Our AAV paper published in Science on Thanksgiving, 4+ years with @PierceOgdenJ, @samsinai, @geochurch, continuing now at @dyno_tx, here’s how: including twists, turns, surprises, and many people to thank along the way!https://science.sciencemag.org/content/366/6469/1139 …
-
-
I joined
@geochurch lab early 2015 aiming to combine 3 technologies for protein engineering: chip-based#DNA synthesis,#NGS &#MachineLearning. I teamed up with the talented@PierceOgdenJ, and@samsinai who joined about a year later to enrich our lives with his ML wisdom.pic.twitter.com/7O3UqEf5oM
Prikaži ovu nit -
(In the course of a epic project like this one, so many people were happy to help in 1e6 different ways, I'll try to acknowledge as many as I can while setting the stage for how and why we set off on this scientific adventure...)
Prikaži ovu nit -
We saw amazing potential combining these 3 technologies to study biology & for engineering, inspired by: PhilRomero,
@francesarnold,@jbkinney,@segal_eran,@srikosuri,@dbgoodman,@jbloom_lab,@dougfowler42,@lea_starita,@erezaterez,@JShendure, @YangKevinK,@deboramarks, ...Prikaži ovu nit -
Why? Proteins are complex and still mysterious in many ways. How can we smartly engineer when we can’t predict what happens when making even a single change to a protein sequence? High-throughput measurement technologies can fill the void in our understanding. But how to focus?
Prikaži ovu nit -
Following
@paulg’s “make something people want”, I was searching for the perfect protein family: 1) state-of-the-art very suboptimal 2) many applications 3) people need it (enough to pay $$$), 4) hard to make it (otherwise why even develop a new technology?), 5) health relevantPrikaži ovu nit -
Discovered
#AAV thanks to@ChewWeiLeong's early work in@geochurch lab delivering#CRISPR#Cas9 with#AAV. It was a perfect fit, an elegant and minimal genetic system, plus bonus: built in genotype-phenotype coupling (capsids package their own genomes!)pic.twitter.com/lEeC1m4Bz4
Prikaži ovu nit -
More than just being interesting, I wanted to be sure there really were real world applications. Thanks to ElizabethLeshen, WhitneySnider, SarahHonig & ElenaViboch who helped me understand the business of
#AAV#GeneTherapy w/@VickiSato in@HarvardHBS CommercializingSciencePrikaži ovu nit -
+ thanks to many who volunteered their time and opinions to set us on the right path! ShenShen, AriFriedland,
@ksbosley,@mannysimons, JamesMcLaughlin, @PKolchinsky, BenAuspitz, FeliciaPagliuca, ...Prikaži ovu nit -
The foundations of
#GeneTherapy today are built on natural#AAV capsids, but these aren’t optimal in many ways: we want them to target new cell types more efficiently and specifically. Plus avoid the immune system, package more DNA, and be cheaper to produce.Prikaži ovu nit -
#AAV capsids are complex! 60 monomers assemble around the DNA genome, then capsids must navigate within the body to cellular targets, enter cells within an endosome, break free of those endosomes, get into the nucleus and finally release their payloads...Prikaži ovu nit -
This multiplicity of critical functions makes
#AAV challenging to engineer. If you improve only a single function, but the other functions get worse, that’s no good. And this happens often, because most changes to the protein are deleterious!Prikaži ovu nit -
Standard protein engineering approaches like rational design and random mutagenesis favor libraries of capsids of either high quality or of high quantity, but struggle to achieve both simultaneously. We thought we could add something unique to the field...
