Cross Validation of AI Suggestions for Nonalcoholic Fatty Liver Disease

A friend got this diagnosis from a naturopath examining blood cells using a microscope. Visual inspection of blood lacks the degree of objectivity that I would prefer (AI imaging of the microscope slides is what I would like to see things evolve to). I went to Microbiome Prescription’s Medical Conditions with Microbiome Shifts from US National Library of Medicine and then look at the candidate suggestions for Nonalcoholic Fatty Liver Disease (nafld) Nonalcoholic .

For information about the Artificial Intelligence approach, see AI Generated Diet Suggestion for Medical Conditions. Note: This is not machine learning but old school fuzzy expert systems.

Out of curiosity, I did a few cross validations for the highest to take and highest to avoid. Everything was in agreement. The few suggested items that I checked were shown to have the desired impact on NAFLD.

The purpose of cross validation is to see how well the Artificial Intelligence Logic (and assumptions) are performing.

I decided to do a deeper cross validation because I found that the treatment literature was relatively abundant. REMEMBER the AI only knows the bacteria shifts and nothing about the diagnosis or treatments.

Items to Take

In priority order, the top items with AI Weight of 20 or more

Net ImpactLinkModifier
RIGHT plantarum (probiotics)
RIGHT starch  
RIGHT milk oligosaccharides (prebiotic, Holigos, Stachyose)
RIGHT bifidum (probiotics)
RIGHT (prebiotic)
RIGHT rhamnosus gg (probiotics)
n/a resistant maltodextrin  
n/a saccharin  
n/a fibre-rich macrobiotic ma-pi 2 diet  
RIGHT casei (probiotics)
RIGHT acidophilus (probiotics)
n/a Goji (berry,juice)  
RIGHT (prebiotic)  
WRONG (sodium chloride)  
RIGHT protein diet  
RIGHT wine  
n/a fasting  
RIGHT Root  
RIGHT (origanum vulgare, oil) |  
RIGHT (polyphenol)  
WRONG salt  
n/a merbromin  
RIGHT subtilis (probiotics)

Items to Avoid

Values with AI Weight of -14 or less

RIGHT carbohydrate diet  
n/a thiamine hydrochloride (vitamin B1)
n/a melatonin supplement
n/a camelina seed  
complex high animal protein diet  
RIGHT diet  
n/a mannooligosaccharide (prebiotic)  
n/a low fodmap diet  
n/a sodium stearoyl lactylate  
Complex high-protein diet  
RIGHT sugar diet  
RIGHT C (ascorbic acid)


There is a strong bias to publish positive results, hence finding many n/a in the to avoid list is expected. A study show an adverse effect is unlikely to see publication. We see the following ratios:

  • To Take: 22 Right to 2 Wrong, i.e. 92% correct
  • To Avoid: 10 Right to 2 Wrong i.e. 83% correct

This suggests that the n/a ones above are likely to be correct.

Leave a Reply

Your email address will not be published. Required fields are marked *