The cynical part of me wonders: if this has been a promising approach for 10+ years, why weren't they able to secure VC funding years ago (or nonprofit biomedical research funding from places like the Gates Foundation that care a lot about infectious disease)?
This. Claims of a universal antiviral are as old as western medicine itself. Literally the only reason anyone knows what the Hippocratic oath is is because Hippocrates was already famous at the time for promoting elderberry as the universal antiviral.
I am totally onboard with the premise (as a TechBio-adjacent person), and some of the approaches you're taking (focused domain-specific models like Orthrus, rather than massive foundation models like Evo2).
I'm curious about what your strategy is for data collection to fuel improved algorithmic design. Are you building out experimental capacity to generate datasets in house, or is that largely farmed out to partners?
We think that Orthrus can be applied in a bunch of ways to non-coding and coding RNA sequences but it's definitely fair we're a bit more focused on RNA sequences currently instead of non-coding parts of the genome like promoters and intergenic sequences.
For the data - Orthrus is trained on non experimentally collected data so our pre-training dataset is large by biological standards. It adds up to about 45 million unique sequences and assuming 1k tokens per sequence it's about 50b tokens.
We're thinking about this as large pre-training run on a bunch of annotation data from Refseq and Gencode in conjunction with more specialized Orthology datasets that are pooling data across 100s of species.
Then for specific applications we are fine tuning or doing linear probing for experimental prediction. For example we can predict half life using publicly available data collected by the awesome paper from: https://genomebiology.biomedcentral.com/articles/10.1186/s13...
Eventually as we ramp up out wet lab data generation we're thinking about what does post-training look like? There is an RL analog here that we can use on these generalizable embeddings to demonstrate "high quality samples".
There are some early attempts at post-training in bio and I think it's a really exciting direction
There are biotech companies like Eikon Therapeutics (https://www.eikontx.com/ ) where super-resolution microscopy in living cells is a central part of the platform.
There is also one widespread approach that isn't mentioned in the article: expansion microscopy. Expansion takes the scifi-sounding approach of: what if you could make your sample physically bigger? See the Wikipedia page for more: https://en.wikipedia.org/wiki/Expansion_microscopy
1. Stephen Hell has been theorizing about how to do super-res microscopy since the mid-90s, so the article saying it was sci-fi "20 years ago" is off by about 10 years.
2. Stephen Hell has recently given the world another new technique, MINFLUX, which seems to be his best gift to super-res researchers so far. :)
"The other problem is with viral vector based gene therapy is you can’t have it again. You develop antibodies which prevent it from working again, and it could
cause a dangerous immune response."
Just wondering - would it make sense to immune-suppress the patient for a short period of administering of the viral-based therapy.
And as they describe that most gene therapies affect only extra-nuclear DNA, and thus have no permanent effect, wouldn't mRNA work better then in such cases - naturally the tech wasn't there 10+ years ago, yet today thanks to COVID it is here.
Edit (due to posting rate limit) in response to comment below:
I was thinking about mRNA coding dystrophin like it was coding COVID protein - should be cheap and easy (well, for some definition of easy in that context) doable, and it would be like a weekly self-injection - no toxicity, etc. Of course fixing the issue once for life would be better, once such cure becomes available, yet for now it would be similar like diabetics have with insulin - hassle for sure, yet it works.
AAV based therapies may have no permanent effect when the cells in question are actively proliferating (and the payload dilutes with each division) but muscle tissue is largely post-mitotic.
mRNA is in comparison very transient (in the range of days, and that's being charitable), even when modified (5' cap, uridine analogs, poly(A) tail) as it was in COVID vaccines. This is fine for vaccines, as you essentially want just a single exposure to the protein with each vaccine dose. You do need dystrophin continuously though (even though the cells are not dividing much, they are still recycling it).
You could argue for delivering gene therapy with mRNA/NLPs in multiple doses over the course of patient's life but that would likely 1) exacerbate toxicity and 2) be super-expensive
Do we have any studies that show this fast clearance? From what I understand at least one of them used a pseudo-uradine that there isn't an efficient direct metabolic pathway to process, which was kind of the whole point. The idea being it would circulate longer and be "more effective"
The uridine modification was intended to reduce immunogenicity of mRNA - some of our immune cells have pattern-seeking receptors in the TLR family that recognize ssRNA and dsRNA. The presence of modified uridines throws this pattern recognition off. (https://doi.org/10.1016/j.jconrel.2015.08.051)
The modifications to increase mRNA half-life concerned mostly the caps and poly(A) tail. But even with those the persistence was in the range of days (sort of depending on how sensitive a method you picked).
