UR ASBMB Newsletter 04/25
Hello science team,
Welcome to the very first edition of what we hope will become your favorite monthly tradition — the UR ASBMB Newsletter! Each month, we’ll drop this delightful little bundle of science surprises and club updates straight into your inbox, absolutely free (yes, science and savings).
We’ll be sending future editions via UR CCC, so if you’re reading this, congrats — you’re already in the loop! No need to be an official member of the national ASBMB (though they have their own great newsletter, which you can subscribe to by creating an account here).
Our goal with this newsletter is simple and straightforward: to keep you up to date on the latest breakthroughs in biochemistry, microbiology, and molecular biology (and other areas), along with what's happening in our club. Each edition will feature:
- A short opinion piece (written by yours truly or a fellow club member)
- A heads-up on upcoming events
- A list of curated picks of the coolest recent publications (open-source only!)
- And much more coming up soon!
Got ideas for what we should include in future issues? We’d love to hear from you — fill out this form and let us know!
Since this is our debut issue, we’re pulling triple duty and spotlighting top articles from the past three months! Here’s what you’ll find in this issue:
- Upcoming Event(s) — April 2025
- Opinion: The Limits of AlphaFold
- Need to Read (March 2025)
- Need to Read (February 2025)
- Need to Read (January 2025)
Without further ado, let’s dive in! – JZ
Upcoming Event - 04/25
Future editions of the newsletter will include a calendar made by our lovely e-board members, but since we only have one event planned this April, I’ll tell you about it right here, right now.
On Friday, April 18, from 6:30 to 7:30 pm, you can join us in Douglas 403 for a fun night of Kahoot and protein folding! And yes, we have prizes you can show off to all your friends. This is an event you definitely don’t want to miss out on!
The Limits of AlphaFold
What happens when AI runs out of training data?
On April 1st (or as I like to call it, March 32nd), OpenAI released a new voice, Monday. It called me out for having the “emotional bandwidth of a worn-out USB cable”, which, honestly, fair. Just a day prior, on March 31st, SoftBank Group promised to invest more than $30 billion (that’s a 3 followed by ten zeros) into OpenAI1. As of now, OpenAI is valued over $300 billion — that’s more than two Pfizer’s (market cap: $150 billion). Maybe it’s time I switch my major to computer science. Then again, I enjoy having personal hygiene.
But this isn’t a tech newsletter — it’s a biotech newsletter. AI isn’t just for doing your homework and replacing your coworkers; it’s also revolutionizing the way we solve protein-folding nightmares. Meet PDB’s best friend: AlphaFold. Okay, you’ve probably met it already (more than once), so I’ll spare you the grand introduction.
AlphaFold 3 was announced way back in May 2024, and it’s apparently better at predicting interactions between proteins and nucleic acids. Still, it struggles with things like cofactors and posttranslational modifications2. Despite its limitations, no one can deny that AlphaFold is one of the most impressive scientific tools developed in the past decade.
But here’s the catch: AI models are only as good as the data you feed them. And we’re starting to hit a wall. AlphaFold 3 pulls from the 233,000+ protein structures publicly available on the PDB, plus a handful of others hidden away in private data vaults3.
Without fresh data, newer versions of AlphaFold (and its lesser-known cousins) will plateau. If we want to build AlphaFold 4, we’re going to need a whole lot more people squinting into cryo-EMs, fiddling with X-ray crystallography, and arguing over amino acid side chains.
Take drug interactions, for example. AlphaFold is notoriously bad at predicting protein-drug complexes. Why? Because pharmaceutical companies aren’t exactly racing to upload their hard-earned structures to the PDB. Only around 6% of PDB entries come from pharma3, and that number probably won’t climb anytime soon — intellectual property and all that.
Instead, some drug companies are developing their own private versions of AlphaFold. They've got the data (allegedly), but instead of sharing it, they’re training their own models and then sharing only the outputs — while doing everything they can to prevent others from reverse-engineering the training data. The kicker? They’re not even sure if it’ll work. If it does, expect them to shout it from the rooftops. If not… well, at least you read about it here.
