bioRxiv Science⌕ Search

Biology subjects

Provencher, B.

Publications and source records attributed to Provencher, B..

2 recordsLinked to original sources
bioRxiv2025

Quantum analysis of protein-ligand binding by integrating structural resolution, sequence homology, and ligand properties

Predicting protein-ligand binding affinity is a fundamental challenge in computational biology and drug discovery, complicated by diverse factors including protein sequence variability, ligand chemical diversity, and structural resolution. Here, we present an integrative study that combines classical machine learning and quantum-enhanced modeling to investigate how crystal structure resolution, sequence similarity, and ligand properties jointly influence binding affinity. Using a curated "refined" dataset from PDBbind and an expanded general dataset, we first conduct correlation and regression analyses to quantify the relationships among binding affinity, ligand descriptors (e.g., molecular weight, logP), and protein structural metrics (resolution, R-factor). We observe moderate positive correlations between ligand size/hydrophobicity and affinity, and a slight negative correlation between resolution and affinity in the refined dataset that largely disappears in the general set. We then train multiple predictive models, including random forests, deep neural networks, and quantum-enhanced approaches--quantum kernel methods, variational quantum circuits, and a hybrid classical-quantum neural network. Experimental results show that quantum-enhanced models perform on par with classical methods in predicting binding affinities and, in some cases, offer modest improvements. Notably, a hybrid quantum-classical model achieves the highest accuracy (Pearson correlation R{approx}0.80R) on the refined dataset. These findings highlight the potential of quantum computing for capturing complex patterns in biomolecular data, laying groundwork for improved structure-based drug design. Our study underscores that while data quality and curation greatly influence observed trends, quantum machine learning despite current hardware limitations can already serve as a competitive and promising tool in computational structural biology. AUTHOR SUMMARYIn our study, we confront a major bottleneck in the creation of new medicines: accurately predicting how strongly a potential drug will attach to its target protein in the body. Getting this right early in the process could save billions of dollars and years of research by preventing dead-ends. We investigated if quantum computing, a new technology that processes information in a fundamentally different way, could provide a better solution. We trained several computer models to predict these binding strengths, comparing standard artificial intelligence (AI) with new models enhanced by quantum machine learning. Our results showed that a hybrid model, which strategically combines classical AI with a quantum-powered component, delivered the most accurate predictions on high-quality, curated data. This work demonstrates that quantum computing is already a competitive tool for real-world biological problems, not just a future possibility. We believe that by further developing these quantum-enhanced approaches, we can create more reliable predictive tools to make the long and difficult search for new drugs faster and more successful.

molecular biology↗
bioRxiv2021

Deep learning-based segmentation of high-resolution computed tomography image data outperforms commonly used automatic bone segmentation methods

Segmenting bone from background is required to quantify bone architecture in computed tomography (CT) image data. A deep learning approach using convolutional neural networks (CNN) is a promising alternative method for automatic segmentation. The study objectives were to evaluate the performance of CNNs in automatic segmentation of human vertebral body (micro-CT) and femoral neck (nano-CT) data and to investigate the performance of CNNs to segment data across scanners. Scans of human L1 vertebral bodies (microCT [North Star Imaging], n=28, 53m3) and femoral necks (nano-CT [GE], n=28, 27m3) were used for evaluation. Six slices were selected for each scan and then manually segmented to create ground truth masks (Dragonfly 4.0, ORS). Two-dimensional U-Net CNNs were trained in Dragonfly 4.0 with images of the [FN] femoral necks only, [VB] vertebral bodies only, and [F+V] combined CT data. Global (i.e., Otsu and Yen) and local (i.e., Otsu r = 100) thresholding methods were applied to each dataset. Segmentation performance was evaluated using the Dice coefficient, a similarity metric of overlap. Kruskal-Wallis and Tukey-Kramer post-hoc tests were used to test for significant differences in the accuracy of segmentation methods. The FN U-Net had significantly higher Dice coefficients (i.e., better performance) than the global (Otsu: p=0.001; Yen: p=0.001) and local (Otsu [r=100]: p=0.001) thresholding methods and the VB U-Net (p=0.001) but there was no significant difference in model performance compared to the FN + VB U-net (p=0.783) on femoral neck image data. The VB U-net had significantly higher Dice coefficients than the global and local Otsu (p=0.001 for both) and FN U-Net (p=0.001) but not compared to the Yen (p=0.462) threshold or FN + VB U-net (p=0.783) on vertebral body image data. The results demonstrate that the U-net architecture outperforms common thresholding methods. Further, a network trained with bone data from a different system (i.e., different image acquisition parameters and voxel size) and a different anatomical site can perform well on unseen data. Finally, a network trained with combined datasets performed well on both datasets, indicating that a network can feasibly be trained with multiple datasets and perform well on varied image data.

bioengineering↗