When faced with a stubborn lock, jiggling the key often proves successful. Similarly, groundbreaking research utilizing artificial intelligence from MIT demonstrates that coronaviruses, including the culprit behind Covid-19, may employ an analogous strategy to deceive cells and gain entry. These AI-enhanced discoveries could revolutionize how we evaluate the danger level of various coronavirus strains and mutations, potentially unveiling innovative pathways for therapeutic development.
Traditional investigations into spike proteins—responsible for the distinctive crown-like appearance of coronaviruses—have primarily focused on biochemical interactions. However, this study, powered by advanced machine learning techniques, adopted a novel approach. By employing sophisticated atomistic simulations, researchers analyzed the mechanical dynamics of spike protein movements, shape transformations, and vibrations. The results, processed through computational modeling algorithms, suggest these vibrational patterns may represent a strategic mechanism that coronaviruses exploit to manipulate cellular locking systems, thereby facilitating viral penetration through cell walls to hijack reproductive machinery.
The research team, leveraging artificial intelligence analysis, discovered a robust correlation between the frequency and intensity of spike vibrations and the virus's ability to penetrate cells. Conversely, they identified an inverse relationship with coronavirus fatality rates. Given that this AI-powered methodology relies on comprehending the intricate molecular architecture of these proteins, the scientists propose it could serve as a rapid-screening tool for emerging coronaviruses or novel Covid-19 mutations, enabling swift assessment of their potential threat levels.
These groundbreaking findings, authored by MIT professor of civil and environmental engineering Markus Buehler and graduate student Yiwen Hu, are being published today in the journal Matter.
According to Buehler, conventional representations of the SARS-CoV-2 virus are somewhat misleading. "The virus doesn't look like that," he explains, because at the nanometer scale of atoms, molecules, and viruses, "everything is in constant motion and vibration. They don't resemble the static images in chemistry textbooks or on websites."
Buehler's laboratory specializes in atom-by-atom computational modeling of biological molecules and their behaviors. As soon as Covid-19 emerged and data about the virus's protein structure became available, Buehler and Hu, a mechanical engineering doctoral student, immediately applied their artificial intelligence-enhanced techniques to investigate whether the mechanical properties of these proteins influenced their interactions with the human body.
The minuscule nanoscale vibrations and shape alterations of these protein molecules present significant observational challenges through conventional experimental methods, making AI-powered atomistic simulations invaluable for understanding these processes. The researchers applied this machine learning approach to examine a critical infection phase—when a viral particle, equipped with its protein spikes, attaches to a human cell receptor known as the ACE2 receptor. Once these spikes establish contact with the receptor, they unlock a channel that permits the virus to infiltrate the cell.
This binding mechanism between proteins and receptors functions similarly to a lock and key system, making vibrations crucial, as Buehler explains. "If it's static, it either fits or it doesn't," he states. However, protein spikes are not static; "they're vibrating and continuously changing their shape slightly, and that's significant. Keys are static and don't change shape, but imagine a key that's constantly transforming—vibrating, moving, morphing slightly? It would fit differently depending on its configuration at the precise moment of insertion."
The researchers theorize that the more the "key" can transform, the higher the probability of achieving a successful fit.
Buehler and Hu employed artificial intelligence algorithms to model the vibrational characteristics of these protein molecules and their interactions, utilizing analytical techniques such as "normal mode analysis." This method, enhanced by machine learning, examines vibration development and propagation by representing atoms as point masses interconnected by springs that symbolize the various forces acting between them.
Through their AI-powered analysis, they discovered that variations in vibrational characteristics strongly correlate with differences in infectivity rates and lethality among various coronaviruses, using data from a global database of confirmed cases and case fatality rates. The viruses studied included SARS-CoV, MERS-CoV, SARS-CoV-2, and one increasingly prevalent mutation of the SARS-CoV-2 virus. Buehler suggests this artificial intelligence-enhanced methodology represents a promising tool for predicting potential risks from new coronaviruses as they inevitably emerge.
In all cases examined through their artificial intelligence framework, Hu notes that a crucial aspect of the process involves fluctuations in an upward swing of one protein molecule branch, which facilitates binding to the receptor. "That movement holds significant functional importance," she explains. Another critical indicator relates to the ratio between two distinct vibrational motions within the molecule. "We've found that these two factors demonstrate a direct relationship with epidemiological data—virus infectivity and lethality," she states.
The correlations uncovered through their machine learning analysis mean that when new viruses or mutations of existing ones appear, "you could screen them from a purely mechanical perspective," Hu says. "You can examine the fluctuations of these spike proteins and predict their epidemiological impact—how infectious and severe the disease might be."
These AI-enhanced findings might also open new research avenues for potential Covid-19 and other coronavirus treatments, Buehler suggests. He speculates about the possibility of discovering a molecule that could bind to spike proteins in a manner that would rigidify them and restrict their vibrations. Another approach might involve inducing opposing vibrations to neutralize the natural ones in the spikes, similar to how noise-canceling headphones eliminate unwanted sounds.
As biologists gain deeper insights into various coronavirus mutations and identify which genomic regions are most susceptible to change, this artificial intelligence methodology could also serve predictive purposes, Buehler explains. The most probable mutation types could all be simulated through computational modeling, with those posing the greatest danger flagged to alert the world to watch for signs of their actual emergence. "The G614 mutation, for instance, currently dominating global Covid-19 spread, is predicted by our findings to be slightly more infectious yet somewhat less lethal," Buehler adds.
Mihri Ozkan, a professor of electrical and computer engineering at the University of California at Riverside, unconnected to this research, comments that this analysis "highlights the direct correlation between nanomechanical features and coronavirus lethality and infection rates. I believe this work significantly advances the field in uncovering insights into disease and infection mechanics."
Ozkan further notes that "If, under natural environmental conditions, the overall flexibility and mobility ratios predicted in this work occur, identifying an effective inhibitor that could lock the spike protein to prevent binding might represent the holy grail of preventing SARS-CoV-2 infections, which we desperately need now."
The research received support from the MIT-IBM Watson AI Lab, the Office of Naval Research, and the National Institutes of Health.
