HN user
Justin Farlow cofounder of serotiny, acquired by J&J
Network effects + energetic fficiencies. On an energy landscape that includes integration over very short and very long lifetimes, the thalweg of utility/energy rests right about where the current codon optimizations are. And any schemes that deviate don't get to share in the others' bounty. Reusing your foods' effort saves a lot, metabolically.
This is a challenge, even for someone who has professionally used the breadth of proteins. I really like the test. I'm actually kind of surprised at my own pulling on knowledge to make a guess - it's an orthogonal way to think about the question than is usually posed.
I wonder if there's a way to ease the difficulty by filling in 'correct' features of the guesses: if your guess is a 'transmembrane' then it reveals that as a property. On the other hand, I don't think the annotations are clean enough - and are often designed for 'at all' rather than 'primary' features. For one of the examples, once I noticed it was an adhesion protein, it would have been interesting to sift through classes or cell types as opposed to just continuing to shoot in the dark based on the structure alone.
I presume you're showing even the 'low confidence' portions of the predicted structure? Please do.
You could also show the primary amino acid sequence too - there's a weird familiarity with those given how often the structures themselves have historically not been so accessible. BLASTING each of the guesses would be another interesting thing to see.
And the atoms in the proteins and DNA that are exactly replicated to the atom each have a feature sizes resolved at fractions of a nanometer in 3 dimensions (and likely in time/dynamics too).
Here's the full sequence of the protein, found in the supplement [1]
KSSEPASVSAAERRAETEQHKLEQENPGIVWLDQHGRVTAENDVALQILGPAGEQSLGVAQDSLEGIDVVQLHPEKSRDKLRFLLQSKDVGGSPVKSPPPVAMMINIPDRILMIKVSSMIAAGGASGTSMIFYDVTDLTTEPSGLPAGGSAPSHHHHHH
It is a protein encoding the PxRcoM-1 heme binding domain with C94S mutation and a C-terminal 6xHis tag (RcoM-HBD-C94S)
[1] https://www.pnas.org/doi/10.1073/pnas.2501389122#supplementa...
This is an AI-generated response, and is inaccurate.
That was one of the first cases of _germline_ gene editing using CRISPR - NOT "the first instance of gene editing." There have been quite a few other genetic editing tools that predate CRISPR, and there have been other edits using CRISPR that were not of the entire human's genome.
"Custom" in that this therapy was designed AFTER a specific patient showed a need, and then given to _that_ patient. In most every other context a particular class of disease is known, a drug designed, and then patients sought that have that disease that matches the purpose of the drug.
What's intriguing is not the 'custom' part, but the speed part (which permits it to be custom). Part of what makes CRISPR so powerful is that it can easily be 'adjusted' to work on different sequences based on a quick (DNA) string change - a day or two. Prior custom protein engineering would take minimum of months at full speed to 'adjust'.
That ease of manipulating DNA strings to enable rapid turnaround is similar to the difference between old-school protein based vaccines and the mRNA based vaccines. When you're manipulating 'source code' nucleic acid sequences you can move very quickly compared to manipulating the 'compiled' protein.
It also can actually allow you to identify positions within the image at a greater resolution than the pixels, or even light itself, would otherwise allow.
In microscopy, this is called 'super-resolution'. You can take many images over and over, and while the light itself is 100s of nanometers large, you actually can calculate the centroid of whatever is producing that light with greater resolution than the size of the light itself.
Serotiny: https://en.wikipedia.org/wiki/Serotiny
to build something but to not know how it actually works and yet it works.
Welcome to Biology!
Clearly RAM-stingy Apple has found a use for RAM - almost certainly in loading local LLMs.
Llama 8G loads and runs pretty well on the new M-series Macs with a reasonable amount of RAM.
There are a number of companies working on 'in vivo' deliveries for CARs. Oftentimes using the same tools as proven out by the Moderna vaccine.
There are two kinds of personalization in a [CAR] [T] therapy:
1) using the patient's own cells [personalization of T Cells]
2) customized therapeutic genetic payload, per patient [personalization of the CAR]
There are current competing factions for #1 - where cells are from just the patient ["Autologous"] (safer, slower, more expensive), and where the cells are from a universal donor ["Allogeneic"] (possible immune response, but can be manufactured at scale).
The therapeutic payload is a DNA sequence encoding a synthetic chimeric receptor ["CAR"]. This sequence is customized based on the details of the patient's particular cancer, but are common across many people. If the cancer has an excess of "Protein X" on it, then the CAR sequence is designed to target Protein X. All patients with a similar cancer profile receive the same CAR sequence as a payload to the T cells. This too could be personalized, to not just profile the _class_ of cancer, but particular to that _specific patent's cancer's profile_ - but this is not yet feasible given the turnaround time to build, test and evaluate a new genetic payload in the context of a person's specific tumor cells.
