A point cloud-based deep learning strategy for protein–ligand binding affinity prediction

Abstract

There is great interest to develop artificial intelligence-based protein–ligand binding affinity models due to their immense applications in drug discovery. In this paper, PointNet and PointTransformer, two pointwise multi-layer perceptrons have been applied for protein–ligand binding affinity prediction for the first time. Three-dimensional point clouds could be rapidly generated from PDBbind-2016 with 3772 and 11 327 individual point clouds derived from the refined or/and general sets, respectively. These point clouds (the refined or the extended set) were used to train PointNet or PointTransformer, resulting in protein–ligand binding affinity prediction models with Pearson correlation coefficients R = 0.795 or 0.833 from the extended data set, respectively, based on the CASF-2016 benchmark test. The analysis of parameters suggests that the two deep learning models were capable to learn many interactions between proteins and their ligands, and some key atoms for the interactions could be visualized. The protein–ligand interaction features learned by PointTransformer could be further adapted for the XGBoost-based machine learning algorithm, resulting in prediction models with an average Rp of 0.827, which is on par with state-of-the-art machine learning models. These results suggest that the point clouds derived from PDBbind data sets are useful to evaluate the performance of 3D point clouds-centered deep learning algorithms, which could learn atomic features of protein–ligand interactions from natural evolution or medicinal chemistry and thus have wide applications in chemistry and biology.

HAC-Net: A Hybrid Attention-Based Convolutional Neural Network for Highly Accurate Protein–Ligand Binding Affinity Prediction

“…There has been continuous progress in applying ML to predict protein–ligand binding affinity, gaining significant popularity in 2010 with NNScore, an ensemble of 10 multi-layer perceptrons (MLPs), and RF-Score, a random forest-based model. Many groups have subsequently utilized random forest-based approaches − or related methods such as gradient-boosted trees − to predict binding affinity, and most other architectures contain one or multiple MLPs as subcomponents. Convolutional neural networks (CNNs) have become increasingly popular for binding affinity prediction due to their success on image detection tasks .…”

Geometric Interaction Graph Neural Network for Predicting Protein–Ligand Binding Affinities from 3D Structures (GIGN)

“…With the increasing amount of high-quality experimentally determined protein–ligand structures and their binding affinities, machine learning (ML) methods have been widely used to predict protein–ligand binding affinities or interactions by identifying useful patterns from data. − ,− Existing ML-based methods for PLA prediction can be divided into two categories, interaction-free and interaction-based methods, according to whether the models make decisions relying on physical interactions, as shown in Figure .…”

“…In contrast, interaction-based models − ,− , make predictions based on the 3D structures of complexes and physical interactions of proteins and ligands. Among the interaction-based models, 3D-convolutional neural networks (3D-CNNs) , and interaction graph neural networks (IGNNs) ,,,, are most commonly used to predict binding affinities from 3D structures with atomic interaction information.…”

DeepBindGCN: Integrating Molecular Vector Representation with Graph Convolutional Neural Networks for Accurate Protein-Ligand Interaction Prediction

“…To check its efficiency in virtual screening, we tested its time spent in virtual Dziubinska et al, 2018), midlevel fusion (Jones et al, 2021), GraphBAR (Son and Kim, 2021), AK-score-ensemble (Kwon et al, 2020), DeepAtom (Li et al, 2019), PointNet(B) (Wang et al, 2022), PointTransform(B) (Wang et al, 2022), AEScore (Meli et al, 2021), ResAtom-Score(Y. , DEELIG (Ahmed et al, 2021), PIGNet (ensemble) (Moon et al, 2022), BAPA (Seo et al, 2021), SE-OnionNet(S. , DeepBindRG(H. Zhang, Liao, Saravanan, et al, 2019). We can see that our DeepBindGCN_RG_x has comparable performance with most state-of-art models.…”