“We are made to persist. That’s how we find out who we are.” ― Tobias Wolff
Two years in the making, with a database of 6000+ individual lab markers and their values in specific pathologies, I’ve finally cracked an algorithm that has eluded me for over a year and a half: How to use machine learning AI to predict the probability of a certain diagnosis based on an individual’s collection of laboratory results. Because the pathologies link to the NCBI MeSH (medical subject headings) they can also be used to weigh specific clinical signs and symptoms to further establish an accurate diagnosis.
More work on my machine learning diagnosis engine. Underneath the likely diagnoses (from the user-supplied lab results) I’ve added a heat map that correlates the known symptoms associated with each pathology resulting from the reported lab data. The meaner the red color the more the symptom is associated with that disorder. The practitioner can also tick off the ones that apply to the client, press a button, and these probabilities will be added to the results from the laboratory data to further refine the guesswork.
Last week I had the honor and privilege to present on my bioinformatics research at the Clinical Applications of Scientific Innovation (CASI) conference in Nashville, TN.
As part of my presentation, I prepared several of the various apps found in Opus23 and Utopia (genomic and microbiome analytics, respectively) to be able to be used ‘open-source’ (without a need for a loaded patient). I also included the Rheingold app we use at HouseCall for checking possible drug/ supplement interactions and an interesting app that allows for better diagnosis.
Response was quite positive. I think many of the physicians in the audience we unprepared for the scope of the coding that I’ve produced over the last few years. One came over to me afterwards and told me that it was the first truly new thing they had heard in the last four conferences they had attended this year.
Although Nashville has an amazing music scene, Martha and I were too involved in the conference to venture out very far from the hotel. I did regret not being able to visit the Parthenon replica there.
Next week I’ll be presenting at the The Academy of Integrative Health & Medicine (AIHM) in San Diego. That will conclude my lecturing for 2019. Hopefully to then devote more time to HouseCall and Volkswagens!
There is a large gap in our knowledge concerning the potential for interactions between the foods we eat, the herbs an and supplements we consume, and the drugs we are prescribed. To be able to address and identify these interactions can help prevent unwanted side-effects and predict circumstances where a drug in a certain person may be getting removed too quickly in order to properly do it’s job.
To do this, I had to first build a large database of agents that influence the function of specialized enzymes in the liver called cytochromes. Cytochromes are the ‘lead off hitter’ in virtually all the detoxification processes that are coordinated by the liver. There are many, usually identified with a number and the prefix ‘CYP’. Then I had to program the analytics that could cross-reference the dug/food/supplement intake of a specific client with their effects on cytochromes.
I was particular motivated to complete this project because I had a client who was taking four prescribed medications for anxiety, insomnia, and stress and quite a few supplements. In addition, with this client I was able to superimpose their Opus23 genomic data, and as can be seen in the image, the red box for cytochrome CYP2D6 shows a double whammy: The enzyme is compromised because of the client’s genetic mutations, plus one of their meds is inhibiting it even further. This is noteworthy because CYP2D6 is known to metabolize as many as 25% of commonly prescribed drugs, including antidepressants, antipsychotics, analgesics and antitussives, beta adrenergic blocking agents, anti-arrythmics and antiemetics.
Rheingold also supplies predictive data with regard to what might be a consequence of adding addition agents to the client’s protocol. All in all a very satisfying coding experience. In this example, the client may well have a problem if prescribed a beta-blocker for hypertension, or codeine for a cough. Both would leave the body rather slowly, increasing chances for a drug side-effect.
The gene for CYP2D6 is highly polymorphic (variable between individuals). Certain variations in genotype (‘alleles’) can result in the ‘poor metabolizer’ phenotype, characterized by a decreased ability of the cytochrome to process its specific toxins or substrates . Some individuals with the ‘poor metabolizer phenotype’ have no functional protein since they carry 2 ‘null’ alleles. In these folks drugs metabolized by CYP2D6 will take a LONG time to detoxify. Other individuals with the ultra-rapid metabolizer phenotype can have 3 or more active copies of the gene. In these folks the enzyme processes the drug or substrate so fast that it is likely to not have much of an effect at all.
Considering the huge numbers of people who take various combinations of drugs, along with other agents, such as vitamins and herbs, it is surprising that so many of these are prescribed without any sort of due diligence with regard to potential cumulative interactions. HouseCall clients can now feel assured that we have their back when it comes to identifying and predicting possible improper drug and supplement combinations.