The Journey Begins with your microbiome

Thanks for joining me!

This is a companion site to the analysis site at:

Most of the content was originally posted on with the pages on the left being a restructuring of selected posts from over a thousand posts on that site.

Also a PodCast:

Recommended Site For Testing

If you have ME/CFS or other financially disastrous condition, there is always a nasty cost factor for testing. My usual recommendation is for the cheapest, high quality provider that provides information for upload to my analysis site. Some sites provide a mountain more of information — but the benefit from that extra information is almost nothing (and it adds $$$$ and complexity).

  • is shutting down. This had been my personal usual site because using a variety of techniques, the cost was $25/sample. Don’t order from there.
  • (EU based) is an excellent buy using our discount code [MICRO]. They have also automated data transfer to our analysis site.
  • Thryve is what I am starting to use. Their reports may be processed here for independent suggestions. I would also recommend

Caveat Lector: Labs and This site

When this site was started, there was one dominant player in retail-provider: uBiome. In June 2018, the first ThryveInside sample was uploaded, A year later, in May 2019, the first American Gut sample. A year later, in July 2020, BiomeSight started rolling in significant numbers — for 10 months, BiomeSight was the most frequent upload type every month. At present, I support 8 upload types and provide an API for any lab that wishes to do a direct transfer. BiomeSight lead the way here. Statistics are here for those interested.

In an early post, The taxonomy nightmare before Christmas…, The quote below says it all!

Standards seekers put the human microbiome in their sights, 2019

My #1 Measuring Stick

The first three labs, uBiome, Thryve and American Gut, all used the NCBI Bacteria Taxonomy systems. These are number and thus easy to store in the database and economic to do analysis on. This is a critical foundation. There are problems using names, because names change overtime. One bacteria has 237 different names. As illustrated below — same bacteria was discovered by many different people. Each person gave it a name and published papers using that name. In time (especially with DNA techniques) it was realized that they were all the same!!

NCBI is an unique identifier just like social security number is for American. Unfortunately, Canadians have SIN numbers. Other nations have Person Numbers. The same thing has happened with lab equipment. The problem is matching identities. With non-Americans in the US, some are issued TIN numbers (and thus we are good for US identity), others do not have TIN numbers. A person is like a bacteria.

Case Studies With Microba and BiomeSight

Microba does not use NCBI numbers. Microba uses the Genome Taxonomy Database (GTDB for taxonomic classification. The question arises, who attempts the mapping of the GTDB identifiers to NSBI — Microba or MicrobiomePrescription or no-one?

With cooperation from them (namely, they provided a reasonably complete list of the GTDB identifiers that they used), I was able to create a mapping table between those names and NCBI numbers that was not 100%, but sufficient to give meaningful results.

With, they added the numbers to their database. I always prefer the lab to take ownership of the mapping – there can be many nuisances specific to the lab equipment that they are using.

Popular Medical Tests that cannot be added to the data

There are two main reasons that these cannot be added:

  • They only measure selected bacteria (see below)
  • Their unit of measure is different. One counts the number of hex nut in a mixture of 1000 nuts; the other counts the number of packages of hex nuts (with a different number of nuts per package) in a carton of nuts. They are simply too different.
Lab NameBacteria Reported
Bioscreen (cfu/gm)17
Biovis Microbiome Plus (cfu/g)40
Diagnostic Solution GI-Map (cfu/gm)34
GanzImmun Diagnostic A6 (cfu/gm)72
GanzImmun Diagnostics AG Befundbericht25
Genova Gi Effects (cfu/g)28
Genova Parasitology (cfu/g)7
InVitaLab (cfu/gm)23
Kyber Kompakt (cfu/g)11
Medivere: Darm Mikrobiom Stuhltest (16s limited)16
Medivere: Darn Magen Diagnostik (16s Limited)16
Medivere: Gesundsheitscheck Darm (16s Limited)17
Metagenomics Stool (De Meirleir) (16s Limited)53
Smart Gut (ubiome 16s – Limited Taxonomy)23
Verisana (cfu/ml) aka (kbe/ml)11
Viome (No objective measures)29

For these test, users must transcribe whether the test indicated too high(↑) or too low (↓) levels. I give the ability to indicate how much…

How the labs represents varies greatly. Their units are not compatible.

