fgimenez

[Note that I work in this field and have co-founded a CNS AAV company]

This isn't the correct takeaway. AAV are one of the most complex drug modalities and carry considerable risk when used incorrectly. This story is tragic and violates pretty much every ethical consideration for a clinician researcher. Especially ones that are treating children of desperate parents.

That said, AAV are one of the most powerful delivery mechanism we have to deliver gene therapies to the brain. Uniqure has shown the first efficiacious treatment of Huntington's disease with intraparenchymal delivery of AAV5, Zolgensma is a brain targeted AAV9 to treat SMA, Kebilidi is an intraparenchymal AAV2 that treats AADC deficiency.

The general approach should be to keep dose as low as possible and minimally expose the periphery. AAV9 at large doses delivered intrathecally without standardized immunosuppression is simply insane.

They have a lot of anecdotal, observational, and emerging RCT evidence on their effects on substance consumption and abuse.

The biggest effect and best tested is on alcohol use disorder. Mechanistically we don't know if it's through some complex reward mechanism, or something simpler like "alcohol is a calorie and you consume fewer calories." The JAMA study showed that GLP-1 reduce Heavy Drinking Days (>2 drinks/day), but did not reduce overall drinking days. This would imply the simple mechanism -> it's hard to drink a lot of calories even if you do enjoy a drink.

More anecdotal evidence showing this effect in opiates, but nothing in an RCT yet.

So far, nothing has worked in stimulants. Cocaine and Meth abuse are insanely difficult to manage therapeutically right now.

Basic research creates foundation knowledge that can drive medical innovation, but rarely does academic research create final composition of matter. Private funded work is all the non-research components of drug discovery - optimization of molecules, regulatory work, commercialization, etc...

To imply that private companies reap the rewards of basic research without contribute much is ignoring the many other components of translational work.

First-gen GLP-1 goes off patent in 2031 (e.g. semaglutide). Seems far off, but is frighteningly close for Novo. Tirzepatide gets genericized in 2039 and has better efficacy, which is why Lilly is in such a strong position right now.

There is an enormous amount of biotech work to develop next-gen versions that have better half-lives, lower adverse events, and most importantly, have long patent lives. But it seems base GLP-1 are good enough that we should see massive societal change starting next decade.

Yes absolutely. Even if you have already been exposed, the vaccine has been shown to prevent manifestations and/or recurrence of HPV-induced pathologies. It's not necessarily in standard of care, but the literature is pretty overwhelming here.

My pithy observation - Early stage investors are fantastic at extrapolating from minimal data points. Late stage investors are fantastic at extrapolating from many noisy data points. They screw up because the methods for each of their extrapolation functions are entirely different, so much so that they manifest as cultural differences in investment firms.

> I wouldn’t be surprised if companies threw billions of dollars worth of research to make it happen

Well you are correct. Altos labs (https://altoslabs.com/) is basically every "who's who" in science doing this with billions of dollars. Including Wolf Reik who is featured in the article.

Go through a staffing agency. They will manage background checks and can introduce you to many people across a variety of skill sets.

I am VC. I invest in hard tech. That said, opinions are my own.

Unless you come in with a strong background reputation or intro from the top 1% of our network, I would probably not read a document that takes 30min. I cannot imagine many people would without enough incentive.

Think about it this way, if you're going to commit to watching a movie, do you do research beforehand to choose? The common joke is you spend more time searching for recommendations than just watching. The psychology is we want to de-risk committing our energy/attention. This similarly applies to founders sending technical tracts. We do eventually read a tremendous amount of technical details (lit reviews, white papers, etc...) but only after we have understood that the opportunity is worth the effort.

Okay, all that said, there is a deeper code smell here. I think you are likely mixing product implementation with market opportunity. Description of the implementation of your product takes lots of time and explanation as you state above. BUT, you should not be doing that in your first pitch. The first meeting should be explanation of the market opportunity. You aren't selling your product. You are selling your market. If I'm sold on the market, I want to hear why your product captures it afterwards, not before.

I cannot comment on a full macro basis, but it is definitely true that being listed on an index such as S&P 500 will spike the market cap. TSLA was a recent example here. Matt Levine has an interesting piece on this (as usual) regarding trading on index inclusion as securities fraud: https://www.bloomberg.com/opinion/articles/2021-01-05/dystop...

I chatted with an affiliate of the counterparty to the bet here - a couple years before the bet ended but clearly when they were gonna lose. They did say two interesting things I'll relay here without necessarily agreeing:

1. Their major thinking is that the growth of index funds is driven by volume of new investors, not necessarily market performance. At some point we hit the diminishing returns of new money into indexes. When that happens we'll see their "guaranteed" growth slow and you'll need to turn to hedge funds for alpha. He thought 10 years was enough for this to play out. Obviously wrong on timing, but not necessarily wrong on outcome.