I’ve always been a big proponent of open access and open source, so this kind of corporate secrecy doesn’t exactly fill me with joy. Still, if these private models genuinely lead to faster or more accurate drug discovery, maybe it’ll help with some of our public health woes — from antibiotic resistance to doctors prescribing “-mycins” for viral infections. But let’s be real: whatever breakthroughs happen behind those pharma firewalls probably won’t reach us academic folks anytime soon.
Beyond drugs and posttranslational modifications, there’s one other glaring omission from AlphaFold’s success story — and some biochemists have been very vocal about it: RNA folding. Yes, RNA folds too. Proteins aren’t the only ones with structural drama.
While PDB boasts over 233,000 protein structures, it has fewer than 2,000 RNA structures4. Granted, it's the Protein Data Bank, not the Ribonucleic Data Bank — maybe, just maybe, we’ll get an RDB someday if RNA biochemists complain loudly enough.
As we’ve established, less data means weaker AI models. RNA structure prediction tools like trRosettaRNA and RhoFold have popped up in the last couple of years, and they’re not bad5,6 — but they still lag behind AlphaFold’s performance due to the limited structural database. RhoFold even spawned an offshoot called RhoDesign, which tries to reverse-engineer an RNA sequence for a desired structure7. Of course, you still need to hit the wet lab to make sure RhoDesign didn’t gaslight you. So, we’re still waiting for that big AlphaFold moment… for RNA.
In the end, AlphaFold’s greatest strength — its ability to synthesize information from massive amounts of high-quality data — is also its biggest limitation. As we start to run low on open-access structural data, we risk slowing the pace of progress. That is, unless we find new ways to generate, share, and build upon our existing infrastructure. Whether it's protein-drug complexes or RNA, the future of AI in biology depends not just on smarter algorithms and smarter scientists, but on a community willing to collaborate across labs, institutions, and industries. Here's hoping the next breakthrough doesn’t sit behind a paywall.
On a final note, on the topic of collaboration, if you want easy access to AlphaFold at your fingertips, look no further than ColabFold, which is available on the web, for free!
You can access ColabFold here.
References:
- Announcement regarding follow-on investments in OpenAI [Internet]. 2025 [cited 2025 Apr 2]. Available from: https://group.softbank/en/news/press/20250401
- Bagdonas H, Fogarty CA, Fadda E, Agirre J. The case for post-predictional modifications in the Alphafold Protein Structure Database [Internet]. Nature Publishing Group; 2021 [cited 2025 Apr 2]. Available from: https://www.nature.com/articles/s41594-021-00680-9
- Callaway E. Alphafold is running out of data - so drug firms are building their own version [Internet]. Nature Publishing Group; 2025 [cited 2025 Apr 2]. Available from: https://www.nature.com/articles/d41586-025-00868-9
- Kwon D. RNA function follows form – why is it so hard to predict? [Internet]. Nature Publishing Group; 2025 [cited 2025 Apr 2]. Available from: https://www.nature.com/articles/d41586-025-00920-8
- Wang W, Feng C, Han R, Wang Z, Ye L, Du Z, et al. TrRosettaRNA: Automated prediction of RNA 3D structure with Transformer Network [Internet]. Nature Publishing Group; 2023 [cited 2025 Apr 2]. Available from: https://www.nature.com/articles/s41467-023-42528-4
- Shen T, Hu Z, Sun S, Liu D, Wong F, Wang J, et al. Accurate RNA 3D structure prediction using a language model-based deep learning approach [Internet]. Nature Publishing Group; 2024 [cited 2025 Apr 2]. Available from: https://www.nature.com/articles/s41592-024-02487-0
- Wong F, He D, Krishnan A, Hong L, Wang AZ, Wang J, et al. Deep generative design of RNA aptamers using structural predictions [Internet]. Nature Publishing Group; 2024 [cited 2025 Apr 2]. Available from: https://www.nature.com/articles/s43588-024-00720-6
Need to Read 03/25 – EBV and Chronic Inflammation
- Staff Pick: Acute Infections and Chronic Consequences
If you only read one paper from this issue, make it this one. For years, virologists like myself have suspected that some infections can lead to chronic, multi-system inflammatory diseases — and this study offers some compelling evidence. Researchers found, in severe cases of SARS-CoV-2 (especially in children with MIS-C), elevated TGFβ levels can reactivate latent Epstein-Barr Virus (EBV), leading to T-cell dysfunction. It’s a big step toward understanding how acute infections can trigger long-term immune devastation. Read it here: TGFβ links EBV to multisystem inflammatory syndrome in children!