Suggestions are based on these rough values and uses the same logic. A key limitation is that their normal ranges are likely computed assuming a bell curve and not Kaltoft-Moltrup Ranges. You may be acting on items that are in the typical ranges seen.

Issue of Missing Hierarchical Layers

If you look at “My Biome View” on Microbiome Prescription, you will see the hierarchy (per NCBI). Most labs do not give the full hierarchy in their reports. Often they will skip layers. The clearest example is Microba. They provide information in only 4 files.

But when this upload is viewed, you see all of the levels!

My Biome View

A more extreme example is the CosmosID’s PDF files, where they only list the species and strains!

The user who submitted this would see the following My Biome View…

Microbiome Prescription “completes” the data by summing up each level into the level above if missing. So I sum the count of all of the species in a genus to get the genus count if it was missing from the upload. There is an unfortunate gotcha. you may have 8000 in a genus and the sum of the species is 6000. If the lab provided the genus count, then we are good — no need to create a record with 6000. If we must create this level, then we are missing 2000 and higher levels are underreporting!.

This issue is also seen in some lab results. They scale the numbers so that the species that they report adds up to the count for the genus. What they do not report on is dropped from all of the parent levels.

When you use the Krona Chart, if there are no “unknown section” the0n this ignoring the not identified is a possible issue with the lab results. You can also do this on the My Biome View by comparing the numbers of the parent to the sum of the children – if they always match, then assume that the not identified are ignored.

Illustrates when the not identified is shown on a Krona Chart

Inconsistent Numbers

Above we have the case of the genus count being more than the sum of it’s species. This is a good state, because the numbers are more accurate. We have the unidentified bacteria being identified as least at the genus level.

I have also found cases where the sum of the species exceeds the genus. This can legitimately happen when alternative hierarchies are used. It becomes a problem when we attempt to keep everything in one hierarchy (“There can only be one!”)

Meme: "There can be only one" - All Templates -
From TV Series Highlander.

As a result, if the sum of the species (using NCBI hierarch) exceeds the genus, then we update the genus number for consistency (if we do not do that, then Krona charts can look bizarre — which a user emailed about).

Bottom Line

“Different strokes for different folks” is the problem. In accepting data from 9 different sources, I need to harmonize. The key that I play in is NCBI. This is a huge benefit because it is used with KEGG: Kyoto Encyclopedia of Genes and Genomes, which really enhances analysis.

Right Solution

It is simple, the labs should add to their websites equivalent pages seen on Microbiome Prescription — but only using their lab results. If their staff lacks the skills, I am a professional developer and can be contracted to do a lot of the backend coding (at my usual commercial rates ).

If you wish to be pro-active.

  • Verify that every bacteria shown on my biome view is shown on the lab results page. If it is not, they are skipping elements of the hierarchy
  • Verify that the count agrees, if not look at what is added up
  • Contact the provider and ask for automatic transfer to be implements. Code wise it is very simple, a few hours of work at most for most developers. What is needed is documented here, including a test site!

I cannot fix the root issue — inconsistent data. You are their customers and by being vocal, you can make a difference. If the upload is correct and complete — I make no modifications, it is only for problematic uploads.

Net Benefit Estimates

This actually exposes some of the challenges in estimating what a gut modifier will do. To illustrate it look at the example below. Most substances have mixed results, what you see listed is not unusual.

There are many ways of computing a net Benefit:

  • Add up those that move in the right direction, subtract those in the wrong, ignore the amount –>Category
  • As above but multiply by the number of bacteria away from the high or log –> Amount
  • Instead of the raw numbers, use log(number), borrowing from logistic regression and other concepts –> log(amount)
  • For each of the above we could also add in the confidence. This is based on the number of studies and the degree of agreement between studies (a.k.a. Fuzzy Logic factor) –> Log * Confidence
Bacteria – Shift – Amount (log)Substance (Confidence)CategoryAmountlog(Amount)Log * Confidence
A High by 1000 (3)Decrease (0.4)1100031.2
B High by 100(2)Decrease (0.8)110021.6
C High by 10 (1)Increase (1.2)-1–10-1-1.2
D Low by 1000 (3)Decrease (0.4)-1-1000-3-1.2
E Low by 100 (2)Increase(0.9)110021.8
F Low by 10 (1)Increase(1.2)11011.2
G High by 10,000 (4)Decreaseno info
H High by 100,000 (5)Increases (0.3)-1-100,000-5-1.5

Above we see two calculations suggesting that it will benefit, and two calculation suggesting it will not benefit. SCREAM!!!