2. One of the conditions of the bet was that they have lunch once a year to discuss bet progress. Given that charity lunches with Warren are going for 4.5mm today, they essentially got 10 lunches for 100k each. That is...quite valuable for a hedge fund manager.

Now, nobody can sell a loss better than a hedge fund, so I take with a grain of salt. But it is some food for thought.

My experience has been to say "yes" to everything in early career when you are not in demand. Then slowly transition to saying "no" as your career develops. The transition point from "yes" to "no" is when you develop more unique and valuable skill sets.

When all you have to offer is energy, excitement, and smarts - "yes" opens doors, creates relationships, and gives you opportunities to learn. That's how you grow, not only in career, but in relationships.

After you can bring more differentiated value to an opportunity, you're going to be in more demand and need to filter the best use of your time and energy.

The Steve Job's part is absolutely correct when you are sitting in Steve Jobs' position. As a thought experiment, probably every Fortune 500 CEO would want to have a 1-2 hr meeting with Jobs circa 2008. That would be 500-1000 hours of meetings where arguably they would derive more value than he would. So while it would be insane for most to pass up on these meetings, he obviously would need to.

Don't confuse the "yes" vs "no" periods of your life.

This is such a cool idea. The gap between the "aha" moment of a programming challenge and the implementation is high for me since I don't get to code very much anymore. Can't wait to check it out!

Really depends on the type of cold email. If an investment banker cold emails me with a “hot deal” I delete immediately. If a founder sends me a template email in a space I clearly dont invest, I delete. If a dev sends me an email with interesting insight or project and the ask is an easy to answer question, I’ll respond. Might go out of my way and chat if I’m impressed.

I’m a VC, so this take is biased. Liberally apply grains of salt.

If you are looking to start a VC-scale startup, I’d suggest talking to VCs who can help introduce you to domain experts. There are a few reasons this is beneficial:

- You get to battle test your idea to hone the story. This will help you recruit non-technical people.

- You build relationships with VC for the moment you need funding.

- You learn which VC are actually helpful. The meme is that all VC have to pitch how much value they add, but the reality is most can’t/won’t/don’t outside of MBA 101 platitudes.

- A great VC is always building out networks for BD and founder identification. You should leverage that to help in your space.

- Even if the idea doesnt pan out, if you prove yourself to be talented and hard working, they are incentivized to help you find a role relevant to your interests.

This is colored by my personal investing philosophy that you should prioritize meeting “makers”. Others prioritize meeting executives, or sources of capital, or influencers, etc... None is a “true” path, there are successes in each. But the ones focused on makers always want to chat with folks like you.

I have met one person who did it. Even then they were a certifiable genius, still had to jump through an enormous amount of bureaucracy with an enormous amount of sponsorship from many faculty whos grant funding was directly tied to this person working on their project.

So in practice, no.

If you want to learn more theory, I do think it's possible to apply and get an MS. Especially now since MS programs are bleeding people due to remote learning.

- Norvig's AI: Doesn't have much deep learning, but you get through it and understand the expansiveness of the field.

- Algorithms - Papadimitrou and Vazirani: I had a professor who described this as a poetry book about algorithms. Alternative is Sipser

- An Introduction to Statistical Learning: This is like a diet form of Elements of Statistical Learning which is much more approachable and pragmatic.

- Janeway's Immunobiology - De facto standard of immunology. Great.

- SICP: duh

- Principles of Data Integration: This is more because the subject matter is so important and nobody really has studied fundamentals. Did you know general data integration is AI-complete? If 99% of work in AI was spent on data integration, the field would move so much faster.

They are a phenomenally cool drug delivery vehicle as you can get differential uptake based on their surface markers. Check this out: https://www.sciencedirect.com/science/article/pii/S221138351...

Happy to geek out more offline. Email in my profile.

Yes, you could frame the Voyager treatment as "better levodopa". Though, the mechanism would feasibly mitigate neuro-degeneration and improve the terrible dose escalation of the drug.

Also correct about general delivery difficulties. Hadn't heard of Biontech's method - almost sounds like voodoo by your description.

I think there is interesting blocking and tackling happening on an organ-by-organ basis. E.g. GalNAc for liver hepatocytes, LNPs for systemic mRNA therapies, direct injection for eye or CNS (cheating, but still works). I'm partial to exosome hype...

Also no such discussion is complete without saying CRISPR, but the point remains that you can conjugate it targetted vehicles like antibodies. Conjugating to antibodies seemed to work for Stem :p

I sincerely think we are in the gene therapy renaissance. Delivery has always been an issue, but there are new tricks coming out with regard to cell-specific targeting. We can piggy back off the work done for RNA-based therapeutics (mRNA, ASOs, RNAi).

