The AI ​​system predicts fragments of proteins that can join or inhibit an objective | MIT News

All biological function depends on how different proteins interact with each other. Protein-protein interactions facilitate everything, from DNA transcription and cell division control to higher level functions in complex organisms.

However, there is much without being clear about how these functions are orocated at the molecular level and how proteins interact with each other, either with other proteins or copies of themselves.

Recent findings have revealed that small protein fragments have a lot of functional potential. Although they are incomplete pieces, the short stretches of amino acids can still join the interfaces of an target protein, recapitulating native interactions. Through this process, they can alter the function of that protein or interrupt their interactions with other proteins.

Therefore, protein fragments could empower both basic research on protein interactions and cellular processes, and could potentially have therapeutic applications.

Recently Posted in Proceedings of the National Academy of SciencesA new method developed in the Biology Department is based on existing artificial intelligence models to computationally predict protein fragments that can join and inhibit full length proteins E. coli. Theoretically, this tool could lead to genetically codifiable inhibitors against any protein.

The work was done in the associate professor of Biology and researcher at the Howard Hughes Medical Institute Gene-Wei Li In collaboration with the laboratory of Jay A. Stein (1968) Professor of Biology, Professor of Biological Engineering and Department Head Amy Keating.

Taking advantage of automatic learning

The program, called Fragfold, takes advantage of Alfafold, an ai model that has led to phenomenal advances in biology in recent years due to its ability to predict the folding of proteins and protein interactions.

The objective of the project was to predict fragments inhibitors, which is a novel Alfafold application. The researchers in this project experimentally confirmed that more than half of the fragflet predictions for the Union or the inhibition were precise, even when the researchers did not have previous structural data on the mechanisms of these interactions.

“Our results suggest that this is a generalizable approach to find union modes that probably inhibit protein function, even for new protein objectives, and you can use these predictions as a starting point for more experiments,” says the author and the author corresponding. Andrew Savinov, a postdoc in the Li Laboratory. “We can really apply this to proteins without known functions, without known interactions, without even known structures, and we can put some credit in these models that we are developing.”

An example is FTSZ, a protein that is key to cell division. It is well studied, but contains an intrinsically disorderly region and, therefore, especially challenging to study. Disordered proteins are dynamic, and their functional interactions are very fleeting, occurring so briefly that current structural biology tools cannot capture a single structure or interaction.

The researchers took advantage of Frafold to explore the activity of the FTSZ fragments, including the fragments of the intrinsically disorderly region, to identify several new union interactions with several proteins. This jump in understanding confirms and expands in previous experiments that measure FTSZ's biological activity.

This progress is significant in part because it was carried out without solving the structure of the disorderly region, and because it exhibits the potential power of fragic.

“This is an example of how Alfafold is fundamentally changing how we can study molecular and cellular biology,” says Keating. “The creative applications of ai methods, such as our work in Fragfold, open unexpected capabilities and new research addresses.”

Inhibition and beyond

The researchers achieved these predictions fragmenting each protein and then modeling how these fragments would join the interaction partners who thought they were relevant.

They compared the predicted union maps throughout the sequence with the effects of those same fragments in living cells, determined using experimental high performance measurements in which millions of cells produce a type of protein fragment.

Alphafold uses coevolutive information to predict folding, and typically evaluates the evolutionary history of protein using something called multiple sequences alignments for each prediction execution. MSA are critical, but they are a bottleneck for large -scale predictions: they can take an amount of time and prohibational computational power.

For Fragfold, the researchers instead pre -heed the MSA for a full length protein once, and used that result to guide the predictions for each fragment of that full length protein.

Savinov, together with Keating Lab Student Sebastian Swanson PhD '23, predicted inhibitory fragments of a diverse set of protein in addition to FTSZ. Among the interactions they explored were a complex between the Lipopolycard and LPT LPT transport proteins. A fragment of LPG protein inhibited this interaction, presumably interrupting the supply of lipopolysaccharide, which is a crucial component of the E. coli External cell membrane essential for cellular fitness.

“The big surprise was that we can predict union with such high precision and, in fact, often predict the union that corresponds to inhibition,” says Savinov. “For each protein we have seen, we have been able to find inhibitors.”

The researchers initially focused on protein fragments such as inhibitors because if a fragment could block an essential function in cells is a relatively simple result to systematically measure. Looking to the future, Savinov is also interested in exploring the function of fragments outside the inhibition, such as fragments that can stabilize the protein to which they join, improve or alter their function, or trigger the degradation of the protein.

Design, in principle

This research is a starting point to develop a systemic understanding of cell design principles, and in what elements can be resorting to deep learning models to make precise predictions.

“There is a broader and greater scope objective than we are building,” says Savinov. “Now that we can predict them, can we use the data we have of predictions and experiments to extract the outstanding characteristics to discover what Alfafold has learned about what makes a good inhibitor?”

Savinov and the collaborators also deepened how protein fragments bind, exploring other protein interactions and mutating specific waste to see how these interactions change how the fragment interacts with its objective.

When experimentally examining the behavior of thousands of mutated fragments within the cells, an approach known as deep mutational scan, revealed key amino acids responsible for inhibition. In some cases, the mutated fragments were inhibitors even more powerful than their natural sequences of full length.

“Unlike the above methods, we do not limit ourselves to the identification of fragments in experimental structural data,” says Swanson. “The central force of this work is the interaction between high -performance experimental inhibition data and predicted structural models: experimental data guide us towards fragments that are particularly interesting, while structural models predicted by fragfold provide a specific hypothesis and probable for how fragments work at the molecular level. “

Savinov is excited about the future of this approach and its innumerable applications.

“When creating compact and genetically codifiable binders, Fragfold opens a wide range of possibilities to manipulate the protein function,” LI agrees. “We can imagine the delivery of functionalized fragments that can modify native proteins, change their subcellular location and even reprogram them to create new tools to study cell biology and treat diseases.”