NetSci: A Library for High Performance Biomolecular Simulation Network Analysis Computation

We present the Netsci program - an open-source scientific software package that leverages GPU acceleration and a k-nearest-neighbor algorithm in order to estimate the mutual information (MI) between data in a set. The GPU acceleration presented here, as an improvement upon existing estimators, enables calculation speeds several orders of magnitude faster than CPU-based implementations, all with dataset size limits determined only by the available hardware. To demonstrate the validity and usefulness of Netsci, we show that the MI is correctly computed for the analytically-verifiable two-dimensional Gaussian distribution, and we also reproduce the generalized correlation (GC) analysis performed in an earlier study on the B1 domain of protein G. In addition, we apply Netsci to the analysis of molecular dynamics simulations of the Sarcoendoplasmic Reticulum Calcium-ATPase (SERCA) pump. Specifically, we use Netsci to understand the allosteric mechanisms and pathways of SERCA, and compare the differential effects of the binding of two nucleotides, ATP and 2'-deoxy-ATP (dATP). We determine that ATP binding to SERCA, compared to dATP, induces differential allosteric effects. The most likely information pathways from the bound nucleotide to the calcium binding domain are also predicted using our MI estimator in combination with network analysis tools on the SERCA pump, which differs based on the bound nucleotide. Netsci is shown to be a useful program for the estimation of MI and GC within general datasets, and for the analysis of intraprotein communication and information transfer, in particular.