Prikaži ovu nit -
It’s an amazing time to be in
#GeneTherapy. We benefited hugely from pioneering works by HiroyukiNakai, LukVandenberghe, AravindAsokan, TomasBjorklund, JimWilson, DirkGrimm, SergeiZolotukhin,@Nicole_Paulk, ThomasWeber,@Thieum001,@bendeverman, + many discussions w/ samePrikaži ovu nit -
Inspired by work from my PhD with
@hattaca,@jimmyx66, NivCohen,@harriswangnyc and@RoyKishony, we knew that a comprehensive scanning library of all possible single mutations would be blueprint for understanding how to engineer the#AAV capsid https://www.ncbi.nlm.nih.gov/pubmed/28009265Prikaži ovu nit -
Just substitute, insert or delete every possible amino acid one by one at each position across the capsid protein, about 30k variants. Simple idea. Two years later, and...success!pic.twitter.com/cZP5h05sGV
Prikaži ovu nit -
Actually, it wasn’t that easy. Biggest challenges occurred during many failed attempts to link mutations and barcodes (beware template switching). We cloned the library 3 different ways before settling on a synthesis-and-split approach that beautifully solved the problem.pic.twitter.com/ybZQBX9U2v
Prikaži ovu nit -
Directly synthesizing the library, we were able to make 99.9% of all single amino acid mutants. Once we had the library assembled, we could measure the function of all variants in a single experiment, even in a single animal! Our goal was to get PoC measurements in mice.
Prikaži ovu nit -
But first came a BIG surprise. Rather than making only amino acid changes, we incorporated all codon changes, thinking these would reveal non-coding effects. Here you can see how mutations in synonymous codons indicate the presence of something odd, definitely a pattern here...pic.twitter.com/mB4ouucH9e
Prikaži ovu nit -
Long story short: we found a new gene in the early VP1 region of the capsid, which had gone unnoticed during more than 50 years of studying
#AAV. One of my most exciting surprises of my career!Prikaži ovu nit -
Back to our main story, our goal was to use the library to understand how
#AAV works in vivo. Could we learn anything about why capsids go toward particular organs within the body?pic.twitter.com/vPnQsibTXK
Prikaži ovu nit -
We injected the virus library into mice to measure biodistribution: most mutants clustered into a few distinct profiles. In other words, it should be possible to build predictive models from a relatively small number of measurements! A promising sign for
#MachineGuideDesign.pic.twitter.com/9bBCTCUvSU
Prikaži ovu nit -
Together with
@samsinai, we put this idea to the test. We designed new multi-mutants sequences (with many changes) using data from our single-mutant dataset. Would this data enable us to build better libraries than a random approach?pic.twitter.com/xYLH0bB1S3
Prikaži ovu nit -
As we had hoped, our
#MachineGuidedDesign strategy was much better at generating viable mutants than were randomly generated sequences. This demonstrates that such data can accelerate the search for synthetic#AAV capsids that have improved properties for#GeneTherapy.pic.twitter.com/XFculK8HCx
Prikaži ovu nit -
Libraries like these will make great training data for machine learning algorithms, for even more informed engineering. Stay tuned for exciting future updates.
#DeepSequenceSpace meets#DeepLearning.Prikaži ovu nit -
Almost a year ago, I left
@harvardmed to focus on translating these technologies into making optimized#AAV for human therapy. Joined by amazing Co-founders including@samsinai,@SSlomovic,@Adrian_Veres_,@geochurch and now many more at@dyno_tx, http://www.dynotx.com pic.twitter.com/4e6OI7xcMh
Prikaži ovu nit -
Many efforts underway to improve these technologies and scale up our high-throughput experimental + data pipelines and
#MachineLearning models. We are rapidly growing and actively hiring! Check out open positions http://dynotx.com/#careers pic.twitter.com/nUkMNFmzMb
Prikaži ovu nit -
Thanks to the many folks in
@geochurch lab,@wyssinstitute and beyond who helped make this possible!@eskamoah,@SurgeBiswas,@glebkuz, NinaJain,@dbgoodman, MaxSchubert, DavidThompson, IsaacHan, DenitsaMilanova, JohnAach, JayCulverwell, MaryTolikas, AyisAntoniou!Prikaži ovu nit -
Anyone interested to learn more check out this review I wrote with
@geochurch on#MachineGuidedDesign for#AAV! https://insights.bio/cell-and-gene-therapy-insights/journal/articles/challenges-and-opportunities-of-machine-guided-capsid-engineering-for-gene-therapy/ …Prikaži ovu nit
Kraj razgovora
Novi razgovor -
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.