pfisherman

I hear you about switching off and not ruminating over programming problems after work hours. I have found that (a) exercising for 30 mins helps put me “in the moment” provides a nice mental break and “palate cleanser” and (b) practicing mindfulness has helped me be more intentional about controlling where my thoughts go.

By this I think they mean “ethical non monogamy” in the US?

People have been screwing around on their partners throughout history. Some cultures turn a blind eye so long as proper discretion is exercised. In others, polygamous marriages are explicitly sanctioned - usually to the detriment of women. The US tends to be a bit puritanical when it comes to sex.

For those who are not familiar, the I-SPY 2 trial is quite remarkable. It is an adaptive trial design - in essence reinforcement learning with human experiments.[1]

For all the criticism FDA gets, they should also get credit for leaning into new approaches to testing drugs like adaptive trials. Part of why the US remains one of the most innovative countries.

1. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7731787/

This is what I wonder about. Locking items behind glass has to have some second order effects? When Target started locking toothpaste and laundry detergent behind glass, I stopped going to that store entirely. They lost the sale of whole basket of goods to prevent shrinkage of one item!

When I see this I can’t help but think that these big box retailers are in a death spiral.

Internet mobs have been mainstream since Kony 2012 made slacktivism a thing. Assigning cancel culture to one part of the political spectrum is not intellectually sound.

Hahaha! I agree that both are high risk, high reward. The difference is that biotech companies semi-regularly produce products that large numbers of people use and derive benefits from. Not saying that won’t happen with Quantum Computing, but QC has felt like was 10 years away for about 10 years now.

Unlikely this could ever be done. We share too much of the same firmware for something like to be possible. The average human genome has about 4-5 M single nucleotide polymorphisms (SNPs) - rate in the order of 1 SNP / 1000 bases. The rate is much lower in functionally important parts of the genome because those tend to be under evolutionary constraint / conservation.

The linked article is pretty ridiculous. I am pretty sure it leaves out a lot of nuance from the underlying work - as these types of press releases normally do - but there is a bunch of stuff in there that just does not make sense.

First the taboo against consanguineous marriages was most likely because of genetic diseases. Biology may not have been very advanced, but people were smart enough to pick up on patterns. Similar to taboos against cannibalism despite having no concept of prion disease.

If this was pushed by the catholic church then explain Italian families!

Is “the hero’s journey” not a story about rugged individualism that has been told over and over in different forms across cultures since the beginning of history?

The skepticism should extend beyond big pharma to anyone trying to sell healthcare products or treatments. There is good reason for the Food, Drugs, and Cosmetics Act.

What gets me is how people will distrust pharmaceutical companies - whose marketing claims are at least regulated by the FDA - and then blindly trust nutraceutical companies, random streamers (i.e. spon con), etc.

Exactly. The article only mentions spending by PhRMA. While they are the biggest, there are other trade groups out there. Also, all of the big pharma companies have their own individual PACs.

For example, Sanofi individually spent $4.77M lobbying efforts in the US apart from any contributions to trade groups.

PDF warning: https://www.sanofi.com/assets/dotcom/content-app/documents/L...

I do not think it is a secret that most pharma companies are vehemently opposed to the drug pricing provisions in the inflation reduction act. I would be surprised if the lobbying spend across industry is not much more than $9M.

Sure. Unfortunately people already thought of this long ago.

Drugs often have more than one active pharmaceutical ingredient in them. For example, rybelsis vs mounjaro. The average senior citizen is taking four medications.

Pase 1 trials test doses of therapeutics far above what ends up on the market. In general there is a trade off between safety and effectiveness. Note that safety is evaluated in the context of the underlying condition. So worse side effects are tolerable for treating something cancer or HIV, than for treating something like a common cold,

The drug in the example given by the author, Alopurinol, is very hepatotoxic so cranking up the dosage is not something you can do without risk of killing someone. Or you could just use rasburicase - recombinant urate-oxidase - instead.

The point of this blog post was totally lost on me. I can’t even say it is a critique of the practice of medicine - or maybe therapeutic development (?) - because literally every example given by the author is a solution rather than a problem.

Also the author clearly does not know what they are talking about. For example, them calling for the use of surrogate endpoints - they clearly did not know the term of the art - when a whole regulatory framework around this concept has already been developed, formalized, and implemented!

To me the piece has a mix of smugness, hubris, and gross ignorance that would make for truly masterful work trolling if I did not believe the author were sincere.

This is not a very good analogy. Bread is not a dual use technology like surveillance tech.

There are plenty of classes of goods that are heavily regulated. For example, explosives.

Depends on what you want to use the data for. In general, you should take a trust but verify approach to new results, methods, data, etc.

No. Makes more sense to license the asset (molecule). And even then, big pharma companies won’t talk to you until your molecule is in phase 2.

If anybody with an AI pharma start up tells you they plan to make money by licensing their platform, then run the other way because it is a red flag.

If you want to see what a real “AI” pharma company looks like check out Vertex pharmaceuticals.

