Computational Platform

What makes us unique is the power of our computational platform, which enables us to reconstruct the series of conformational rearrangements underlying the process of protein folding at an all-atom level of detail.

The folding pathway of biologically-relevant proteins is characterized by the presence of intermediate states separated by high energy barriers. Due to the exponential dependency between the transition rate and the height of the barrier, the conformational changes between metastable states only rarely occur. For this reason, protein folding can last milliseconds to minutes. Standard molecular dynamics techniques are unable to reach such timescales for biologically-relevant proteins, even by employing the most powerful existing special-purpose supercomputer.

In contrast, our algorithms are capable of driving the polypeptide chain through the conformational landscape, overcoming the process of trial and error responsible for the timescale-inaccessibility typical of protein folding. Our scheme enables us to accurately characterize the folding pathways of proteins with size up to 400 amino acids in less than a month.

Our sole requirement to fully characterize the folding mechanism of a target protein is the structure of its native state. Such information is used by our approach to generate a collective variable that defines the reaction progress. Our algorithm relies on such information to drive the dynamics by employing a custom implementation of adiabatic-biased molecular dynamics. We then ensure the reliability of the sampled pathways by using filtering schemes derived from theoretical physics.

Density-based clustering algorithms are then employed to deliver a network of states and transitions accurately representing the mechanism from which relevant folding intermediates can be identified. Once the candidate structures are selected, we scout their surfaces to identify druggable pockets absent in the native state.

These sites can be used as the starting point of an in-house virtual screening campaign, eventually delivering promising hit compounds for the target of interest.

Our platform is powered by a large GPU cluster, able to run molecular dynamics simulations with a computing performance of hundreds of TFLOPS.

We continuously improve our algorithms and develop new techniques to push innovation forward. The reconstruction of the protein folding mechanism can be improved by taking into account the complexity of the cellular environment in which proteins acquire their native fold.

One of the key elements of folding in vivo is that it occurs sequentially during protein synthesis. Indeed, vectoriality can significantly influence the folding mechanism of proteins and should be considered for more accurate characterization of folding intermediates. Here at Sibylla, we have developed dedicated algorithms that, for the first time, enable simulating the folding process during synthesis by the ribosome and translocation across the endoplasmic reticulum membrane.