MP-NeRF: A Massively Parallel Method for Accelerating Protein Structure Reconstruction from Internal Coordinates

The conversion of proteins between internal and cartesian coordinates is a limiting step in many pipelines, such as molecular dynamics simulations and machine learning models. This conversion is typically carried out by sequential or parallel applications of the Natural extension of Reference Frame (NeRF) algorithm. This work proposes a massively parallel NeRF implementation which, depending on the polymer length, achieves speedups between 400-1000x over the previous state-of-the-art NeRF implementation. It accomplishes this by dividing the conversion into three main phases: a parallel composition of the monomer backbone, the assembly of backbone subunits, and the parallel elongation of sidechains; and by batching computations into a minimal number of efficient matrix operations. Special emphasis is placed on reusability and ease of use within diverse pipelines. We open source the code (available at https://github.com/EleutherAI/mp_nerf) and provide a corresponding python package.

Correlated mutations distinguish misfolded and properly folded proteins

Knowledge about the three dimensional structure of proteins is crucial in order to learn about their behavior, stability, or role as a target in drug design. Unfortunately, traditional experimental methods used in structure determination such as X-ray crystallography and NMR are costly and time-consuming. Therefore, computational methods that allow for protein structure reconstruction from sequence only are greatly desired. One of these is the recently developed direct coupling analysis (DCA) method [1, 2] which achieves the best results in residue-residue contact prediction from multiple sequence alignments only. Predicted contacts are used as restraints in the reconstruction of the three-dimensional structure of a protein. Unfortunately, the accuracy of DCA methods is on the order of 40% among the 100 strongest predicted contacts, which is insufficient for ab initio protein structure reconstruction. However, the results of DCA can support protein structure reconstruction in a different way. Our results show that DCA can indicate the best protein structure among its structural variants by the prediction of residue-residue contacts [3]. We counted the number of correctly predicted contacts within the strongest 100 DCA predictions for a set of obsolete PDB entries and their successors and for 22 proteins for which the Decoys 'R' Us database [4] provided properly folded and misfolded structures. These numbers were related to structure similarity scores, such as RMSD or TM-score [5]. DCA correctly predicts significantly more contacts for properly folded structures than for misfolded ones. Our method works much better for structures determined with X-ray crystallography than with the NMR spectroscopy [3]. The method will not detect misfolded proteins per se, but when a protein structure experimentalist needs to choose between alternative folds for the same protein, DCA can help. [1] F. Morcos et al., Direct-coupling analysis of residue coevolution captures native contacts across many protein families, 2011, Proc Natl Acad Sci U S A 108(49):E1293-301. [2] C. Feinauer et al., Improving contact prediction along three dimensions, 2014, PLoS Comput Biol., 10(10):e1003847. [3] P.P. Wozniak, G. Vriend, M. Kotulska, Correlated mutations select misfolded from properly folded proteins, 2016, Bioinformatics, (article accepted). [4] R. Samudrala, M. Levitt, Decoys 'R' Us: A database of incorrect protein conformations to improve protein structure prediction, 2000, Protein Science 9: 1399-1401. [5] Y. Zhang, J. Skolnick, TM-align: A protein structure alignment algorithm based on TM-score, 2005, Nucleic Acids Research, 33: 2302-2309.

Correlated mutations distinguish misfolded and properly folded proteins

Knowledge about the three dimensional structure of proteins is crucial in order to learn about their behavior, stability, or role as a target in drug design. Unfortunately, traditional experimental methods used in structure determination such as X-ray crystallography and NMR are costly and time-consuming. Therefore, computational methods that allow for protein structure reconstruction from sequence only are greatly desired. One of these is the recently developed direct coupling analysis (DCA) method [1, 2] which achieves the best results in residue-residue contact prediction from multiple sequence alignments only. Predicted contacts are used as restraints in the reconstruction of the three-dimensional structure of a protein. Unfortunately, the accuracy of DCA methods is on the order of 40% among the 100 strongest predicted contacts, which is insufficient for ab initio protein structure reconstruction. However, the results of DCA can support protein structure reconstruction in a different way. Our results show that DCA can indicate the best protein structure among its structural variants by the prediction of residue-residue contacts [3]. We counted the number of correctly predicted contacts within the strongest 100 DCA predictions for a set of obsolete PDB entries and their successors and for 22 proteins for which the Decoys 'R' Us database [4] provided properly folded and misfolded structures. These numbers were related to structure similarity scores, such as RMSD or TM-score [5]. DCA correctly predicts significantly more contacts for properly folded structures than for misfolded ones. Our method works much better for structures determined with X-ray crystallography than with the NMR spectroscopy [3]. The method will not detect misfolded proteins per se, but when a protein structure experimentalist needs to choose between alternative folds for the same protein, DCA can help. [1] F. Morcos et al., Direct-coupling analysis of residue coevolution captures native contacts across many protein families, 2011, Proc Natl Acad Sci U S A 108(49):E1293-301. [2] C. Feinauer et al., Improving contact prediction along three dimensions, 2014, PLoS Comput Biol., 10(10):e1003847. [3] P.P. Wozniak, G. Vriend, M. Kotulska, Correlated mutations select misfolded from properly folded proteins, 2016, Bioinformatics, (article accepted). [4] R. Samudrala, M. Levitt, Decoys 'R' Us: A database of incorrect protein conformations to improve protein structure prediction, 2000, Protein Science 9: 1399-1401. [5] Y. Zhang, J. Skolnick, TM-align: A protein structure alignment algorithm based on TM-score, 2005, Nucleic Acids Research, 33: 2302-2309.