I mean what you suggested already kind of exists in several forms.[0,1]

Third party data analyses is a thing, but this is often bundled with conducting experiments by CROs.

See my other comment about market size. The problem is that you can make A LOT more money selling drugs than you can selling software to pharma companies. So if your software is any good, then you should use it to make drugs and just be a pharma company.

0. https://www.opentargets.org/

1. https://www.citeline.com/en/products-services/clinical/pharm...

I already gave you a real life example with the uniprot reference. Here is another flagship knowledge base that heavily leverages NLP extraction.[0] Here is another one that gets used in what seems like every network biology article.[1]

Meta analyses? Automating meta analyses is not a real need. They have their place, but it’s like a quaint cottage industry type thing - like custom haberdashery.

Also the most valuable knowledge is not in any publication. If you are reading about it in an article then you are already 2-3 years too late.

0. https://geneontology.org

1. https://string-db.org/

I think the big case study everybody gives is Schrödinger who took 20 years to IPO.

The problem is that the market is just not that big. Assume that every single pharma company buys your product - like Schrödinger - where does your revenue top out?

You can beat that as a small pharma company with one or two assets that make it to market. So the question is: if the computational platform is so good then why not just be pharma company?

Uniprot fills a pretty big niche - proteins. Your body is literally made out of proteins. They carry out most of the work and chemical reactions that amount to what we call life, and almost all drugs work by modulating some protein target in some way. UniprotIDs are the canonical identifiers that computational biologists use for proteins.

But they are just one knowledge base. There are others. Each with their own focus area. And they are all associated with prominent bioNLP / biomedical AI research labs and employ human SME curators.

> 5-10 avg salary non-star ML FTEs likely produce very mediocre and likely non usable results

LOL!

Word of advice to all those who are chomping at the bit to disrupt pharma with AI.

A pharmaceutical company that heavily leverages computation is called a pharmaceutical company. All modern pharmaceutical companies heavily leverage computational tools - including some powered by deep learning.

Any company building a computational platform to accelerate drug discovery / development is not a pharmaceutical company. They are a software company who wants to sell software to pharmaceutical companies, which is a terrible business to be in.

The product the author is selling already exists for free. All of the big knowledge bases use automated ML tooling for curation and extraction. And the people who run them and QC them are world class experts in their domains. I mean, just take a look at uniprot.[0]

And the only types of pharma companies who would want a big knowledge graph would be the large ones with active programs across multiple therapeutic areas. Most of the small companies / start ups tend to be focused on getting one or two assets to market in a therapeutic domain. And the big companies possibly have their own ML teams doing literature extraction - because a team of 5-10 FTE ML / software engineers is like a rounding error in their R&D budget.

The other thing is that the most valuable knowledge is the stuff that is not in the literature. That’s why we do experiments.

[0] https://academic.oup.com/nar/article/51/D1/D523/6835362

How much of this is finding the right data structure (graph) vs translating the problem into a new domain?

I think maybe the former follows from the latter?

Why the company and not the employee?

I think that the taxpaying public receives benefits from the remote work arrangement; such as less traffic, noise, pollution, and consumption of public resources such as roads.

I like the idea, but not the implementation. I think it should be done as a tax credit for the employee or the company instead.

Doing so creates positive incentives for companies to hire remote workers in California. Requiring the company to pay up by force of law disincentivizes companies from hiring remote workers in California.

I also think that it would be more equitable since an employee’s home office setup belongs to them and not their employer.

A quick search on tells me there are. I am speaking from personal experience. If I do an intense workout and don’t eat within the next few hours I get anxious and depressed.

Took me a while to realize that diet was the culprit; but after a lot of self experimentation I am 100 percent certain.

Have you ever tried heated yoga? Apparently it is very popular in Houston -- as if the oppressive heat outdoors wasn’t enough.

Yes they buried the caveats at the very end of the article. They don’t know whether effects are specific to hot yoga or exercise in general.

My experience is that 30 mins of vigorous exercise at least 3 days per week has huge benefits on overall wellbeing - including mental health. The caveat there is that diet has to be tuned to support your level of activity. If you are exercising and not eating enough, low blood sugar can bring on the bad feels / thoughts.

I’ll have to double check, but I think they tested in an animal (mouse) model. Not sure how far the compound got in the pipeline got for the original indication, but since it is from the repurposing hub they might be able to jump to phase 2.

They already do. QSAR has been a thing for a long time. The limiting factor is almost always the data. The publicly available SAR data in chembl / pubchem is trash, and why would you spend time and money designing and generating a good training dataset for a model when you could just be optimizing the compound?

Keep in mind that the dataset generated during optimization and the one you would prefer to train a model are quite different.

Also, in practice optimizing a lead compound is about more than just potency. You are also worried about things like toxicity, bioavailability, off target effects (selectivity), synthesis routes / yields (manufacturability), etc. And functional readouts (the assays you run) are not consistent across diseases / indications / targets / compounds.

That being said I am almost certain someone has used Bayesian optimization for an assist.

Alex Tropsha has some pretty solid articles on best practices.