That's right, they use N1-Methylpseudouridine instead of uridine (the nucleoside contained in uracil, which is the U in mRNA sequences) to last a bit longer (but not forever) and to avoid triggering immune reactions to the mRNA itself (the immune system can detect foreign mRNA).
Certainly the vaccine's mRNA sequence breaks down into separate nucleotides. If it did not, continued production of the antigens would cause a chronic immune reaction and/or immune exhaustion that would make the vaccine ineffective.
I don't know what happens to the N1-Methylpseudouridine though. That's an interesting question.
> Certainly the vaccine's mRNA sequence breaks down into separate nucleotides. If it did not, continued production of the antigens would cause a chronic immune reaction and/or immune exhaustion that would make the vaccine ineffective.
I suspect you just described "long COVID" or "vaccine injury" for some fraction of folks.
Now, that particular study is in whatever cell line, highly dubious how it pertains to a human body, a few steps removed. But if you say "will you see this if you vaccinate 500 million times in 500 million people each with 500 trillion cells" - yea probably you would
Numerous studies have found vax-derived spike persisting for months and even years after vaccination, giving rise to concerns expression of spike can continue long after the claimed 24-48 hours.
A recent study found spike protein persisting for 17 months in the cerebral arteries of stroke victims. [1]
> In our study, in situ hybridization detected both mRNA derived from the vaccine and mRNA from the SARS-CoV-2 virus. ... our in situ hybridization method has high sensitivity and could detect trace amounts of mRNA, possibly reflecting unrecognized asymptomatic infections. These findings emphasize the need for caution in interpreting the presence of spike protein as exclusively vaccine-related.
We should also note that the study doesn't show that the original vaccine mRNA somehow survived for months, only that mRNA matching the vaccine sequence was detected by complementary probes.
I wonder if, in these cases, the vaccine was administered to someone with an active (but asymptomatic) COVID infection, and the vaccine mRNA was copied by the same RNA-dependent RNA polymerase that copies the viral RNA.
That might explain why both vaccine and viral RNA were found.
DeepMind/Google does a lot more than the other places that most HN readers would think about first (Amazon, Meta, etc). But there is a lot of excellent work with equal ambition and scale happening in pharma and biotech, that is less visible to the average HN reader. There is also excellent work happening in academic science as well (frequently as a collaboration with industry for compute). NVIDIA partners with whoever they can to get you committed to their tech stack.
For instance, Evo2 by the Arc Institute is a DNA Foundation Model that can do some really remarkable things to understand/interpret/design DNA sequences, and there are now multiple open weight models for working with biomolecules at a structural level that are equivalent to AlphaFold 3.
There are also techniques that combine both. In my experience (as an experimental structural biologist working in drug design), they frequently disagree.
It is possible that they are licensing technology that was developed in academic science and are raising money to scale it up and get it ultra-standarized for commercial scale.
I agree that the modern Silicon Valley model of VC funding has been spoiled by SaaS startups, where the capital expense is smaller, the timeline to exit is shorter, and pivots are easier. It is not great for deeptech innovation because those require more capital, time, and are more technology-constrained than software. Ironically, modern VC was developed to support semiconductor startups (1970s-90s), but has drifted from that technology-heavy origin.
Indeed, now is the moment to step on the gas in biotech. The past 15 years have been nothing short of extraordinary in the field. We finally have the tools needed to effectively measure biology, manipulate biology, and increasingly predict biology. More recently, we have been able to turn more and more problems into computational problems.
With all of this coming together, we should be accelerating both public and private investment in biotechnology because we're getting closer and closer to transformative therapies. But...we're failing to rise to the occasion and meet the moment.
The entire class of "biologics" drugs only came about in the past 15 years thanks to advances in sequencing and biotech. They are the mainstays of treatment for dozens of serious dermatologic, rheumatologic, and GI diseases, not to mention they directly cured multiple cancers.
Not op, but I’m in the field and can give you some things to read about:
- CAR-T
- CRISPR
- PRIME editing
- Base editing
- Modified mRNA
- PD-1 inhibitors
- On the cusp of personalized cancer vaccines
- ADCs
- Structure correctors
- Targeted protein degraders
- siRNAs
These have all really hit their stride in the past 15 years. Guess where all of them initially came from? Random ass government-funded academic research. Sure, you can split hairs with me on the 15 years and NIH/NSF etc funding, but it’s basically true. We are killing the golden goose…
Delivery vectors for nucleic acid have really progressed too. Peptide design and screening (also high throughput tools in general) have developed and led to great advances in peptide conjugates, such as peptide radioligands.
Sorry, I am not involved with biotech. Genuinely curious.
Edit. My impression of bio tech is that upfront costs are high and timeline for commercialization is long, and the only real biotech firm that I am aware of is Theranos. So I am probably coming from a place of ignorance.
reply