The ideal is that all will be in agreement, if not, your philosophy or model ends up making the call. For my suggestions algorithms I use something similar to Log * Confidence.

Video on this post

You can roll your own…

All of the studies and sources on how some substance modifies different bacteria is shown on the site. The distributions are shown on the site. You can do your own data extract and apply your own logic and algorithms.

Warning: Suggestions from Labs and some Medical Professionals

Often I have found that lab gives suggestions. The better ones will give the medical citations behind the suggestions. Typically this is a single citation, often 30 years old study. In some cases, I have found later studies having opposite results — they have not keep current with all of the studies 🙁 OR they do not want to have to deal with contradictory results (thus grab one and use it for suggestions).

This also applies to some medical professionals. They keep their suggestions in their brains, usually reducing it to a single fact (like the labs). Keeping current is time consuming.

Strains strain our knowledge of probiotics

In an earlier post, I had illustrate the problem of whether L.Reuteri produced histamine. The answer is “Not sufficient information to answer” — why is shown below. It depends on which strain you have! The source (human/not human) is not sufficient. In general, the probiotic species is insufficient to answer the histamine question.

From prediction to function using evolutionary genomics: human-specific ecotypes of Lactobacillusreuteri have diverse probiotic functions[2014].

Lactobacillus Casei and Paracasei

On many studies, this is reported to reduce hay fever and allergies. If you check the web, you will find that it is also reported as a histamine producer. How can this be true since increased histamines would make allergies worst. The answer is simple. BOTH ARE CORRECT when you look at the fine print… (and you need the fine print that may be missing on the probiotic label).

  • “histamine production were found in … Lactobacillus casei 18, isolated from cider)” [2013]
  • “According to the results,  Lactobacillus casei CCDM 198 exhibited the best degradation abilities…. significantly (P < 0.05) reduced BAs (putrescine, histamine, tyramine, cadaverine), up to 25% decline in 48 h.” [2020]
  • ” Lb. paracasei subsp. paracasei CB9CT and another strain (CACIO6CT) of the same species that was able to degrade all the BAs were singly used as adjunct starters for decreasing the concentration of histamine ” [2016]
  • “Seventeen isolates were found that were able to degrade tyramine and histamine in broth culture. All 17 isolates were identified by 16S rRNA sequencing as belonging to Lactobacillus casei.” [2012]

For Lb Casei and Paracasei, most of the studies suggests that it degrades histamine.

Worked Example

We use L. Casi and L. Paracasei from Custom Probiotics, for two main reasons, they are the cheapest per BCFU, they have no fillers, pre-biotics, etc so we do not have to deal with counter-indicated formulation that often happens with commercial probiotics blends (often using a marketing-driven formulation).

So looking up the strains, I see Lc-11 and Lpc-37, Time to search for information on these:

On LPC-37

On Lc-11

Table 1.
Semanic Scholar

Bottom Line

IMHO, you need to:

  • Know every strain in your probiotics (not just species!)
  • You need to be able to locate studies using that strains (Lb. Casei Snakeoild may be just a marketing name to hide that fact that the manufacturer/packager does not know the strain)
    • I have seen some product literature claiming benefits from a different strain because they were the same species —FALSE LOGIC.
  • Ideally, you will find some relevant studies using these strains — ideally on humans!

If you are using antibiotics, you may wish to search for the probiotics antibiogram. Ideally, the manufacturer/seller would provide all of that information with a simple email requesting it.

The Brain’s Microbiome – what reaches it?

This is an area that I became aware of a decade ago and used this knowledge to discard some suggestions and take other suggestions based on the physical characteristics of the brain. This has major implication for brain fog, autism, ME/CFS, depression, Alzheimer’s, Long Haul Covid and many many more conditions.