- Proteasomes: Not Just for Presentation
Sure, proteasomes are best known for chewing up proteins and helping present antigens to your adaptive immune cells, but what if they’re also low-key defenders against bacteria? This study found that proteasome-derived defense peptides (PDDPs) can inhibit bacterial growth by disrupting membrane integrity — acting as an unexpected line of innate immunity. Think of them as your cells’ in-house broad-spectrum antibiotics. Read it here: Cell-autonomous innate immunity by proteasome-derived defence peptides!
- A Tumor’s Surprise Headache
Aspirin might be doing more than just fighting headaches — it could be helping the immune system fight cancer! This study shows that aspirin disrupts platelet signaling via the TXA2-ARHGEF1 pathway, which normally suppresses T-cell activity. By lifting this suppression, T cells are better equipped to reject metastatic cancer cells. It sure makes having an upset stomach slightly more bearable when you realize the same drug is also fighting cancer. Read it here: Aspirin prevents metastasis by limiting platelet TXA2 suppression of T cell immunity!
- A Nuclear Traffic Jam
The nuclear pore complex is like the bouncer of the nucleus — letting the right stuff in and out. Using two-color MINFLUX imaging, researchers discovered that import and export share overlapping regions within the pore. Surprisingly, the whole process is slower than previously thought, likely due to binding interactions or restricted diffusion. Still secure, but not exactly the Bundesautobahn. Read it here: Overlapping nuclear import and export paths unveiled by two-colour MINFLUX!
- Supermycin? No, Mandimycin
Fungal infections are becoming harder to treat, and new antifungal drugs are few and far between. Fear no more, Mandimycin is here! It’s a newly discovered polyene macrolide that targets multiple membrane phospholipids — a rare and promising mode of action. By disrupting fungal membranes in a broad-spectrum fashion, Mandimycin could become a valuable weapon in the fight against multidrug-resistant fungi. Read it here: A polyene macrolide targeting phospholipids in the fungal cell membrane!
Need to Read 02/25 — Calories, Codons, and COVID
- Binge Eat Today, Brain Fog Tomorrow
Turns out, that weekend binge might be messing with more than just your waistband. Researchers found that even short-term high-calorie diets can lead to increased liver fat retention and a noticeable reduction in brain insulin sensitivity — especially in memory-related areas like the hippocampus. What’s more, these effects linger even after returning to a normal diet, subtly altering reward behavior without changing diet patterns. Read here: A short-term, high-caloric diet has prolonged effects on brain insulin action in men!
- Keep Beating with Lab Grown Solutions
In a remarkable step toward regenerative medicine, scientists used lab-grown human heart muscle (derived from iPS cells) to repair chronically failing hearts in both rhesus macaques and humans. The grafts not only survived long-term (6+ months) — they also integrated with blood vessels and improved heart function. No major safety issues were observed, paving the way for a first-in-human trial with promising early results. Read it here: Engineered heart muscle allografts for heart repair in primates and humans!
