Authors:
Joseph Azzopardi
1
and
Jean Paul Paul Ebejer
1
;
2
Affiliations:
1
Department of Artificial Intelligence, University of Malta, Msida, MSD 2080, Malta
;
2
Centre for Molecular Medicine and Biobanking, University of Malta, Msida, MSD 2080, Malta
Keyword(s):
Virtual Screening, Structure-based Virtual Screening, Scoring Function, Pharmacophoric Interaction Point, Machine Learning, Deep-Learning, Convolutional Neural Networks, LigityScore.
Abstract:
Scoring functions are at the heart of structure-based drug design and are used to estimate the binding of ligands to a target. Seeking a scoring function that can accurately predict the binding affinity is key for successful virtual screening methods. Deep learning approaches have recently seen a rise in popularity as a means to improve the scoring function having as a key advantage the automatic extraction of features and the creation of a complex representation without feature engineering and expert knowledge. In this study we present LigityScore1D and LigityScore3D, which are rotationally invariant scoring functions based on convolutional neural networks. LigityScore descriptors are extracted directly from the structural and interacting properties of the protein-ligand complex which are input to a CNN for automatic feature extraction and binding affinity prediction. This representation uses the spatial distribution of Pharmacophoric Interaction Points, derived from interaction fea
tures from the protein-ligand complex based on pharmacophoric features conformant to specific family types and distance thresholds. The data representation component and the CNN architecture together, constitute the LigityScore scoring function. The main contribution for this study is to present a novel protein-ligand representation for use as a CNN based SF for binding affinity prediction. LigityScore models are evaluated for scoring power on the latest two CASF benchmarks. The Pearson Correlation Coefficient, and the standard deviation in linear regression were used to compare and rank LigityScore with the benchmark model, and also to other models recently published in literature. LigityScore3D has achieved better overall results and showed similar performance in both CASF benchmarks. LigityScore3D ranked 5th place for the CASF-2013 benchmark , and 8th for CASF-2016, with an average R-score performance of 0.713 and 0.725 respectively. LigityScore1D ranked 8th place for the CASF-2013 and 7th place for CASF-2016 with an R-score performance of 0.635 and 0.741 respectively. Our methods show relatively good performance when compared to the Pafnucy model (one of the best performing CNN based scoring functions), on the CASF-2013 benchmark using a less computationally complex model that can be trained 16 times faster.
(More)