This particular therapy has the cells be from the patient (personalized), but the CAR sequence provided to the cells is common for all people that have the same cancer profile (semi-personalized). In this case, the cancer profile includes those that have an abundance of the protein called Claudin-6.
I've designed a device that utilizes mechanical force to transmit information that was around 5nm in diameter. It was based on the human Notch receptor. It's a few hundred amino acids in length, folded to produce a protein that senses force transmission, is cleaved upon unfolding, and releases a transcription factor the nucleus of a cell.
I kind of find the distinction of 'robots' vs cells funny, as once you get down to the (sub)nanometer level one's intuition should flip: organic material acts stiffer and more lego-like than metals - which act more like unreliable putties. A "device" that becomes small enough is much more likely to be made of organic molecules than metallic molecules - cells ARE those futuristic robots...
The kinesin motor proteins are pretty cool too [1], but those are naturally occurring machines that I suspect we'll be imitating for a long time.
Using the above chemistry, you can attach DNA oligos to lipids, DNA oligos to proteins, fluorophores to DNA, fluorophores to proteins at particular locations or other complex drugs to DNA, protein or lipids.
Once you get DNA oligos in there you can do computation, as X binds X', and Y to Y'. So you can have all sorts of complex synthetic & designed interactions using chemistry that is both seamless and doesn't interfere with normal molecular biolgy.
Once you have proteins, you can localize particular chemistries.
Serotiny | Remote (US), Bay Area, CA | Full-Time | https://serotiny.com
Serotiny invents synthetic proteins to treat cancer and genetic diseases. We have a novel approach to designing proteins that couples synthetic biology, high-throughput screening, and machine learning. We've had success in both cell and gene therapy contexts.
We're a cross-functional team of scientists and developers and we're looking to expand our NGS processing capabilities & scale our design software. Please reach out of you're interested in this role or software development in biotech: jobs_platform@serotiny.com
Serotiny | Bioinformatics Scientist - Next Generation Sequencing + Protein Design | Remote (US), Bay Area, CA | https://serotiny.com/
Serotiny invents synthetic proteins to treat cancer and genetic diseases. We have a novel approach to designing proteins that couples synthetic biology, high-throughput screening, and machine learning. We've had success in both cell and gene therapy contexts.
We're a cross-functional team of scientists and developers and we're looking to expand our NGS processing capabilities & custom protein design software. Please reach out of you're interested in this role or software development in biotech: jobs_platform@serotiny.com.
Serotiny | Bioinformatics Scientist - Next Generation Sequencing + Protein Design | Remote (US), Bay Area, CA | https://serotiny.com/ Serotiny invents synthetic proteins to treat cancer and genetic diseases. We have a novel approach to designing proteins that couples synthetic biology, high-throughput screening, and machine learning. We've had success in both cell and gene therapy contexts.
We're a cross-functional team of scientists and developers and we're looking to expand our NGS processing capabilities & custom protein design software. Please reach out of you're interested in this role or software development in biotech: jobs_platform@serotiny.com.
Serotiny | Bioinformatics Scientist - Next Generation Sequencing + Protein Design | Remote (US), Bay Area, CA | https://serotiny.com/
Serotiny invents synthetic proteins to treat cancer and genetic diseases. We have a novel approach to designing proteins that couples synthetic biology, high-throughput screening, and machine learning. We've had success in both cell and gene therapy contexts.
We're a cross-functional team of scientists and developers and we're looking to expand our NGS processing capabilities. Please reach out of you're interested in this role or software development in biotech: jobs_platform@serotiny.com.
The Base Editor (and Prime Editor, etc.) _is_ CRISPR (Cas9) but with additional components fused to it that provide it additional features allowing it to edit in a more elegant way over the naked Cas9.
The number of [nucleic-acid-delivered] gene therapies in Phase II & Phase III trials right now is huge - because of this progress in delivery [of nucleic acids]. Gene therapies for the eye, for hemophilia, for sickle cell, many many cancer therapies all rely on the ability to 'deliver' nucleic acid payloads to cells. Of those, only 3 or 4 have been approved - and all in the past 2 years, but there are a huge number that are behind that tip of the iceberg - quite precisely because it's relatively straightforward to do 'same thing but with a different sequence' once the first one works.
Delivery of the RNA is hard. To the right cell type, not immediately degraded, not accidentally integrated into a critical part of the genome, with a payload that is actually effective, etc.
The original gene therapies (early 2000s) were essentially RNA therapies (adenovirus). And their unethical rush and subsequent failures caused a bit of a 'gene therapy winter' [1]. We've since made enormous progress on both the ability to safely deliver genes, but also our ability to generate/design new useful genes.