The literature

“Research on disease-modifying treatments for  central nervous system  [CNS] diseases have generated a cemetery of failed drugs, rejected in part because of their incapacity to cross the BBB” [3,4,5,6,7]. 

The Molecular Weight [MW] threshold of BBB drug transport of small molecules has been demonstrated previously in studies of drug penetration into the brain.3334 Blood–brain barrier permeation decreases 100-fold when the size of the drug is increased from an MW of 300 Da, which corresponds to a surface area of 50 square angstroms, to an MW of 450 Da, which corresponds to a surface area of 100 square angstroms.35

Drug transport across the blood–brain barrier [2012]

I remember that for my treatment of ME/CFS back in 2000, Low Molecular Weight Heparin was what was advocated. It was found far more effective than normal heparin (and costed 30+time more!!)

Getting information on Molecular Weight

What is Da? It’s Dalton — “Measure of molecular weight or molecular mass. One molecular hydrogen molecular atom has molecular mass of 1 Da, so 1 Da = 1 g/mol.” This information is often available on Wikipedia in the left hand column as molecular mass. i.e. 457.483 g·mol−1 –> 457 Da

Information is also available at

If there is a choice, you want the smallest number. Remember we are talking 100x LESS getting thru with an increase of weight by 50%! Some examples:

For some items (lacking the specific chemical formula) we may not get that data.

Bottom Line

The concept of “needing to take anti-inflammatories for brain inflammation” is correct — the gotcha is that many of the items you may take will never reach the brain because they are too big! Go thru your lists (and suggestions from others) and trim them down to the light molecular weight ones. You will likely get much better success — I did!

Hypothesis Testing: 16s Results to detect SIBO

I’ve recently added computations for Methane and Hydrogen using KEGG data to Microbiome Prescription. Checking the contributed symptoms, I had over 120 samples with SIBO indicated. So it was time to test the hypothesis.


15% was 90%ile r over, 10% was expected
Poorer match:: 5% was over 90%ile, 10% expected

What about the old Methane?

The old computation was done on adhoc gathered associations from the literature. It also did not show any patterns.

Bottom Line

With the obvious path being unsuccessful, then time to examine where we did find associations.

(4R,5S)-4,5,6-trihydroxy-2,3-dioxohexanoateBetween 33%ile to 66%ile
cob(II)yrinate a,c diamideBetween 66%ile to 100%ile
D-tagatoseBetween 0%ile to 33%ile
undecaprenyl phosphateBetween 0%ile to 33%ile
Vitamin B9 (Folic Acid/Folate)Between 66%ile to 100%ile
Lactic acidBetween 0%ile to 33%ile
2-ButanoneBetween 66%ile to 100%ile
Hydrogen cyanideBetween 66%ile to 100%ile
Methyl thiocyanideBetween 66%ile to 100%ile
PropionateBetween 66%ile to 100%ile
Vitamin KBetween 66%ile to 100%ile
Vitamin B7 (biotin)Between 66%ile to 100%ile
Sialic acidBetween 66%ile to 100%ile
NorepinephrineBetween 66%ile to 100%ile
succinyl-CoA:acetate CoA-transferaseBetween 33%ile to 66%ile
phosphoenolpyruvate carboxykinase (GTP)Between 33%ile to 66%ile
Enzymes are a wash — all of focused on typical values
Pentose phosphate pathway (Pentose phosphate cycle)Between 75%ile to 100%ile
Pentose phosphate pathway, oxidative phase, glucose 6P => ribulose 5PBetween 75%ile to 100%ile
Pentose phosphate pathway, non-oxidative phase, fructose 6P => ribose 5PBetween 75%ile to 100%ile
Serine biosynthesis, glycerate-3P => serineBetween 75%ile to 100%ile
Histidine degradation, histidine => N-formiminoglutamate => glutamateBetween 75%ile to 100%ile
Riboflavin biosynthesis, plants and bacteria, GTP => riboflavin/FMN/FADBetween 75%ile to 100%ile
Tetrahydrofolate biosynthesis, GTP => THFBetween 75%ile to 100%ile
CAM (Crassulacean acid metabolism), lightBetween 75%ile to 100%ile
Lysine biosynthesis, DAP dehydrogenase pathway, aspartate => lysineBetween 75%ile to 100%ile
We find high levels of glutamate, serine, lysine, ribose and ribulose being scents to look at. Unfortunately nothing was found on PubMed
Blautia hanseniispeciesBetween 33%ile to 66%ile
Filifactor alocisspeciesBetween 66%ile to 100%ile
BacteroidiaclassBetween 66%ile to 100%ile
Bacteroides gallinarumspeciesBetween 66%ile to 100%ile
By bacteria found no strong associations

