These are good questions to raise. I'll try to give my own perspective as to why I am not greatly concerned about AI or espionage as significant threats.
AI has great potential as a tool for drug discovery. It can be effective in domains that I would characterise as information rich. That would include things like protein and peptide design, or small molecule design for binding to receptor sites, or perhaps novel synthesis pathways. In these domains we have huge datasets - protein sequence libraries, protein structure databases, very many drug libraries to support high throughput combinatoric screening, a vast literature on synthetic methods, etc. Machine learning and AI methods are interesting when you have large amounts of training data like this, and a well defined problem.
Pentosan polysulfate production is not this kind of problem. It is principally a process engineering problem. The problem that needs to be solved is how, from a given starting material, do you process beech bark through a sequence of modification, separation and purification steps, to produce a material that conforms to a spec that is held in confidence by the FDA, TGA etc. This is not an information theory or design problem, it is a very boring and tedious process technology problem.
It will involve things like how long and at what temperature and pH do you digest the beech bark, what kinds of columns do you do the fractionation on, how do you regenerate them, what solvent mixes or gradients get used, what kind of cleanup steps you do, etc. I believe there are dozens of process steps along the way, all with process parameters to optimize, and complex process variable interactions. Each process also requires development of analytical methods, so that when that step completes you can tell if it did what it was designed to do. It's all very empirical and hands on. It's not the sort of problem AI is good at.
For similar reasons I am also not too worried about industrial espionage. There probably is a lot of internal knowhow and espionage could short cut some development. But just as important as the process itself is the process engineering culture that supports it. It's the tacit and often unwritten knowledge about how to troubleshoot a process if it goes out of spec and you need to bring it back under control. Its not specific aspects of a recipe or formula that are important as the expertise that underpins the whole process and the various ways it could run off the rails, and the institutional/cultural ability to bring it back into spec if it does.
If OA was a simple disease, with only one or two mechanisms involved, AI would be a great tool to look at a particular receptor and start proposing novel peptides or drug molecules that bind to that receptor. That's lead generation, and it's just the start of the story. As we know there is a huge pathway from theoretical binding constants to lab to preclinical to clinical trials with real world requirements for efficacy, safety, side effect profile, etc. So for a new drug candidate all the hard work of drug development still lies ahead. And OA is not a simple disease, and pentosan does not work via a single or simple mechanism.
I don't discount the risk of biosimilars, but I don't lose much sleep over AI or espionage. Hope thats helpful.
Expand
(20min delay)