Bagaimanakah ML memajukan pembangunan biologi struktur? Para saintis Harvard menggunakan AI untuk mengkaji pembangunan manusia pada skala terkecil

Editor | Cabbage Leaf
For structural biologist Lucas Farnung, there is no question more fascinating than how a single fertilized egg develops into a fully functional human. He is working to study this process at the smallest scales: trillions of atoms must work in sync to achieve this process.
“I don’t see a big difference between solving a 5,000-piece puzzle and what we do in the lab,” said Farnung, assistant professor of cell biology at the Blavatnik Institute at Harvard Medical School. We try to figure out visually what this process looks like, and then we can form an idea about how it works. Nearly all cells in the human body contain the same genetic material, but what happens to these cells during development. What tissue type (for example, becomes liver or skin) is largely determined by gene expression, which determines which genes are turned on and off.
Gene expression is regulated by the transcription process, and transcription is the focus of Farnung's research. During transcription, molecular machines read the instructions contained in the genetic blueprint stored within DNA and generate the molecular RNA that carries out the instructions. Other molecular machines read the RNA and use the information to make proteins that power nearly every activity in the body.
Farnung studies the structure and function of the molecular machines responsible for transcription.
In an interview with the media, Farnung discussed his work and how machine learning is accelerating research in this field.
Q: What is the core question your research is trying to answer?
Farnung: I always say that we are interested in the smallest logical problems. The human genome is present in nearly every cell, and if you stretched the DNA that makes up the genome, it would be about two meters long, or six and a half feet. But the two-meter-long molecule must be crammed into the cell nucleus, which is only a few microns in size.
That’s the equivalent of stuffing a fishing line that stretches from Boston to New Haven, Connecticut (about 150 miles) into a football.
To achieve this, our cells compact the DNA into a structure called chromatin, but the genomic information on the DNA is no longer accessible to molecular machines.
This creates a conflict because the DNA needs to be compact enough to fit within the nucleus, but the molecular machines must be able to access the genomic information on the DNA.
We are particularly interested in observing how a molecular machine called RNA polymerase II acquires genomic information and transcribes DNA into RNA.
Q: What techniques does your team use to visualize molecular machines?
Farnung: Our general approach is to isolate the molecular machinery from the cell and then observe it using a specific type of microscope or X-ray beam.
To do this, we introduce genetic material encoding a human molecular machine of interest into an insect or bacterial cell so that the cell makes the machine in large quantities.
We then use purification techniques to separate the machinery from the cell so that we can study it individually.
However, this is complicated because we are usually not just interested in individual molecular machines, which we also call proteins.
There are thousands of protein interactions that regulate transcription, so we have to repeat this process thousands of times to understand these protein-protein interactions.
Q: Artificial intelligence is beginning to penetrate into all aspects of basic biology. Has it changed the way you approach structural biology research?
Farnung: Over the past thirty or forty years, research in my field has been a tedious process. A PhD student's scientific research career may only focus on studying one or two proteins, but understanding the interactions of proteins in cells requires a workload that thousands of students may not be able to complete.
In the past two or three years, however, we have increasingly looked to computational methods to predict protein interactions. Google DeepMind released AlphaFold, a machine learning model that can predict protein folding, which is a major breakthrough. Importantly, the way proteins fold determines their function and interactions.
We are now using artificial intelligence to predict tens of thousands of protein-protein interactions, many of which have never been experimentally described before. In fact, not all of these interactions occur inside the cell, but we can verify them through laboratory experiments.
This is very exciting because it really accelerates our scientific research. When I look back on my PhD, the first three years were largely a failure—I failed to discover any protein-protein interactions.
Now, armed with these computational predictions, a PhD student or postdoc in my lab can be very confident that the lab's experiments validating protein-protein interactions will be successful. I call it an enhanced version of molecular biology - but legitimate - because we can now find the actual questions we want answered much faster.
Q: In addition to efficiency and speed, in what ways is artificial intelligence reshaping your field?
Farnung: An exciting change is that we can now test any protein in the human body against any other protein in an unbiased way to see if they are likely to interact. Machine learning tools in our field are causing disruption similar to what personal computers did to society.
When I first became a researcher, people were using X-ray crystallography to reveal the structure of individual proteins - it's a wonderful high-resolution technology, but it can take many years. Later, during my PhD and postdoc years, cryo-electron microscopy (cryo-EM for short) came into being. This technique allows us to observe larger, more dynamic protein complexes at high resolution.
Cryo-electron microscopy has enabled great advances in our understanding of biology and accelerated drug development over the past decade.
I consider myself lucky to be part of the so-called resolution revolution brought about by cryo-electron microscopy. But now, it feels like machine learning for protein prediction is bringing about a second revolution, which to me is really amazing and makes me wonder how much more acceleration we will see.
In my estimation, we are probably doing research 5 to 10 times faster now than we were 10 years ago. It will be interesting to see how machine learning changes the way we conduct biological research in the next 10 years.
Of course, we have to manage these tools carefully, but I’m excited to find answers to questions I’ve been thinking about for a long time 10x faster.
Q: In addition to the laboratory, what are the downstream applications of your work?
Farnung: We are understanding how biology works in the human body at a fundamental level, but we always believe that understanding basic biological mechanisms can help us develop effective treatments for various diseases. For example, disruption of DNA chromatin structure by molecular machines has been shown to be one of the main drivers of many cancers. Once we figure out the structure of these molecular machines, we can understand the effects of changing a few atoms to replicate the mutations that cause cancer, at which point we can begin to design drugs that target the proteins.
We have just launched a project in collaboration with HMS Therapeutics Programs that is studying chromatin remodelers, a protein that is severely mutated in prostate cancer. We recently obtained the structure of this protein and are conducting virtual screening to see which compounds can bind to it.
We hope to be able to design a compound that inhibits this protein and potentially develop it into a proven drug that could slow down the progression of prostate cancer. We are also studying proteins associated with neurodevelopmental disorders such as autism. Machine learning can help us here because the tools we use to predict protein structure and protein-protein interactions can also predict how small molecule compounds bind to proteins.
Q: Speaking of collaboration, how important is working across research areas and disciplines to your research?
Farnung: Collaboration is very important to my research. The field of biology has become extremely complex, with so many different areas of study that it is impossible to know everything. By collaborating, we can bring people with different expertise together to study important biological questions, such as how molecular machines access the human genome.
We collaborate with other researchers at Harvard Medical School on many different levels. Sometimes we draw on structural expertise to support work in other laboratories. Sometimes we have solved the structure of a protein, but we need collaboration to understand the role of that protein in the broader cellular environment. We also collaborate with laboratories using other molecular biology methods. Collaboration is critical to driving progress and better understanding of biology.
Related content: https://hms.harvard.edu/news/how-machine-learning-propelling-structural-biology

Atas ialah kandungan terperinci Bagaimanakah ML memajukan pembangunan biologi struktur? Para saintis Harvard menggunakan AI untuk mengkaji pembangunan manusia pada skala terkecil. Untuk maklumat lanjut, sila ikut artikel berkaitan lain di laman web China PHP!