Last, we look at what studies reported

Tax RankTax NameShiftDistributionCitation Link
genusEnterococcus (NCBI:1350 )HighDistribution   📚 PubMed
genusKlebsiella (NCBI:570 )HighDistribution   📚 PubMed
genusPrevotella (NCBI:838 )HighDistribution   📚 PubMed
genusSalmonella (NCBI:590 )HighDistribution   📚 PubMed
genusStaphylococcus (NCBI:1279 )HighDistribution   📚 PubMed
genusStreptococcus (NCBI:1301 )HighDistribution   📚 PubMed
phylumFirmicutes (NCBI:1239 )LowDistribution   📚 PubMed
speciesAcinetobacter baumannii (NCBI:470 )HighDistribution   📚 PubMed
speciesBacteroides fragilis (NCBI:573 )HighDistribution   📚 PubMed
speciesBifidobacterium longum (NCBI:216816 )LowDistribution   📚 PubMed
speciesEnterococcus faecalis (NCBI:1351 )HighDistribution   📚 PubMed
speciesEnterococcus faecalis (NCBI:1351 )LowDistribution   📚 PubMed
speciesEnterococcus faecium (NCBI:1352 )HighDistribution   📚 PubMed
speciesEscherichia coli (NCBI:562 )HighDistribution   📚 PubMed
speciesEscherichia coli (NCBI:562 )HighDistribution   📚 PubMed
speciesMethanobrevibacter smithii (NCBI:2173 )HighDistribution   📚 PubMed
speciesMethanobrevibacter smithii (NCBI:2173 )HighDistribution   📚 PubMed
speciesPseudomonas aeruginosa (NCBI:287 )HighDistribution   📚 PubMed
We have some agreement, Bacteroidia is HIGH above and Firmicutes  is low.

Looking at the “usual suspect” for SIBO, Methanobrevibacter smithii, a methane producer, we found only 23 samples with any (and all of the labs associated with these samples reports this bacteria so over 100 (80%) of the people reporting SIBO had none appearing)

There is no clear association of this bacteria to SIBO in our samples

Tentatively conclusion, SIBO does not leave any clear tracks in a 16s Sample.

Hay Fever / Allergic Rhinitis – What we know

It is that season again — and some areas are reporting much higher levels than usual (with predictions for it getting worst). Some people will load themselves up daily on antihistamine, for example, Diphenhydramine HCl , which impacts over 800 different bacteria. We do have a profile of the bacteria shifts seen on Microbiome Prescription.


The bad news is that we have lots of studies, but no good studies — all of them have problems…

” A total of 57 062 articles were derived from searching seven online databases and evidence from 48 RCTs and 10 observational studies were reviewed for methodological quality and risk of bias. No qualitative studies meeting the inclusion criteria could be found, therefore only a quantitative review was performed. ”

Health supplements for allergic rhinitis: A mixed-methods systematic review [2020]



“Probiotics may be beneficial in improving symptoms and quality of life in patients with allergic rhinitis; however, current evidence remains limited due to study heterogeneity and variable outcome measures. Additional high-quality studies are needed to establish appropriate recommendations.”

A systematic review and meta-analysis of probiotics for the treatment of allergic rhinitis [2015]

Probiotic Potential of Lactobacillus Species in Allergic Rhinitis [2021 – full text] is a recent review with two appearing to be most likely effective (to some extent): Lactobacillus Casei and the closely related Lactobacillus Paracasei with dosages up to 30 billion CFU/day. These happen to be less than the suggested dosages from Custom Probiotic.


