Proteins are the workhorses that keep our cells running, and there are various 1000’s of kinds of proteins in our cells, each performing a specialized function. Researchers have long known that the structure of a protein determines what it may do. More recently, researchers are coming to understand that a protein’s localization can also be critical for its function. Cells are stuffed with compartments that help to arrange their many denizens. Together with the well-known organelles that adorn the pages of biology textbooks, these spaces also include quite a lot of dynamic, membrane-less compartments that concentrate certain molecules together to perform shared functions. Knowing where a given protein localizes, and who it co-localizes with, can subsequently be useful for higher understanding that protein and its role within the healthy or diseased cell, but researchers have lacked a scientific option to predict this information.
Meanwhile, protein structure has been studied for over half-a-century, culminating in the synthetic intelligence tool AlphaFold, which may predict protein structure from a protein’s amino acid code, the linear string of constructing blocks inside it that folds to create its structure. AlphaFold and models prefer it have turn out to be widely used tools in research.
Proteins also contain regions of amino acids that don’t fold into a set structure, but are as a substitute essential for helping proteins join dynamic compartments within the cell. MIT Professor Richard Young and colleagues wondered whether the code in those regions may very well be used to predict protein localization in the identical way that other regions are used to predict structure. Other researchers have discovered some protein sequences that code for protein localization, and a few have begun developing predictive models for protein localization. Nonetheless, researchers didn’t know whether a protein’s localization to any dynamic compartment may very well be predicted based on its sequence, nor did they’ve a comparable tool to AlphaFold for predicting localization.
Now, Young, also member of the Whitehead Institute for Biological Research; Young lab postdoc Henry Kilgore; Regina Barzilay, the School of Engineering Distinguished Professor for AI and Health in MIT’s Department of Electrical Engineering and Computer Science and principal investigator within the Computer Science and Artificial Intelligence Laboratory (CSAIL); and colleagues have built such a model, which they call ProtGPS. In a paper published on Feb. 6 within the journal , with first authors Kilgore and Barzilay lab graduate students Itamar Chinn, Peter Mikhael, and Ilan Mitnikov, the cross-disciplinary team debuts their model. The researchers show that ProtGPS can predict to which of 12 known kinds of compartments a protein will localize, in addition to whether a disease-associated mutation will change that localization. Moreover, the research team developed a generative algorithm that may design novel proteins to localize to specific compartments.
“My hope is that it is a first step towards a strong platform that allows people studying proteins to do their research,” Young says, “and that it helps us understand how humans turn into the complex organisms that they’re, how mutations disrupt those natural processes, and how you can generate therapeutic hypotheses and design drugs to treat dysfunction in a cell.”
The researchers also validated most of the model’s predictions with experimental tests in cells.
“It really excited me to find a way to go from computational design all of the option to trying these items within the lab,” Barzilay says. “There are plenty of exciting papers on this area of AI, but 99.9 percent of those never get tested in real systems. Due to our collaboration with the Young lab, we were in a position to test, and really learn the way well our algorithm is doing.”
Developing the model
The researchers trained and tested ProtGPS on two batches of proteins with known localizations. They found that it could accurately predict where proteins find yourself with high accuracy. The researchers also tested how well ProtGPS could predict changes in protein localization based on disease-associated mutations inside a protein. Many mutations — changes to the sequence for a gene and its corresponding protein — have been found to contribute to or cause disease based on association studies, however the ways through which the mutations result in disease symptoms remain unknown.
Determining the mechanism for the way a mutation contributes to disease is significant because then researchers can develop therapies to repair that mechanism, stopping or treating the disease. Young and colleagues suspected that many disease-associated mutations might contribute to disease by changing protein localization. For instance, a mutation could make a protein unable to affix a compartment containing essential partners.
They tested this hypothesis by feeding ProtGOS greater than 200,000 proteins with disease-associated mutations, after which asking it to each predict where those mutated proteins would localize and measure how much its prediction modified for a given protein from the conventional to the mutated version. A big shift within the prediction indicates a probable change in localization.
The researchers found many cases through which a disease-associated mutation appeared to vary a protein’s localization. They tested 20 examples in cells, using fluorescence to match where within the cell a traditional protein and the mutated version of it ended up. The experiments confirmed ProtGPS’s predictions. Altogether, the findings support the researchers’ suspicion that mis-localization could also be an underappreciated mechanism of disease, and show the worth of ProtGPS as a tool for understanding disease and identifying latest therapeutic avenues.
“The cell is such an advanced system, with so many components and sophisticated networks of interactions,” Mitnikov says. “It’s super interesting to think that with this approach, we will perturb the system, see the end result of that, and so drive discovery of mechanisms within the cell, and even develop therapeutics based on that.”
The researchers hope that others begin using ProtGPS in the identical way that they use predictive structural models like AlphaFold, advancing various projects on protein function, dysfunction, and disease.
Moving beyond prediction to novel generation
The researchers were excited in regards to the possible uses of their prediction model, but in addition they wanted their model to transcend predicting localizations of existing proteins, and permit them to design completely latest proteins. The goal was for the model to make up entirely latest amino acid sequences that, when formed in a cell, would localize to a desired location. Generating a novel protein that may actually accomplish a function — on this case, the function of localizing to a particular cellular compartment — is incredibly difficult. As a way to improve their model’s probabilities of success, the researchers constrained their algorithm to only design proteins like those present in nature. That is an approach commonly utilized in drug design, for logical reasons; nature has had billions of years to work out which protein sequences work well and which don’t.
Due to the collaboration with the Young lab, the machine learning team was in a position to test whether their protein generator worked. The model had good results. In a single round, it generated 10 proteins intended to localize to the nucleolus. When the researchers tested these proteins within the cell, they found that 4 of them strongly localized to the nucleolus, and others could have had slight biases toward that location as well.
“The collaboration between our labs has been so generative for all of us,” Mikhael says. “We’ve learned how you can speak one another’s languages, in our case learned loads about how cells work, and by having the prospect to experimentally test our model, we’ve been in a position to work out what we want to do to truly make the model work, after which make it work higher.”
Having the ability to generate functional proteins in this manner could improve researchers’ ability to develop therapies. For instance, if a drug must interact with a goal that localizes inside a certain compartment, then researchers could use this model to design a drug to also localize there. This could make the drug simpler and reduce unwanted side effects, because the drug will spend more time engaging with its goal and fewer time interacting with other molecules, causing off-target effects.
The machine learning team members are enthused in regards to the prospect of using what they’ve learned from this collaboration to design novel proteins with other functions beyond localization, which might expand the chances for therapeutic design and other applications.
“A whole lot of papers show they will design a protein that could be expressed in a cell, but not that the protein has a specific function,” Chinn says. “We actually had functional protein design, and a comparatively huge success rate in comparison with other generative models. That’s really exciting to us, and something we would really like to construct on.”
The entire researchers involved see ProtGPS as an exciting starting. They anticipate that their tool will probably be used to learn more in regards to the roles of localization in protein function and mis-localization in disease. As well as, they’re inquisitive about expanding the model’s localization predictions to incorporate more kinds of compartments, testing more therapeutic hypotheses, and designing increasingly functional proteins for therapies or other applications.
“Now that we all know that this protein code for localization exists, and that machine learning models could make sense of that code and even create functional proteins using its logic, that opens up the door for therefore many potential studies and applications,” Kilgore says.
