A search tool is fine but what database is it going to be looking in? Are you talking about something like google or duckduckgo but with a focus on biochemistry? How would your search tool be better than simply using them? You are going to end up with a huge pile of semi-related studies and anecdotes. The relationships between various factors will not stand out until you painstakingly analyze the results
Thats why I say some form of ai is going to be needed. It will do the hard work of finding relationships between various drugs, various factors in the cell and so on. If your search tool simply uses the net, it will come up with a lot of junk. I think you can pay an annual fee to see all pubmed articles free can't you? If so that will help but if your search gives you 1000 results in order of relevance, you have a big job ahead of you to sift through it to find what you want. If ai is too expensive/ time consuming to do and search is easier, that may be the place to start. Later perhaps fine tune it and give it some ai capabilities.
The search tool will build its own database using user provided data. It will not search the net nor will it do any other sort of data extraction at least initially.
I will bootstrap it with data I've accumulated and users can add their own data or up/down vote existing data. Users could also add studies that confirm/refute relationships other people (including me) have defined.
There are a few main advantages to the search tool. There are disadvantages of course and many of them have been pointed out already (mostly by you, thank you for the good questions) .I don't expect the tool to be perfect but I see a niche for it. PubMed, Longecity, Google, etc. will still have their place. I opened this thread to see if the tool is useful enough for people besides me to use it
Here are 8 positives to the tool which might help explain why I want to create it and why people might want to use it
1) It will be a network graph.
It will be something like a much broader, dynamic version of a static network graph like https://www.google.c...iact=mrc&uact=8
It will show the complex series of relationships involved in chemistry much better than looking at studies one by one can.
A google search on Huperzine A will show that it decreases AChE and may even show that it increases ACh but it will not show
A) how ACh increases other things (like growth hormone).
B) show what else increases ACh (like sauna use)
C) show what decreases ACh (like CBD)
2) It will be focused.
2A) There will be one relationship per pair of objects.
Plain old googling will have many leads which will be hard to understand without a systematic representation of the data
2B) It will be focused on what its users care about
I threw out a bunch of my data as I learned what things are less relevant to me. I don't particularly care about motility at this point and knowing lemon balm's affect on it will merely clutter the data I care about. Of course if I ever get numerous users they may add motility relationships, but the tool will reflect actionable data to its userbase.
3) It will include anecdotes.
Just because there is no study for something doesn't mean it isn't sometimes true (and vice versa). For example, the tool might have a relationship between lemon balm and sleep which shows lemon balm increasing sleep as supported by https://www.ncbi.nlm...les/PMC3230760/ . However, I personally have a hard time sleeping after consuming lemon balm presumably since it is an AChE inhibitor. So while there would be the sleep/lemon balm relationship in the tool there would also be a lemon balm/AChE relationship which would help some users make more educated decisions
Personally, the two people who say that a supplement gives them energy are very valuable to me. I personally do rather poorly on most energetic supplements so rightly or wrongly this information lets me prioritize which strategy to try first. This anecdote will be especially useful if it is backed up by a mechanism (such as the supplement increasing NE)
4) It will be user driven. Mistakes will hopefully be hightlighted and perhaps eventually automatically removed.
5) It will be dynamic
Since there is too much data to show at once clicking on a node in the graph will expand the breadth of the data visualized. So if you search for huperzine A it will display one or two levels of relationships. Assuming I default the tool to 2 levels of visualization it would show that huperzine A decreases AChE which increases ACh. If you clicked on ACh you would see one what else influences ACh and what ACh influences. Clicking the ACh node again would hide this information
6) It will eventually be configurable
If there are enough users I will add logins so that we can learn about what the user wants. For example, if you don't care about motility you could click on the motility concept and tell it to permanently hide motility relationships. On the other hand, if you like me are affected by dopamine, NE, and ACh these relationships could automatically be highlighted and expanded
7) It will provide relationships between exogenous compounds
A search for lemon balm would show what classes it falls into. Besides showing that it is an AChE inhibitor and allowing you to see other AChE inhibitors it will list its common components. For example, you could see that lemon balm contains rosmarinic acid and then see what rosmarinic acid influences and what other exogenous compounds have rosmarinic acid. I seem to poorly respond to Rosemary (another rosmarinic acid source) and there is some mechanism to explain why I might do so. While this certainly isn't hard science it does help rule out potentially harmful treatments or at least helps to prioritize them.
8) The tool will highlight and attempt to resolve contradictory data/studies
I plan to have something like a green circle with a number representing the number of positive studies and a red circle providing the number of contradictory studies. It will also have thumbs up and thumbs down features like longecity.
----------------------------------------------------------------------------------------
I agree that semi related studies will be a problem. I am sourcing my initial relationships from this site, selfhacked and mybiohack. I have seen things in all of them that seem wrong or that at least I can't understand. To remedy this, I plan on including whatever study information I can find and I will eventually try to rank the relationship's quality based on this data. For example, was a study in mice and was it in vivo? If I don't link to a study other users could always do so.
I also agree that some AI will be useful. I personally don't care about every finding out every relationship between every possible system/compound, but I would like to
1) Test existing relationships by validating them with AI detected studies
2) Find new relationships between existing end points. For example, if Schisandra and sleep are two things users care about I would like to know the relationship between the two. If none of my users care about decreasing baldness then I will not care about what Schisandra does about baldness.
I feel an organic slow development will be the only way to go here. If I don't get additional development support or tons of users I will probably only add the features I need and add them as I need them or have time to do so.