If we look at Nat.Lib.of Medicine studies at Microbiome Prescription, we do not see any reports of low Lactobacillus. What we do see is low Bifidobacterium) ( Bifidobacterium longum , Bifidobacterium adolescentis, Bifidobacterium catenulatum) and Clostridium butyricum. Logically those seem better candidates!

Bifidobacterium longum produces a rich collection of end products (1,380), the absence of which may account for hay fever.

Bottom Line

I suspect Placebo effect and poor study construction has resulted in the fuzziness for supplements and lactobacillus probiotics. The Nat.Lib.of Medicine profiles points to some specific bacteria that are low and the available studies, appear to suggest that taking those bacteria as probiotics will significantly improve hay fever. The list is:

Adequate Vitamins D and E supplementation may also help. I use the word adequate because often the dosages suggested in some studies are insufficient to alter blood level by any reasonable amount in a month (see this post for a formula ) – hence “no effect”.

There is one more path to consider, getting suggestions explicit for the shifts reports.

The results are shown below

The full details

We see L.Casei, L. Paracasei and Clostridium butyricum on the recommended list — in agreement with the above. Further down, we see Selenium (cited above) also listed

The above is evidence based on the microbiome shifts seen with hay fever.

What is your NEXT diagnosis?

This morning I chatted 90 minutes with another data scientist about the microbiome. After the video chat he sent me a link to this recent article: From IBS to ME – The dysbiotic march hypothesis [2020]

” The pathogenesis of the relationship is unknown. Intestinal dysbiosis may be a common abnormality, but based on 1100 consecutive IBS patients examined over a nine years period, we hypothesize that the development of the disease, often from IBS to ME, actually manifests a “dysbiotic march”. In analogy with “the atopic march” in allergic diseases, we suggest “a dysbiotic march” in IBS; initiated by extensive use of antibiotics during childhood, often before school age. Various abdominal complaints including IBS may develop soon thereafter, while systemic symptom like CFS, fibromyalgia and ME may appear years later.”

Related to the above:

Personally, I have seen someone progress from GERDs to IBS to Chronic Fatigue Syndrome to atypical Crohn’s disease. The progression is not deterministic, with DNA being a significant factor.

I have had on my todo list, creating a microbiome progression map. I have just added it (based solely on gold-standards PubMed studies). It can be seen via


When you click the crystal ball . you will be taken to a new page. For example, IBS

Associated medical conditions to IBS
For Chronic Fatigue Syndrome
For Autism

Bottom Line

This is based on PubMed studies which are often hit and miss for depth of analysis and reporting shifts. Over time, I expect data to improve and the forecasts on this page to improve.

COVID Microbiome and implications for Long Haulers

A Periodic Review of PubMed for COVID Fecal Microbiome finally found some studies:

One paper reported very very good results!

The optimal eight oral microbial markers (seven faecal microbial markers) were selected by fivefold cross-validation
on a random forest model, and the classifier based on the optimal microbial markers was constructed and achieved an area under the curve (AUC) of 98.06% (99.74% in the faecal microbiome).”

Alterations in the human oral and gut microbiomes and lipidomics in COVID-19 [2020]

“The heatmap showed that the faecal microbial community in CPRs (Confirmed Patients Recovered) was different from that in CPs (Confirmed Patients) and HCs,(Healthy Controls)” [SP is suspected Patient, SPR is suspected Patient recovered] which appears to confirm my hypothesis that most infections will leave a “garbage state” in the microbiome which will generally shift slowly back to the healthy state. This return to a healthy state is not certain and when it fails to happen, then we have diagnosis such as long haul covid, chronic fatigue syndrome, and post-infection syndrome.

More coming…

Alzheimer’s treatment via the microbiome.

This month (Feb 2021) there as a major article Structural and Functional Dysbiosis of Fecal Microbiota in Chinese Patients With Alzheimer’s Disease released. I have updated my list of bacteria (with links to source studies),

This post is for caregivers that are interest in low risk treatment that theoretically have a high probability of success and low cost.

Short Summary of Approach

The microbiome produces some 4000+ different chemicals. For many conditions, especially “untreatable”, it appears that imbalances in these chemical mixtures result in cells, including brain cells, malfunctioning.