- A Tale of Two Breaks
Meiotic recombination — the magical shuffle that mixes up our genes — starts with DNA double-strand breaks (DSBs) made by the protein Spo11. Scientists finally managed to reconstitute the mouse Spo11-Top6BL complex in vitro and confirmed its ability to bind DNA and cleave it (in an ATP-independent and Mg2+-dependent manner). A long-sought breakthrough for understanding one of biology’s most fundamental processes! Read it here: In vitro reconstitution of meiotic DNA double-strand-break formation!
- One Codon to Stop Them All
Who needs three stop codons when one will do? This team engineered a strain of E. coli (Ochre) that uses only the UAA stop codon, freeing up UAG and UGA for new functions — like incorporating noncanonical amino acids. The result? A cleaner, more programmable genetic code with high accuracy and a glimpse into a less redundant biological future. Could the days of degeneracy finally be over? Read it here: Engineering a genomically recoded organism with one stop codon!
- SARS-CoV-2: Omicron Boogaloo
SARS-CoV-2 isn’t done with us, yet. Using population data from Qatar, researchers tracked how immunity from prior infections held up over time. Before Omicron, people had strong and long-lasting protection against reinfection. But once Omicron took over, with its sneaky immune-evasion tricks (more than 50 new mutations), that protection waned fast — basically gone after a year. The takeaway? We’re going to need seasonal COVID boosters, much like the flu shot. Read it here: Differential protection against SARS-CoV-2 reinfection pre- and post-Omicron!
Need to Read 01/25 — Venom, Viruses, etc.
- Deep Learning vs. Deadly Snake Venom
Snakebites are infrequent, terrifying, and often really bad news (for the human, not so much the snake). Antivenoms are expensive, tough to store, and don’t always work well – especially against fast-acting toxins, such as the infamous three-finger toxins. A group of scientists have enlisted deep learning to design de novo proteins that can effectively neutralize these venomous villains. Better yet, they are easily synthesized and stored. No horsing around needed. Read it here: De novo designed proteins neutralize lethal snake venom toxins!
- Loneliness Hurts… Literally
We all know loneliness stings, but it turns out being alone also messes with your biology. A group of scientists investigated just how bad things can get when you constantly say “you’re busy” to your friends. Social isolation has skyrocketed in our screen-dominated world, so has negative health correlates like increased inflammation and chronic diseases. It turns out, being antisocial isn’t good for your body. Time to introduce yourself to the class and say three fun facts. Read it here: Plasma proteomic signatures of social isolation and loneliness associated with morbidity and mortality!
- Cancer’s Trojan Mitochondria
The more we learn about cancer, the more we realize how sneaky they can get. And this time, some cancers have figured out how to mess with your immune system’s power supply. By dumping faulty mitochondria into cytotoxic T cells, these cancers make it that much harder for your immune system to fight back in the long run. Clever yet infuriating. Read it here: Immune evasion through mitochondrial transfer in the tumour microenvironment!
- Batman Never Gets Sick from Ebola
Bats: the tiny mammals with superhero immune systems. While they carry viruses that could wipe us out, the bats somehow stay chill. The Bat1K genome project uncovered how bats pull off this biological balancing act — including antiviral genes that promote viral tolerance without causing the immune system to overreact. Take notes, humans. Read it here: Bat genomes illuminate adaptations to viral tolerance and disease resistance!
- The Aging Brain, Cell by Cell
What happens to our brains as we age? Using single-cell transcriptomics, researchers mapped how individual neurons and glial cells change over time (in mice). As it turns out, certain brain regions are especially vulnerable — such as the hypothalamus. Maybe one day we’ll see how these changes are mirrored in the human brain as well. Read it here: Brain-wide cell-type-specific transcriptomic signatures of healthy ageing in mice!
PS: If you made it this far, congratulations, you're my new best friend! If you scrolled through without reading the entire thing, shame on you! In any case, our upcoming May newsletter will feature an article on everyone's favorite birdflu. Stay tuned, stay healthy, and stay curious! - JZ
|
