MASS: predict the global qualities of individual protein models using random forests and novel statistical potentials

Abstract

Background: Protein model quality assessment (QA) is an essential procedure in protein structure prediction. QA methods can predict the qualities of protein models and identify good models from decoys. Clustering-based methods need a certain number of models as input. However, if a pool of models are not available, methods that only need a single model as input are indispensable. Results: We developed MASS, a QA method to predict the global qualities of individual protein models using random forests and various novel energy functions. We designed six novel energy functions or statistical potentials that can capture the structural characteristics of a protein model, which can also be used in other protein-related bioinformatics research. MASS potentials demonstrated higher importance than the energy functions of RWplus, GOAP, DFIRE and Rosetta when the scores they generated are used as machine learning features. MASS outperforms almost all of the four CASP11 top-performing single-model methods for global quality assessment in terms of all of the four evaluation criteria officially used by CASP, which measure the abilities to assign relative and absolute scores, identify the best model from decoys, and distinguish between good and bad models. MASS has also achieved comparable performances with the leading QA methods in CASP12 and CASP13.

In Silico Structural Modeling and Binding Site Analysis of Cerebroside Sulfotransferase (CST): A Therapeutic Target for Developing Substrate Reduction Therapy for Metachromatic Leukodystrophy

“…R g represents the mass-weighted root-mean-square distance of atoms from their respective centers of mass. R g is an indicator of 3D model compactness. , The calculated average R g for the CST protein was found to be nominal with the value of 1.8 nm throughout the simulation time, strongly supporting the stability and compactness of the model structure (Figure C).…”

Peptonics: A new family of cell‐protecting surfactants for the recombinant expression of therapeutic proteins in mammalian cell cultures

“…[ 45 ] Similar to smart polymers, the size of self‐assembled structures formed by both peptide‐ and peptoid‐based Peptonics does not vary significantly with concentration, but increases 10‐fold upon heating their solutions in PBS from room temperature to 37°C, rising from 5–11 nm to 100–200 nm; at 0.1 mg mL −1 , below the CMC of the tested Peptonics, no aggregates were detected above 10 nm, whereas above the CMC the size of aggregates were rather constant in the concentration range 0.5–2 mg mL −1 . The comparison of the hydrodynamic radius of Peptonic particles, which is estimated to range between 0.8 and 1.8 nm, [ 46 ] with the values of size and polydispersity (PDI) listed in Table S5 suggests that the self‐assembled structures formed at room temperature are not ordered micelles, but rather aggregates of a few Peptonics; conversely, the structures formed at room temperature are likely micelle‐like structures formed by a large number of Peptonic chains.…”

Predicting residue‐specific qualities of individual protein models using residual neural networks and graph neural networks

“…We used the models from CASP7 to CASP12 for training and validation, where models are trimmed into evaluation units (EUs) 21 . For each residue of a protein model, we generated features of six categories: one‐hot encoding of amino acid sequence (1 × 21), position‐specific scoring matrix (PSSM) created by PSI‐BLAST with the MSA (1 × 20), normalized Rosetta energies (1 × 20), SOV_refine 22 local scores (1 × 6) and global scores (1 × 6) between predicted (from sequences) and parsed (from models) SS and solvent accessibilities, MASS protein statistical potentials 14 (1 × 6), and (6) sinusoidal positional encoding (1 × 4).…”

“…Our previous MASS system developed the MASS potentials 14 as machine learning features. MASS potentials introduce energy functions and statistical potentials, capturing features from models.…”

“…The authors of QAcon 13 found that the performance of its two‐layer neural network was improved by adding the novel feature of residue‐residue contact information, which illustrates the value of using the contact information in protein QA. Our previously developed MASS 14 is a single‐model method that only predicts global quality scores using random forests based on six protein statistical potentials that indicate structural or energetic properties of protein models. In comparison, one of the methods presented in this article, MASS2, was trained to predict residue‐specific qualities and was based on completely different methodologies.…”