of building.

Sometimes I get asked how bloodsmart.ai compares to other blood chemistry programs. I used the other programs for years before coding my own, and rather than ML, they use what I call "hand-rolled algorithms." For example, if alkaline phosphatase is low, then it must be a zinc deficiency. Unfortunately, biology is way more complicated than that, and supplementing with zinc with just one indicator never helps.

On this podcast, my Scientific Director Megan Hall and I are discussing how to interpret the forecast on a bloodsmart.ai report and how we use the results in our work with clients. We talk a little about how the algorithms work under the hood and how we know the forecasts have predictive value. We also explain what might be going on when the forecasts don't match direct testing.

To get the most out of this podcast, be sure to follow along with Megan's outline.

Here's the outline of this podcast with Megan Hall:

[00:04:39] bloodsmart.ai software.

[00:04:47] Supervised machine learning.

[00:06:36] Pain as the amazing protectometer; Video: Pain, the brain and your amazing protectometer - Lorimer Moseley.

[00:08:25] Karl Friston.

[00:09:38] eLife podcast and eLife Journal.

[00:10:06] Machine learning in embryology: Bormann, Charles L., et al. "Performance of a deep learning based neural network in the selection of human blastocysts for implantation." Elife 9 (2020): e55301.

[00:12:16] Machine learning for identifying prostate cancer: Hood, Simon P., et al. "Identifying prostate cancer and its clinical risk in asymptomatic men using machine learning of high dimensional peripheral blood flow cytometric natural killer cell subset phenotyping data." Elife 9 (2020): e50936.

[00:13:18] Podcast: How to Interpret Your White Blood Cell Count with Megan Hall.

[00:14:38] Paper: Wood, Thomas R., et al. "An interpretable machine learning model of biological age." F1000Research 8.17 (2019): 17.

[00:14:53] Podcast: How to Measure Your Biological Age, with Megan Hall.

[00:15:24] How do we know the models have skill? Article: A Gentle Introduction to k-fold Cross-Validation.

[00:17:40] What the forecasts are and what they're not.

[00:19:18] A "cloudy crystal ball".

[00:23:21] Using bloodsmart.ai forecasts in clinical practice.

[00:24:25] Book: How to Decide: Simple Tools for Making Better Choices, by Annie Duke.

[00:26:17] The "Archer's Mindset": The value of taking aim.

[00:28:09] Podcast: Environmental Pollutants and the Gut Microbiome, with Jodi Flaws, PhD.

[00:28:45] Article: How to do better at darts and life.

[00:32:33] Health history and symptoms; Health Assessment Questionnaire (HAQ) (example).

[00:35:30] 7 minute analysis.

[00:36:53] bloodsmart.ai bar chart (example).

[00:37:56] Food journaling.

[00:42:27] Environmental Working Group's Skin Deep and Healthy Living App; Think Dirty Shop Clean App.

[00:43:03] Podcast: Air Pollution Is a Cause of Endothelial Injury, Systemic Inflammation and Cardiovascular Disease, with Arden Pope, PhD.

[00:44:23] Titanium bottle kickstarter: Keego.

[00:46:04] Discrepancies between forecast and directly measured marker.

[00:48:42] Forecasts that tend to be seen together.

[00:53:34] Forecast detail view (example).

[00:55:30] Josh Turknett's 4-Quadrant Model.

[00:58:22] Podcast: How to Win at Angry Birds: The Ancestral Paradigm for a Therapeutic Revolution, with Josh Turknett, MD.

[01:01:38] Book a free 15-minute starter session.