As for the CNS - Voyager therapeutics has had some great readouts in September in gene therapy for Parkinson's. More data coming out in Q1 to see second part of study.

Most biotechs don't have any revenue when they go public. They are traded on scientific progress through clinical trials. The idea being that if they do get through a phase 3, they are immediately valuable since they have a monopoly on a valuable therapeutic. Most likely, they get acquired by large pharma before that point - see Kite & Gilead.

Incredible. Also, getting treated by Steve Rosenberg is so awesome. Your results are literally the reason he'll get the Nobel prize within 10 years.

While robots are a stretch, there are a ton of companies taking blood samples and sequencing DNA floating around.

Karius does exactly what you mentioned by sequencing and searching for bacterial/viral dna in your blood.

Guardant and Grail do this to detect and diagnose cancer from tumor DNA in your blood.

Naterra and Ariosa do this to diagnose genetic diseases in fetuses of pregnant women.

I fund people like this a lot. They're always hiring developers for the bioinformatics parts of their products

I like to borrow from the decision theory field. There are massive frameworks around the "value of information" or "value of clairvoyance"[1].

The best summation of their framework is "Make a decision when the probability of new information changing your mind is minimal."

Of course, we need to quantify "new information" and "minimal", but it's a good framework to consider what new data points would change your existing course.

[1] http://lesswrong.com/lw/85x/value_of_information_four_exampl...

Check out Lyn Dupre's "Bugs in Writing." She was the department editor for Stanfords Biomedical Informatics program for several years and synthesized her work into that book. It's required reading for incoming students.

I have seen a lot of startups coming out to "re-invent the clinic" essentially by re-inventing DPC. They generally focus on on a wealthy clientelle with the (tesla-like) promise of moving down market. They aren't small concierge physicians, though. It ends up looking a lot more like one medical on steroids.

Not sure if it is a reasonable model yet, since these firms are incentivized to minimize patient contact. In the positive view, it means they want to focus on preventative care. In the negative view, they will get suffocated with hypocondriacs and have to institute limits to care that invalidates the allure of an unlimited insurance-free provider. Plus, these places effectively have to underwrite like insurers to remain profitable which is not easy when competing woth sophisticated giants. Not to mention the risk pf dealing with a very unstable ACA that will likely change the landscape.

Thanks for this cool writeup. My undergrad probability professor had us prove the CLT using cumulants, but I never kept the notes and had trouble finding the same proof later in life. Do you know of any reference that goes in depth about cumulants/FT/CLT?

Catenus Science | San Francisco | Full-time | On-site

What: Summer Data Science Fellowship

Program: The Catenus Science Apprenticeship Program identifies top data scientists who will raise the bar when hired at a startup. To help meet this goal, the program will train qualified candidates to have immediate, meaningful impact as data scientists in some of the top data startups in the world. This program will hone their skills in statistics, machine-learning, programming, and product development by presenting them with real-world challenges put forth by startups in Silicon Valley and the Bay Area.

What We Offer: Catenus Science offers a fully-paid, 13-week apprenticeship during which we reinforce technical and business skills. We do this via a mix of formal instruction and hands-on application of data science in some of the best startups in the world. Every startup is different, and each has its own unique culture and opportunities. That is why we want to give apprentices exposure to several places so they can find the right fit. Apprentices will have the opportunity to do rotations through three startups in a three month period while receiving training and mentorship the entire time. These startups will have well-scoped, immediately impactful data science problems for apprentices to solve. The program is structured such that apprentices have four days a week of full-time employment at a startup and one day a week of instruction. Apprentices are expected to already be knowledgable in data science, so we view these instructions as a way to hone their skills. In addition, we will use this time for apprentices to present the problems they are working on to the group. This way they can teach about new insights and seek help when faced with obstacles.

Who We Are Looking For: We are looking for candidates who posses technical virtuosity, communication skills, and business sense to use data in impactful ways. Necessary, but not sufficient, mathematical skills include a mastery of statistics and machine learning, and the ability to tell the difference between these two fields. Apprentices should have the hacking proficiency to code any idea regardless of programming language. Nobody can understand those who cannot communicate. You should be able to clearly convey your thoughts, both in writing and orally. Communication does not end at expressing yourself, though. We expect apprentices who recognize the strengths of new ideas by listening to and understanding others. Finally, we have seen time and time again that the best data scientists are ruthlessly pragmatic. The ones creating state-of-the-art products understand the domain-problems they are solving and how to deliver those solutions efficiently. Apprentices should be able to formulate solutions to relevant business problems with data science tools. Furthermore, they should be able to identify new business opportunities by looking at data.

Learn more and apply: http://catenus.com

Contact us: [email protected]