Pinned Tweet
Our work on ENTER is finally out @CellCellPress! Like📱that can 📷📞🎧, ENTER is a modular viral platform that unifies multiple applications:
⭐️Decode Ligand-Receptor interaction
⭐️Receptor-specific cargo delivery
⭐️Link receptor specificity with cell state at single cell
⭐️Decode Ligand-Receptor interaction
⭐️Receptor-specific cargo delivery
⭐️Link receptor specificity with cell state at single cell
8
97
453
Bingfei Yu retweeted
While FIFA has captured everyone’s attention, something consequential happened in academia this week, especially for international students in PhD programs.
DHS issued a final rule changing how long F-1 students may remain in the United States.
A thread on what it means and what students and universities can do. 🧵
DHS issued a final rule changing how long F-1 students may remain in the United States.
A thread on what it means and what students and universities can do. 🧵
5
70
333
90,331
Bingfei Yu retweeted
Excited to share our new preprint on Hi-APEX — a truly H2O2-independent proximity labeling method!
We discovered a new probe, tetrazine-phenol, which enables APEX2 labeling without needing exogenous H2O2. Best part? No changes to the APEX2 enzyme are required.
We discovered a new probe, tetrazine-phenol, which enables APEX2 labeling without needing exogenous H2O2. Best part? No changes to the APEX2 enzyme are required.
3
13
59
6,663
Bingfei Yu retweeted
The results of the Nipah Protein Design Competition are out!
🧬 1200 proteins experimentally validated (3x more than last year)
📈 99 novel binders against the target protein (a challenging tetramer with little prior work)
💪 26 single digit nM or better binders, with the it picomolar affinity!
All data now available open-source on Proteinbase!
Let's take a look at the results ⬇️
🧬 1200 proteins experimentally validated (3x more than last year)
📈 99 novel binders against the target protein (a challenging tetramer with little prior work)
💪 26 single digit nM or better binders, with the it picomolar affinity!
All data now available open-source on Proteinbase!
Let's take a look at the results ⬇️
5
26
142
88,748
Bingfei Yu retweeted
I was trying to get Gemini 2.5 pro deep research to help me with the background research for a grant I'm writing. This report I got blew me away. I'm linking it here with my prompt that I used: docs.google.com/document/d/13H… It reads like a thought provoking review article.
19
31
245
34,726
Bingfei Yu retweeted
The peer-reviewed version of expansion in situ genome sequencing is now out in Science!
The news is bittersweet – when we first revealed this last September, I never guessed it would be my final paper in academia, but a lot has changed. A few parting thoughts:
The news is bittersweet – when we first revealed this last September, I never guessed it would be my final paper in academia, but a lot has changed. A few parting thoughts:
18
295
1,469
200,542
Bingfei Yu retweeted
Just a few months ago, the Baker and Garcia labs released con-concurrent preprints with incredible results. They designed de novo miniproteins that bound specific peptides presented in MHC-I.
Now let me show how you can use BindCraft at @AriaxBio to do the same thing.
Now let me show how you can use BindCraft at @AriaxBio to do the same thing.
9
86
569
57,857
Bingfei Yu retweeted
I saw this paper, and several people sent it to me. Quite similar to what I've been doing for a while, so I wanted to give my review! Which is honestly mostly a review of what I've been up to in comparison to that paper, but should give you a nice perspective from someone close to all this.
===
While both our methods use oligo pools, GoldenGate, and unique primer pairs to pull out subpools, they differ in that I pull out oligo subpools instantiating a single gene into a single well, whereas they pull out many genes into single subpools. I think this is simply because we are optimizing for different things.
My goal has been to get full clonal genes to people for ultra low cost. The main thing enabling that is sequencing cost reduction. For me, that means a cloning round costs, on order, $800 for 100 genes. $400 is for ~100kbp of DNA - this is the size which you can reasonably order the oligo pool often from commercial providers like Twist or Agilent, which is necessary if you want to ship genes to people often. About $400 in actual plastics / reagents / competent cells.
10% of the cost is in plastic (most of which is used at sequencing colonies), 7% on magnetic beads, and 21% on enzymes (including sequencing). If we only include build steps, I spend roughly $50 per 96 genes on enzymes (from NEB), $25 in plastic, and $25 in beads. So $100 per 96 in a roughly comparable way to their method. (35% is actually in competent cells, womp womp, working on making my own)
In Freschlin et al's method, each subpool contains multiple genes, but it appears the process is overall quite similar. The reason I did not go for this is because if I don't find a colony for a particular gene, I need to go back and pull it out to try again: but if I have it in a subpool, I have to go through synthesis again into a new subpool, or try to pick a shit ton of colonies. But it is true that their method, if you are fine with that limitation, dramatically decreases the cost of assembly by approximately 10x, while increasing the cost of screening IF you need clonal DNA. They don't, and so it truly does decrease the price per gene in a meaningful way.
