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

Abstract

Applying deep learning concepts from image detection and graph theory has greatly advanced protein−ligand binding affinity prediction, a challenge with enormous ramifications for both drug discovery and protein engineering. We build upon these advances by designing a novel deep learning architecture consisting of a 3-dimensional convolutional neural network utilizing channel-wise attention and two graph convolutional networks utilizing attention-based aggregation of node features. HAC-Net (Hybrid Attention-Based Convolutional Neural Network) obtains state-of-the-art results on the PDBbind v.2016 core set, the most widely recognized benchmark in the field. We extensively assess the generalizability of our model using multiple train-test splits, each of which maximizes differences between either protein structures, protein sequences, or ligand extendedconnectivity fingerprints of complexes in the training and test sets. Furthermore, we perform 10-fold cross-validation with a similarity cutoff between SMILES strings of ligands in the training and test sets and also evaluate the performance of HAC-Net on lowerquality data. We envision that this model can be extended to a broad range of supervised learning problems related to structurebased biomolecular property prediction. All of our software is available as an open-source repository at https://github.com/gregorykyro/HAC-Net/, and the HACNet Python package is available through PyPI.

Ligand binding affinity prediction with fusion of graph neural networks and 3D structure-based complex graph

“…In terms of the models trained on the PDBbind refined set, TopBP 75 ( R p = 0.861) is the best one based on CNN and MP-GNN 76 ( R p = 0.852) is the best one based on GNN. Other methods, such as 3D CNN and GNN based HAC-Net, 77 the graph-based DL model GIGN, 78 InteractionGraphNet 79 and PLANET, 80 the topological fingerprints-based model (TNet-Bp 81 ), and the 3D CNN-based model (AK-Score 43 ), also show good scoring power in the CASF-2016 test. Other ML-based methods also perform good, for example PerSpect ML 82 ( R p = 0.84, RMSE = 1.27).…”

Open-Source Machine Learning in Computational Chemistry

“…Several groups have worked on using ML to predict binding affinities. Ligand–receptor binding affinities can be computed using Pafnucy, DeepAffinity, GLXE, OctSurf, OnionNet-2, , and Hybrid Attention-Based Convolutional Neural Network (HAC-Net) . Pafnucy’s convolutional NN makes use of a 3D grid input representation with 1 Å resolution for affinity prediction.…”

“…The OctSurf algorithm computes the 3D surface areas of the protein pocket and ligand to predict a resulting binding affinity . The two-dimensional convolutional NN OnionNet-2 generates rotation-free pairwise contacts between protein and ligand atoms to predict binding free energies. , HAC-Net is one of the newest algorithms, which combines the concept of attention with a 3D convolutional NN to compute protein–ligand binding affinity …”

Quantum Convolutional Neural Networks for Multi-Channel Supervised Learning

“…There has been recent work demonstrating the feasibility and advantages of substituting components of classical ML architectures with quantum analogues [36][37][38][39], such as quantum circuits in place of classical convolutional kernels in convolutional neural networks (CNNs) [40][41][42][43][44][45][46][47][48]. Classical CNNs are state-of-the-art for image, video, and sound recognition tasks [49,50] and also have applications in the natural sciences [35,[51][52][53][54][55][56]. CNNs that incorporate quantum circuits to function as kernels, referred to as Quantum CNNs (QCNNs), have performed well on classification tasks involving simple data such as the MNIST dataset of handwritten digits [42,48], as well as multi-channel data such as the CIFAR-10 dataset [45,57].…”