Photo by National Cancer Institute on Unsplash
This Musing is about earlier-stage invention and geekier than usual, but it gives a glimpse of the massive wave that's coming.
The combination of medical research and AI, particularly Large Language Models (LLMs). This Musing will only touch the surface of what's happening, but it will point the way to the massive advance that's coming to medicine. There are lots of categories we could cover, but here are 5 particularly interesting categories.
CAR-T signaling domains
We talked about CAR-T before, but that was just one successful event for an obscure condition. It leads off because it is the closest to a useful product. Earlier stage research is proliferating using CAR-T.
Chimeric antigen receptor T cell technology, in which cells of the immune system are modified with customized receptors, has proved effective in cancer therapy. To explore the range of cell responses that can be encoded in such receptors and to make their design more quantitative and predictive, Daniels et al. tested about 200 of 2400 possible combinations of 13 signaling motifs found in such receptors and used machine learning to predict other effective combinations. Using these design rules, the authors constructed receptors in human T cells with improved signaling characteristics that contributed to better tumor control in a mouse model.
Cancer tends to obscure itself from detection by T-cells, so it can spread untreated. The human body has T-cells that could remove the cancer if it weren't obscured from detection. Enter CAR-T signaling. Once the signaling takes place, ordinary T-cells can go to work to remove the cancer without surgery or drugs -- obviously a gigantic step forward in treatment. There are several late stage trials which could soon be products.
Terminators
Science doesn't yet fully understand how genetic conditions produce disease. Humans have about 1 million instruction bases that create proteins and makes up about 1% of the human genome. The rest regulate when, how and how much of the protein is made. APARENT is a software package that modifies AI to work on the gene code, much like AlphaFold is a package for DeepMind to work on protein prediction.
We apply APARENT to forward engineer functional polyadenylation signals with precisely defined cleavage position and isoform usage and validate predictions experimentally. Finally, we use APARENT to quantify the impact of genetic variants on APA. Our approach detects pathogenic variants in a wide range of disease contexts, expanding our understanding of the genetic origins of disease.
APA (Alternative polyadenylation) is a regulatory process by which types of processes can be derived by a single gene. The millions of degenerate gene sequences would take eons to unravel by hand, but AI can put huge swaths of APA in libraries for scientists to start to work on. It's still a laborious process, but one that could take decades rather than centuries. This is the hardest research task on the list, but also the one that will unravel the most about how DNA works.
Transcription factors
Now that AlphaFold has predicted the structure of all known human proteins, researchers are busily studying what all the proteins do and why. They are studying how protein break up, recombine and are controlled. Proteins are made up of amino acids, and can break up and recombine when exposed to other proteins.
Human gene expression is regulated by over two thousand transcription factors and chromatin regulators. Effector domains within these proteins can activate or repress transcription. However, for many of these regulators we do not know what type of transcriptional effector domains they contain, their location in the protein, their activation and repression strengths, and the amino acids that are necessary for their functions.
While there aren't too many transcription factors (2,000~), the combination of creating or repressing (and figuring the amount that should be created) makes transcription factors a huge target for AI.
RNA switches
Your RNA does the job of creating and regulating the biochemical reactions in your body. How and when do these processes get "switched" off and on? And can we create tools that modify RNA? Well, Moderna and Pfizer proved we can with messenger RNA (mRNA). But that is merely one example of potentially millions of interventions that could regulate the body's reaction to events.
Engineered RNA elements are programmable tools capable of detecting small molecules, proteins, and nucleic acids. Predicting the behavior of these synthetic biology components remains a challenge, a situation that could be addressed through enhanced pattern recognition from deep learning.
In addition to mRNA, researchers are now exploring transfer RNA (tRNA) and ribosomal RNA (rRNA) among others. It's a huge AI field.
Transcription Factor Promoters
Previous research has found families of things that produce salutatory results, but each combination of thing and result had to be worked out laboriously by hand. LLM to the rescue. In this example the thing promoted is synthetic yeast, but nearly anything could be promoted
How transcription factors (TFs) interpret cis-regulatory DNA sequence to control gene expression remains unclear, largely because past studies using native and engineered sequences had insufficient scale. Here, we measure the expression output of >100 million synthetic yeast promoter sequences that are fully random. These sequences yield diverse, reproducible expression levels that can be explained by their chance inclusion of functional TF binding sites.
Figuring out which of 100 million yeast promoters sequences will yield beneficial results would take centuries by conventional means. Being able to get results in minutes will do the work of centuries of laborious work. This work in yeast promoter sequences may or may not generate useful results, but yeasts are only one of hundreds of fields where this kind of work is being done.
Conclusion
We keep going back to medicine and AI, usually as separate topics. But listing AI as a topic by itself is a mistake. AI will revolutionize medicine just as it has started to do with internet searches, academic research, writing (CNET was writing many of their stories with AI until someone figured it out), drawing pictures, creating videos and so much more.
AI is a tool that will be used to disrupt much of the economy. Using a table of tasks related to every job listed by the Department of Labor, a group of researchers has concluded:
For example, depending on variations in research methodology (e.g., the entire occupation is automated or just a specific task), anywhere between 9% and 47% of jobs will be displaced by artificial intelligence.
19% of workers agree that AI can help alleviate the drudgery of their jobs, and nine out of ten tech executives agree that AI-powered machines will handle mundane tasks, thereby freeing up workers to enjoy more creative work.
Serious researchers are being quoted as saying:
The first person to live to be 1,000 years old is alive today.
A lot of science has to take place before that can happen. The quote is shocking. But even more shocking is the number of serious scientists who have repeated it. If you aren't investing in medical research, what rock are you under? Just keep individual investments small. New research is just around the corner.
Please forward to interested third-parties. If you wish to have the regular Wednesday Musings in your email box, but aren’t ready for a paid subscription, just click on “Subscribe” and select “none”. If you are one of the few who wish to sponsor my effort, bless you. But you get the same messages as everyone.
I try to answer all questions or comments.