I get a few emails asking about apparent contradictions from Microbiome Prescription AI Engine. With the new “Just give me suggestions!” option, they have been reduced — this post explains the root issue of these conflicts. It does not solve the issue — only time and a lot more published studies will resolve it.
Data From Studies
The data entry attempts to keep true to what is actually reported in the studies on the US National Library of Medicine. A simple example, the following are subjects of different studies:
- Grape seed extract
- Grape polyphenols
- Grape Fiber
- Red Wine
- Red wine polyphenols
It is human nature to try to consolidate information. There are numerous historic examples where such consolidation failed. A simple one is that all antibiotics are the same. In some nations, antibiotics are over the counter — so if you have an infection like tuberculosis, some people would walk in and buy the cheapest antibiotics expecting it to work. It is no surprise that it would have no effect. “They are almost the same” is not an approach that I subscribe to.
The second aspect of studies is that they report on different levels of the bacteria hierarchy, and rarely report on multiple levels. In some cases, the report is only on the highest levels.
When looking at your sample. You may be high at the genus level but normal in the family and order levels. So data from studies about what is impacted at the species, family, order or class level may be ignored. Some people, including medical practitioners, may consolidate this information and after reading one study that mentions a genus apply that to all related species, family, order or class — occasionally to the phylum level. The human tendency to consolidate information strikes again.
Last, the study may have been done on people with a specific condition or type of diet. Diet is often based on location in the world: India (many eat no meat), America (lots of junk food), China (high rice content). The shifts seen with some modifiers with different conditions or diets are different and sometimes in opposite directions. We may not get consistent studies. The human tendency to consolidate information by deeming everyone to be the same strikes again.
Below we have seven similar items — but very different information on what the studies report on.
|Sum of Count||Column Labels|
|grape seed extract||1||8||1|
|red wine polyphenols||5||1||3|
While they are similar, there are difference between them that may be significant. One contains sugars, another contains fiber, another contains alcohol — these minor differences can alter different bacteria significantly.
This is why you may see apparent contradictions in suggestions. We a have a mixture of information about your microbiome and each of the above is a sieve of different mesh and fabric.
I choose not to consolidate information. I keep the information as reported. A medical practitioners is not able to keep all of this information in their head. They will proceed to consolidate, and consolidate and simplify – the art of medicine. The Artificial Intelligence Engine is the result of processing over 41 Gigabytes of information — that is likely far more than any medical practitioners had actually read in their career. Is the detail needed? That’s a personal judgement. I prefer to have it. Using AI, I can work with all of the information available without needing to consolidate or simplify.
When there is a contradiction – which should I choose? My standing answer, is avoid the substance. We do not know with confidence what the outcome may be. On the other hand, I often seen suggestions reinforcing each other — for example Positive: Gluten Free, Negative: Wheats.
We do not want to take gambles with our health — keep to where there is consensus.
“Just give me suggestions”
This picks common items often seen on the internet for a variety of conditions. When there are “similar” items, the one with the most data will usually be selected. It gives higher confidence. These choices are evolving as I review the data.
The intent is avoid showing contradictions, and work where we have the most data. It’s a simple best path forward.