PaleAle 5.0: prediction of protein relative solvent accessibility by deep learning

Abstract

Predicting the three-dimensional structure of proteins is a long-standing challenge of computational biology, as the structure (or lack of a rigid structure) is well known to determine a protein's function. Predicting relative solvent accessibility (RSA) of amino acids within a protein is a significant step towards resolving the protein structure prediction challenge especially in cases in which structural information about a protein is not available by homology transfer. Today, arguably the core of the most powerful prediction methods for predicting RSA and other structural features of proteins is some form of Deep Learning, and all the state-of-the-art protein structure prediction tools rely on some Machine Learning algorithm. In this article we present a deep neural network architecture composed of stacks of Bidirectional Recurrent Neural Networks and Convolutional layers which is capable of mining information from long-range interactions within a protein sequence, and apply it to the prediction of protein RSA using a novel encoding method that we shall call "clipped". The final system we present, PaleAle 5.0, which is available as a public server, predicts RSA into two, three and four classes at an accuracy exceeding 80% in two classes, surpassing the performances of all the other predictors we have benchmarked.

Solvent Accessibility of Residues Undergoing Pathogenic Variations in Humans: From Protein Structures to Protein Sequences

“…Finally, all residues with RSA ≥ 20% were labeled as exposed (E) or buried (B) otherwise. This threshold (or similar ones, in the range of 15-25% RSA) is routinely adopted for computing the protein surfaces and deriving classification datasets in many studies (Thompson and Goldstein, 1996;Mucchielli-Giorgi et al, 1999;Pollastri et al, 2002;Kaleel et al, 2019), since it roughly divides the set of residues in a protein in two equally-sized subsets. In HVAR3D, using a 20% RSA threshold, we obtain 55% and 45% of residues classified as buried and exposed, respectively, corresponding to a realistic characterization of the protein interior (accounting for completely and partially buried residues) and surface (Miller et al, 1987).…”

“…With the advent of machine and deep learning-based approaches (Baldi, 2018), many methods became available for predicting RSA and ASA. They differ mainly in the machine learning approach, the volume of the database of protein structures and the predicted output (ASA, RSA, or binary classification) (Rost and Sander, 1994;Pollastri et al, 2002;Drozdetskiy et al, 2015;Ma and Wang, 2015;Fan et al, 2016;Wu et al, 2017;Kaleel et al, 2019;Klausen et al, 2019).…”

<strong>O-&beta;-GlcNAcylation,</strong><strong> Chloroquine and </strong><strong>2-Hydroxybenzohydrazine May Hamper SARS-CoV-2 entry to Human via Inhibition of ACE2 Phosphorylation at Ser787 but Also Induce Disruption of Virus-ACE2 Binding</strong>

“…NetSurfP-2.0 (http://www.cbs.dtu.dk/services/NetSurfP/) (Klausen et al, 2019) was used to predict the surface accessibility of every Ser and Thr residue of human ACE2 3D structure. PaleAle 5.0 (http://distilldeep.ucd.ie/paleale/) (Kaleel et al, 2019) was used to predict the available area of Ser and Thr residues for ligand interaction. Yin-Yang sites were also predicted on 3D structure of ACE2 protein through docking experiments with the addition of the phosphate group, and UDP-GlcNAc individually using SwissDock server (http://www.swissdock.ch/) (Grosdidier et al, 2011).…”

Using physical features of protein core packing to distinguish real proteins from decoys

“…Thus, we will also implement recurrent neural networks to predict the rSASA values for each residue. 67,68 This model can then be concatenated with the model developed here. Additionally, machine learning methods can be used to predict the particular f c for each amino acid sequence to estimate the difference, Δf c , between the prediction and the actual value.…”

“…For example, we have shown that the identification of core residues is one of the most important aspects for determining a predicted structure's accuracy. Thus, we will also implement recurrent neural networks to predict the rSASA values for each residue 67,68 . This model can then be concatenated with the model developed here.…”