The per-gene estimates on synthesis are approximately accurate with my own pool calculations, which you hit $1.58 per kbp with a $7,000 pool from Twist and $1.60 per kbp with a $9,833 pool from Agilent (factoring in the edges you need for GoldenGate enzymes and primer bind sites). If you want to do this commercially, note Twist will murder you if you sell products using their oligo pools and Agilent will ask to see the lease on your "lab space", as well as zoning documentation. Rough for a guy like me.
A key point in their paper:
"The final OMEGA-assembled genes likely contain errors originating from the starting oligopool. Most manufacturers report synthesis error rates between errors per nucleotide, most of which are point mutations." This is not true. Their citation comes from an oligo pool analysis off of IDT pools, but they use Twist pools: the mutational profile is different. In fact, if you look at most pools (from twist, to dynegene, to genscript), you see something similar, which shocks me about IDT, one pool I haven't analyzed - almost everyone has mostly indels.
Here is some real data from yours truly,
The subtlety comes here:```I picked 1536 colonies, ran my sequencing pipeline, got a total of 925 wells sequenced, 523 plasmids sequence correct, with 173 unique-and-correct plasmids (I picked a few colonies). In total, I sequenced 760,449 plasmid bases with a coverage of at least 5x, with a median of 4 colonies screened, mean of 4.57, with a standard deviation of 3.43.I got the following mutations: Point: 82, PointIndel: 22, Indel: 268, Insertion: 20Of the point mutations: map[{A C}:2 {A G}:5 {A T}:8 {C A}:7 {C T}:12 {G A}:6 {G C}:1 {G T}:33 {T A}:3 {T C}:2 {T G}:3]Mutation rate: 1 in 1939.920918367347Average gene size: ~1,000This was on a single flongle flow cell.```(one caveat, I had yet to implement my nanopore filtering yet, but those appear to mostly not be indels for my data set)
"Across all tested GG set sizes (20 to 70 sites), over 90% of sequenced assemblies were correct with most assemblies exhibiting >95% correct product... We assessed correct assemblies based on fragment order and did not consider sequence-correctness. Thus, these errors reflect the impact of misligation events on assembly fidelity". Now, I give them credit for the fact, since they weren't actually sequencing colonies after, they wouldn't have any way of knowing that, in fact, 31% of their assemblies have some form of indel/insertion. And that is with a 900bp gene! Imagine what it gets to once you hit those real big numbers of like 2.6kbp! (you don't have to: at 2.6kbp 66% will have some kind of indel/insertion)
On the other hand, they do say they sequenced the oligo pool with an illumina sequencer, but don't provide mutation rate data from that. I've never done that. I'd love to see the actual data! My process might just introduce a whole bunch of indels somewhere! My software might be wrong! Let's see the data.
Their method is a clever way to get around screening. My hardest challenge has been competent cells - I can't afford those big ole 50uL plates, so I've been trying to hack it together with cheaper cells, but those don't work so well. So I've started trying to make my own large batch competent cells. I didn't expect it to be competent cells, but my sequencing method worked first time (which I might talk about later), but the damn competent cells have been TOUGH (and before you ask, yes, I've tried most methods, and am going back to the classics in just large format).
One other thing I need to check: ligation fidelity has never been a problem. But ligation efficiency has been. I wonder if they have ligation efficiency checks in place.
===
While both our methods use oligo pools, GoldenGate, and unique primer pairs to pull out subpools, they differ in that I pull out oligo subpools instantiating a single gene into a single well, whereas they pull out many genes into single subpools. I think this is simply because we are optimizing for different things.
My goal has been to get full clonal genes to people for ultra low cost. The main thing enabling that is sequencing cost reduction. For me, that means a cloning round costs, on order, $800 for 100 genes. $400 is for ~100kbp of DNA - this is the size which you can reasonably order the oligo pool often from commercial providers like Twist or Agilent, which is necessary if you want to ship genes to people often. About $400 in actual plastics / reagents / competent cells.
