Luka Gebel, Chander Velu & Antonio Vidal-Puig, The strategy behind one of the most successful labs in the world, Nature, Vol 630 | 27 June 2024, https://doi.org/10.1038/d41586-024-02085-2
The opening paragraph:
The Medical Research Council’s Laboratory of Molecular Biology (LMB) in Cambridge, UK, is a world leader in basic biology research. The lab’s list of breakthroughs is enviable, from the structure of DNA and proteins to genetic sequencing. Since its origins in the late 1940s, the institute — currently with around 700 staff members — has produced a dozen Nobel prizewinners, including DNA decipherers James Watson, Francis Crick and Fred Sanger. Four LMB scientists received their awards in the past 15 years: Venkatraman Ramakrishnan for determining the structure of ribosomes, Michael Levitt for computer models of chemical reactions, Richard Henderson for cryo-electron microscopy (cryo-EM) and Gregory Winter for work on the evolution of antibodies (see Figure S1 in Supplementary information; SI). Between 2015 and 2019, more than one-third (36%) of the LMB’s output was in the top 10% of the world’s most-cited papers.
After much reporting:
Although these rules govern the LMB, the outcomes are more than the sum of their parts. The organization’s management strategy gives rise to emergent behaviours and deliverables that align with its long-term research goals. The management model has emerged from a set of actions taken by management over time that collectively result in a coherent approach to achieving the overall aim of the LMB4. In management theory terms, the LMB is a complex adaptive system, similar to an ecosystem.
A complex adaptive system is a self-organizing system with distinctive behaviour that emerges from interactions between its components in a manner that is usually not easy to predict5. Components might include individuals and their activities; material parts, such as technologies; and the ideas generated from these interactions6.
Effective management of this complex adaptive system is fundamental to the LMB’s success. Through continual adaptation and evolution, the LMB can generate new knowledge more effectively than most other institutions can.
Final paragraphs:
In recent years, some funders have pulled out of basic bioscience. For example, more of the US National Institutes of Health’s extramural funding over the past decade has gone to translational and applied research than to basic science (see Science 382, 863; 2023). Some highly reputable basic-science research institutions have suffered as a result and have even been dissolved, such as the Skirball Institute in New York City10. However, it is crucial to resist the temptation of dismantling basic science research, considering the complexity and difficulty of re-establishing it.
In response, a lab such as the LMB might enhance the translation of its discoveries by strengthening connections with the clinical academic sciences and private-sector industries. Leveraging strengths in the pharmaceutical industry — in areas such as artificial intelligence and in silico modelling — can bolster basic science without compromising a research lab’s focus. The LMB’s Blue Sky collaboration with the biopharmaceutical firm AstraZeneca is a step in this direction (see go.nature.com/3rnsvyu).
Third, it is becoming increasingly challenging for basic science labs to recruit and retain the best scientific minds. Translational research institutes are proliferating globally. Biotechnology and pharma firms can pay higher salaries to leading researchers. And researchers might be put off by the large failure rates for high-risk projects in fundamental research, as well as by the difficulties of getting tenure in a competitive lab such as the LMB.
Bill, I'm keen to hear your take on compute vs humans. Or proxy - money
ReplyDelete"ESM3: Simulating 500M years of evolution with a language model (evolutionaryscale.ai)33 points by wawayanda 1 day ago | hide | past | favorite | 9 comments
https://news.ycombinator.com/item?id=40787800
On the success of raw compute and immense data NOT in the Laboratory of Molecular Biology (LMB) in Cambridge, UK.
AWS - Bezos + Nvidia - Jensen Huang + 2 x money guys = LMB will become a customer and slowly be crushed under the weight - 5 tonnes - of compute - "10^24 flops, 1 trillion teraflops"
ESM3
A NEW TYPE OF LANGUAGE MODEL
Enabling scientists to understand, imagine, and create proteins.
Biology is fundamentally programmable. Every living organism shares the same genetic code across the same 20 amino acids—life’s alphabet. ESM3 understands all of this biological data, translates it, and speaks it fluently to be used as a generative tool
A model trained across all of evolution.
10^24 flops
1 trillion teraflops
of computing power.
2.78 billion
natural proteins sampled from various organisms and biomes.
771 billion
unique tokens of training data
98 billion
parameters in our largest evolutionary-scale model.
https://www.evolutionaryscale.ai
https://www.evolutionaryscale.ai/blog/esm3-release
November 29, 2023
AWS Teases 65 Exaflop ‘Ultra-Cluster’ with Nvidia, Launches New Chips
https://www.datanami.com/2023/11/29/aws-teases-65-exaflop-ultra-cluster-with-nvidia-launches-new-chips/
Not sure what you're asking. I don't know enough about molecular biology to have a serious opinion about the use of transformers to build models over strings encoding molecular structures. On one level, a string is a string & transformers can find latent structure in large collections of strings. So the use seems plausible. Beyond that....
ReplyDeleteI suppose I am asking will raw compute relegate humans to trials of manufactured protiens instead if using a lab ala Cambridge. No need of knowledge of biology.
DeleteEvolutionary Scale is touting 500m years of evolution done in... a week. With pockets deep enough to buy out Cambridge.
In this context I think "evolution" is doing a lot of work that's not very well-defined.
Delete