The two golden rules of using AI in content creation: know your shit, and know what's shit. www.linkedin.com/pulse/two-go... Inspired by some particularly egregious recent incidents I experienced, and @naomialderman.bsky.social writing about the importance of discernment when using these tools.
We’re launching the Trillion Gene Atlas with Anthropic, NVIDIA, PacBio, and UltimaGenomics.
🌍 The Goal: To expand known genomes 100×, sequencing 100M+ species to build foundation models grounded in evolution itself.
🔗 Read more: www.prnewswire.com/news-release...
Introducing EDEN — a 28B-parameter biological foundation model trained on 9.7T nucleotide tokens, including >10B novel genes from >1M species. This evolutionary scale is what makes generalisable biological design possible.
Paper: www.biorxiv.org/content/10.6...
EDEN: a family of genomic language models trained on up to 9.7 trillion nucleotides from @basecamp-research.bsky.social's BaseData can design large serine recombinases, bridge recombinases, and antimicrobial peptides.
www.biorxiv.org/content/10.6...
Happy to have played a small part in this!
How can AI and nature’s hidden code lead to programmable gene therapies?
Basecamp Research Co-founder & CEO GlenGowers.bsky.social explores this question at @financialtimes.com Times Live Global Pharma and Biotech Summit (Nov 11–12) with industry leaders.
👉 For full details, visit bit.ly/4pPcJpd
A real shame that the future of CASP is in doubt. There is still a lot of progress to be assessed (complexes, nucleic acids, ligands, ensembles). CASP has also been a pioneer in how to run blind assessments.
www.science.org/content/arti...
Research public spending (not limited to medical research), US spends ~ $760 billions and combined EU ~ $400 billions (not including UK), so there is indeed a big gap, but the graph is very misleading as it compares US funding with only samples of European funding.
Seen on LinkedIn, a bit on the nose, comparing Machine learning to AI and now GenAI and AgenticAI
First roses of the year.
youtu.be/aW2esYwKxiU?...
We are recruiting @ox.ac.uk @dunnschool.bsky.social for a joint project between Kevin Foster's group and mine! Cluster hire of two postdocs interested in:
1) Mucosal immunology and oral vaccine development
my.corehr.com/pls/uoxrecru...
2) Microbiology
my.corehr.com/pls/uoxrecru...
Please share!
youtu.be/G-lh0TUTtvI?... original speech in French
youtu.be/q32Zz2dBkos
If you want to know how Europeans feel listen to Claude Malhuret, He is the leader of the independents in France. This is how they see the US now.
That's the full translation.
Worth a listen
I feel that it is more akeen to a Molotov-Riebentropp like agreement between Trump and Putin.
Francis Collins, the NIH Director for 12 years, led the Human Genome Project and other NIH efforts for 32 years, resigned today. Key words from his resignation letter
www.nytimes.com/2025/03/01/u...
I wonder if it is still available from within the US? The error is a DNS issue, like the path for the browser to find the site had been deleted. If it is the case it is very worrying.
Not working for me in the UK, tried bypassing DNS and still no luck. Looks more serious than just server being down. And it is not only pubmed, but the whole NCBI page.
Pass it on.
#StandWithUkraine
There are many reasons for the decrease in reduction of NO2, probably concomitant, urea injection could be quanfied by measuring increase in ammonia in the air, might not be that simple as ammonia might react with other atmospheric components/pollutants.
Apparently, Python is a particularly tricky programming language for LLMs as indentation is important and can be achieved in different ways (different numbers of \s or \t as long as consistent within the same code), this screws up tokenisation.
Either:
1/ Most NO2 is produced in Paris and other parts of the region benefit from these rules.
2/ Other parts of the region have implemented the same rules, to the same effect.
3/ These rules have negligible effect and the decrease is due to other reasons.
I tried to ask Gemini to optimise a sample python code I wrote. Interestingly, the logic offered to improve the code made actual sense, however the code suggested did not reflect that logic and was not functional. I suspect, it doesn't have logic, just glorified auto completion.
Favourite paper of 2017
"Protein structure determination using metagenome sequence data" by Ovchinnikov et al. (1/4)
www.science.org/doi/10.1126/...
Good article on the OpenFold work of @moalquraishi.bsky.social, who poses the interesting question: If we can predict how proteins fold without understanding how they do it, “are we even legitimately doing science anymore, or is it something different?“ magazine.hms.harvard.edu/articles/did...
I'm a bit concerned about that red alpha helix sticking out like a sore thumb, doesn't look too realistic.
Basecamp Research cofounder and CEO Glen Gowers, Ph.D., spoke about AI and drug discovery at the 2024 Forbes Healthcare Summit in New York City. #ForbesHealth
trib.al/aVWj2tF
Thanks to @forbes.com and @thealexknapp.bsky.social for an amazing summit a thought-provoking event.
Conclusions from RNA section at #CASP16, without a template all fail. All small targets had a template.