10% of the cost is in plastic (most of which is used at sequencing colonies), 7% on magnetic beads, and 21% on enzymes (including sequencing). If we only include build steps, I spend roughly $50 per 96 genes on enzymes (from NEB), $25 in plastic, and $25 in beads. So $100 per 96 in a roughly comparable way to their method. (35% is actually in competent cells, womp womp, working on making my own)
In Freschlin et al's method, each subpool contains multiple genes, but it appears the process is overall quite similar. The reason I did not go for this is because if I don't find a colony for a particular gene, I need to go back and pull it out to try again: but if I have it in a subpool, I have to go through synthesis again into a new subpool, or try to pick a shit ton of colonies. But it is true that their method, if you are fine with that limitation, dramatically decreases the cost of assembly by approximately 10x, while increasing the cost of screening IF you need clonal DNA. They don't, and so it truly does decrease the price per gene in a meaningful way.
The per-gene estimates on synthesis are approximately accurate with my own pool calculations, which you hit $1.58 per kbp with a $7,000 pool from Twist and $1.60 per kbp with a $9,833 pool from Agilent (factoring in the edges you need for GoldenGate enzymes and primer bind sites). If you want to do this commercially, note Twist will murder you if you sell products using their oligo pools and Agilent will ask to see the lease on your "lab space", as well as zoning documentation. Rough for a guy like me.
A key point in their paper:
"The final OMEGA-assembled genes likely contain errors originating from the starting oligopool. Most manufacturers report synthesis error rates between errors per nucleotide, most of which are point mutations." This is not true. Their citation comes from an oligo pool analysis off of IDT pools, but they use Twist pools: the mutational profile is different. In fact, if you look at most pools (from twist, to dynegene, to genscript), you see something similar, which shocks me about IDT, one pool I haven't analyzed - almost everyone has mostly indels.
Here is some real data from yours truly,
The subtlety comes here:```I picked 1536 colonies, ran my sequencing pipeline, got a total of 925 wells sequenced, 523 plasmids sequence correct, with 173 unique-and-correct plasmids (I picked a few colonies). In total, I sequenced 760,449 plasmid bases with a coverage of at least 5x, with a median of 4 colonies screened, mean of 4.57, with a standard deviation of 3.43.I got the following mutations: Point: 82, PointIndel: 22, Indel: 268, Insertion: 20Of the point mutations: map[{A C}:2 {A G}:5 {A T}:8 {C A}:7 {C T}:12 {G A}:6 {G C}:1 {G T}:33 {T A}:3 {T C}:2 {T G}:3]Mutation rate: 1 in 1939.920918367347Average gene size: ~1,000This was on a single flongle flow cell.```(one caveat, I had yet to implement my nanopore filtering yet, but those appear to mostly not be indels for my data set)
"Across all tested GG set sizes (20 to 70 sites), over 90% of sequenced assemblies were correct with most assemblies exhibiting >95% correct product... We assessed correct assemblies based on fragment order and did not consider sequence-correctness. Thus, these errors reflect the impact of misligation events on assembly fidelity". Now, I give them credit for the fact, since they weren't actually sequencing colonies after, they wouldn't have any way of knowing that, in fact, 31% of their assemblies have some form of indel/insertion. And that is with a 900bp gene! Imagine what it gets to once you hit those real big numbers of like 2.6kbp! (you don't have to: at 2.6kbp 66% will have some kind of indel/insertion)
On the other hand, they do say they sequenced the oligo pool with an illumina sequencer, but don't provide mutation rate data from that. I've never done that. I'd love to see the actual data! My process might just introduce a whole bunch of indels somewhere! My software might be wrong! Let's see the data.
Their method is a clever way to get around screening. My hardest challenge has been competent cells - I can't afford those big ole 50uL plates, so I've been trying to hack it together with cheaper cells, but those don't work so well. So I've started trying to make my own large batch competent cells. I didn't expect it to be competent cells, but my sequencing method worked first time (which I might talk about later), but the damn competent cells have been TOUGH (and before you ask, yes, I've tried most methods, and am going back to the classics in just large format).
One other thing I need to check: ligation fidelity has never been a problem. But ligation efficiency has been. I wonder if they have ligation efficiency checks in place.
Gene synthesis is often the most expensive part of protein engineering with generative models.
Happy to have played a small part in this work, where Chase developed a method for precision library construction at scale, with per-gene costs as low as $1.50.
@romerolab1
9
39
208
41,109
Bingfei Yu retweeted
🚨BREAKING. From a program officer at the National Science Foundation, a list of keywords that can cause a grant to be pulled. I will be sharing screenshots of these keywords along with a decision tree. Please share widely. This is a crisis for academic freedom & science.