Some drugs help — and often those drugs were seen to alter the microbiome, correcting some of these shifts. The stupid question is this, if we know the bacteria that are involved — then why not starve or feed to put it into better balance.

IMHO It works! In my 50’s I had a sudden onset of cognitive issues, including memory. A SPECT scan was read as Early Onset Alzheimer’s. I also had another diagnosis. That other diagnosis has a bacteria shift pattern reported in 1998 in Australia. Making changes to alter that pattern caused the cognitive issues to fade and disappear.


You need to have a microbiome sample (done by taking a little bit of a stool and sending it to a lab). Then the data need to be upload to the free citizen science site, Microbiome Prescription. Not all labs are supported (i.e. they do not make their data available in a suitable format); those that are supported are listed here (with discount codes).

Once the data is uploaded, there are two Quick Suggestions links that generates suggestions using Fuzzy Logic Artificial Intelligence techniques.

There is a demo logic that show all of the features…. BiomeSight Example Login

There are a lot of tools there, depending on you skill sets and devotion to seeking improvement.

There is a YouTube Channel showing how to use this site and discussion of issues.

Comparing Extreme 3% to Kaltoft-Moltrup Selection

A reader contacted me about a disagreement and the cause was a bug in the code for Kaltoft-Moltrup — subsequently fixed. This post looks at the bacteria selected by each for similarities and differences — so people can better understand the difference (which is a little abstract).

I am going to use one the demo samples from BiomeSight (BiomeSight:2019-06-10 Self).

  • Extreme 3% picked 29 bacteria
  • KM picked 24 bacteria

I sorted their selections below in alphabetical order, 13 are in common (just over 50% of the KM choices).

Kaltoft-MoltrupExtreme 3%
Actinomyces : Too HighActinomyces : Too High
Actinomyces naturae : Too High
Anaerofilum : Too High
Actinomycetaceae : Too High
Bacillales Family X. Incertae Sedis : Too HighBacillales Family X. Incertae Sedis : Too High
Bacteroides cellulosilyticus : Too High
Bacteroides denticanum : Too High
Bacteroides dorei : Too Low
Bacteroides intestinalis : Too LowBacteroides intestinalis : Too Low
Bacteroides rodentium : Too HighBacteroides rodentium : Too High
Bacteroides sartorii : Too HighBacteroides sartorii : Too High
Bacteroides thetaiotaomicron : Too High
Bacteroides vulgatus : Too LowBacteroides vulgatus : Too Low
Blautia : Too High
Blautia obeum : Too LowBlautia obeum : Too Low
Brochothrix : Too High
Brochothrix thermosphacta : Too High
Chitinophagaceae : Too High
Clostridium paradoxum : Too HighClostridium paradoxum : Too High
Coprobacillus : Too High
Coprococcus : Too HighCoprococcus : Too High
cunicula : Too Low
Dehalogenimonas : Too High
Desulfovibrio vietnamensis : Too Low
Johnsonella : Too HighJohnsonella : Too High
Johnsonella ignava : Too HighJohnsonella ignava : Too High
Lachnospira : Too High
Lactococcus : Too HighLactococcus : Too High
Leuconostoc : Too High
Listeriaceae : Too High
Micrococcaceae : Too High
Oscillospira : Too High
Prevotellaceae : Too Low
Streptococcaceae : Too High
Streptococcus vestibularis : Too HighStreptococcus vestibularis : Too High
Sutterella : Too Low
Syntrophobacteraceae : Too High
Tetragenococcus : Too High
Thiothrix : Too High
Turicibacter sanguinis : Too Low

Looking at a chart of Prevotellaceae, we see that KM low is 2.25%, thus be this sample being between 2.25 and 3 resulted it being excluded on one and included on another. For Listeriaceae : Too High, KM used 95.6% instead of 97%.

For Sutterella, KM uses 22% for low, hence it included. This is reasonable because there is a distinctive drop off around that!

For Coprobacillus, it looks like I need to do some adjustments of the KM, a chunk of unusual data caused a “step” that incorrectly triggered the high computation.

Bottom Line

We have good overlap with the differences being due to the curves being different. With the extreme 3% approach, we are insensitive to the difference of shapes. With KM we are sensitive (and some parameters to the algorithm needs a little adjustment).