Allosteric regulation of Paramyxovirus Entry into Host Cells: Machine learning to the rescue

Sameer Varma
University of South Florida

Nipah viruses are highly virulent and cause recurring encephalitis in humans with 77% mortality. The entry of these viruses into host cells is triggered when specific proteins on the viral membrane, called attachment proteins, bind to their appropriate receptors on the host cell membrane. The attachment proteins have separate domains for receptor binding and mediating virus-host membrane fusion. However, the molecular details of how the receptor-binding signal transduces from the receptor-binding domain to the fusion-mediating domain remains unknown. Understanding this process has been challenging mainly because receptor binding induces only minor structural changes in the receptor binding domain (mean deviation < 0.2 nm). This implies that signal transduction occurs primarily via changes in side-chain rotations and fluctuations. Consequently, to understand signaling in such scenarios, one needs to look beyond examining differences between two protein structures. An understanding of signal transduction in such systems requires a quantitative assessment of differences in structural ensembles.
Here we present the development of a new method based on support vector machines to quantitatively evaluate differences in conformational ensembles. The primary challenge that this method overcomes is associated with comparing two high dimensional vector spaces. In addition, we will present how we have used this method in conjunction with accelerated conformational sampling techniques to illuminate the molecular details underlying the allosteric regulation of Nipah entry into host cells. These studies highlight, in general, how signals can be transferred across nanometer long distances in proteins without major backbone rearrangements. We anticipate that our method and approach will be applicable to other systems where allosteric signaling is achieved via small changes in protein structure.

Back to Machine Learning for Many-Particle Systems