[1] https://www.labiotech.eu/in-depth/gene-therapy-history/
In 1972, a paper titled ‘Gene therapy for human genetic disease?’ was published in Science by US scientists Theodore Friedmann and Richard Roblin, who outlined the immense potential of incorporating DNA sequences into patients’ cells for treating people with genetic disorders. However, they urged caution in the development of the technology, pointing out several key bottlenecks in scientific understanding that still needed to be addressed.
Even the cells infected with the mRNA will need to be removed by the immune system using cytotoxic T-cells
I don't think that this is accurate.
In short - certain ones, yes. This should be one step (that was a bottleneck) in helping a company with a fixed budget do an order or magnitude more 'experiments' with the same amount of resources. Lab resources are expensive and fixed, so if you can pre-compute what you need, you can get right to the more powerful results.
We design proteins for immunotherapies - this kind of thing would help us more rapidly design our proteins (and more efficiently use our wet-lab resources to speed existing projects). For others, some drugs are hard build without knowing how they will interact - this could both provide new 'targets' to go after, but also might help prevent projects that would otherwise accidentally target an important protein.
Very cool! I've actually built similar models for myself to build that intuition. They're so helpful and generally hard to find. I agree that scale is generally poorly taught.
I personally think having even a modest intuition for how to map physical knowledge to its appropriate 'scale' (time/space, from plank to universe) is one of the most straightforward ways to be "smart". And a great way to get to know the limits of our knowledge.
I still haven't seen a video that matches my intuition. Which is unfortunate. I want to build a version of your scaler there, but for VR that you can slide up and down along at least time/scale axes - and maybe additional ones too.
Goodsell's book, "Machinery of Life" is fantastic at giving a first-pass representation of the basic interactions in a cell that are otherwise invisible.
Search space for the first billion or so years was likely around atoms/chemistry. Search space for the next billion or so years was likely around mutations in well-defined genetic space (4 nucleic acids as building blocks, 20 amino acids as building blocks). Search space for the past 700M years has included 'recombination' (copy/swap/paste) of larger chunks of (existing) genetic code (in addition to the prior two mechanics). There are probably many such abstraction layers, including like you mention, copying/pasting of entire organisms, environments, biospheres...
Brownian motion dominates at the scale of a protein. In other words, the energy imparted by the ambient temperature is orders of magnitude greater than that that could be established through momentum of objects with that little mass. So everything bumps into everything, and the only distinguishing factor between various bumps is how long they stay stuck together. Most things just bounce right off. Sometimes they are repulsed, and in rare cases (when they've been selected for), multiple proteins can meet in just the right orientation that they stay stuck for a while (now scale this so that 3 or 4 objects must meet at the same time, and now you get into the 'regulation' of a cell at perception-relevant time-scales). This measurement of stickiness is called the "dissociation constant (Kd)" and is a measurement of what concentration of component A you need alongside component B such that half of component A is bound to component B. It is one of the primary drivers of most biochemical processes in a cell.
One issue is that the scales don't well-match. And there are actually many time-scales beneath us, not just a few (and that biology/life is working at them all - including those on the order of millions of years.
An event like a "protein folding" can take milliseconds, in a tube [1]. While atomic/biophysical/biochemical simulations have time-steps of femto-seconds.
[1] https://www.youtube.com/watch?v=gFcp2Xpd29I
From https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3890418/
"Traditional MD [Molecular Dynamics (protein folding)] simulations are limited in length by timestep limits. Studies by our group and others have shown that traditional MD is limited to timesteps of about 2 fs due to high-frequency resonance frequencies.1–3 Many biologically relevant motions occur on the microsecond to millisecond range, which is 9 to 12 orders of magnitude greater than the timesteps possible with traditional MD. Further, each step requires a costly force calculation (O(N) to O(N2)). As such, simulating medium-size proteins often requires months of computer time on a large distributed cluster such as Folding@home4,5 to simulate milliseconds of dynamics, while simulating a large protein (e.g. the β-2 Adrenergic Receptor) on biologically-relevant time scales (milliseconds through hours) using a standard desktop computer would take years. Thus, it is not feasible to simulate timescales of biological interest without substantial advances in MD methods."
12 orders of magnitude of difference in time is akin to the difference between causal events happening once per millisecond and those happening once per century. How do you show a movie capturing the nuance of someone's blink reflex, along with their birth life and death...
Most vaccines come with an 'adjuvant' [1] that activates the immune system in order to encourage the development of antibodies.
Sometimes these are chunks of toxins that your body recognizes a toxic, but without the actual part of the toxin that causes the toxicity. A kind of flag but without the army behind it.
^ That is a _fantastic_ talk for someone who know's some physics get a quick grasp of the state of the field, the different approaches being taken from a set of first principles, and how to evaluate them.