468
6,338
20,747
2,572,307
Bingfei Yu retweeted
One of the most important skills I teach my students&postdocs is how to communicate their research. Last week at NSF, I gave a keynote to (mostly) junior faculty on how to pitch their work. The talk resonated with many folks, so I'm sharing it publicly: mit.edu/~fadel/papers/…
12
124
807
84,036
Bingfei Yu retweeted
Out today in @jitcancer is an exciting an eye-opening project led by Kristin DePeaux, PhD, former student in the lab! In it, we found a peculiar feature of oncolytic virus immunotherapy: viruses can infect some populations of immune cells! A brief 🧵 1/ bit.ly/4fNNNtn
2
6
28
4,114
Bingfei Yu retweeted
🌟1/7 Check out our new pre-print “A cancer #immunotherapy modality based on dendritic cell #reprogramming in vivo”, where we reprogram #tumor cells in situ to efficiently present antigens as type 1 conventional dendritic cells (cDC1s). How effective is it?
A cancer immunotherapy modality based on dendritic cell reprogramming in vivo biorxiv.org/cgi/content/sh… #bioRxiv
1
18
79
12,593
Bingfei Yu retweeted
A new congressional blueprint for restructuring the NIH has been released. It proposes consolidating NIH's 27 IC's and centers into 15 categories, plus term limits for directors. It adds a national security review to grant apps. Public comments due Aug. 16 energycommerce.house.gov/posts/chair-ro…
1
40
49
97,630
Bingfei Yu retweeted
New preprint from our lab is now up on BioRxiv. This time it's a short but sweet one - a very useful method called POLAR-seq developed by @KlaudiaCiurkot in her 2 years in our group. We're keen to share this work early considering how useful it. biorxiv.org/content/10.110…
2
29
84
10,267
Bingfei Yu retweeted
So excited to finally share this project!
We've discovered #seekRNA -a new #RNA that can be reprogrammed to take any cargo DNA and insert it into any target DNA sequence. This has big implications for gene editing and biotech.
Read the fascinating story👇
biorxiv.org/content/10.110…
We've discovered #seekRNA -a new #RNA that can be reprogrammed to take any cargo DNA and insert it into any target DNA sequence. This has big implications for gene editing and biotech.
Read the fascinating story👇
biorxiv.org/content/10.110…
3
52
212
37,947
Bingfei Yu retweeted
📌Preprint alert!
Can naked extracellular RNA enter cells? Can it escape from endosomes into the cytosol? Can naked extracellular mRNAs be translated if not encapsulated in EVs or LNPs?
Why do we have so many RNases in the blood? 🤔
A thread 🧵🪡
biorxiv.org/content/10.110…
Can naked extracellular RNA enter cells? Can it escape from endosomes into the cytosol? Can naked extracellular mRNAs be translated if not encapsulated in EVs or LNPs?
Why do we have so many RNases in the blood? 🤔
A thread 🧵🪡
biorxiv.org/content/10.110…
6
28
123
26,250
Bingfei Yu retweeted
Can AI rewrite our human genome? ⌨️🧬
Today, we announce the successful editing of DNA in human cells with gene editors fully designed with AI. Not only that, we've decided to freely release the molecules under the @ProfluentBio OpenCRISPR initiative.
Lots to unpack👇
Today, we announce the successful editing of DNA in human cells with gene editors fully designed with AI. Not only that, we've decided to freely release the molecules under the @ProfluentBio OpenCRISPR initiative.
Lots to unpack👇
260
898
3,559
1,246,358
Bingfei Yu retweeted
NEW content online! Hallmarks of sex bias in immuno-oncology: mechanisms and therapeutic implications dlvr.it/T5Fz27
0
85
227
32,335
Bingfei Yu retweeted
Latest #BioRxiv on scRNA-seq reveals critical issues with droplet-based beads: Truncation compromises UMI identification, leading to inaccuracies in differential expression analysis. A must-read if your using 10X Genomics or dropseq in your research. tinyurl.com/mvapbxxw
7
38
136
35,281
Bingfei Yu retweeted
Today starts a new era of genome engineering!
We use RNA-only delivery to safely and efficiently deliver genes of interest to the human genome! @CollinsRNPLab @berkeleyMCB
A thread 🧵:
nature.com/articles/s4158…
We use RNA-only delivery to safely and efficiently deliver genes of interest to the human genome! @CollinsRNPLab @berkeleyMCB
A thread 🧵:
nature.com/articles/s4158…
3
107
415
54,688
