Digital discovery最新文献

{"title":"Challenges and opportunities for machine learning potentials in transition path sampling: alanine dipeptide and azobenzene studies†","authors":"Nikita Fedik, Wei Li, Nicholas Lubbers, Benjamin Nebgen, Sergei Tretiak and Ying Wai Li","doi":"10.1039/D4DD00265B","DOIUrl":"https://doi.org/10.1039/D4DD00265B","url":null,"abstract":"<p >The growing interest in machine learning (ML) tools within chemistry and material science stems from their novelty and ability to predict properties almost as accurately as underlying electronic structure calculations or experiments. Transition path sampling (TPS) offers a practical way to explore transition routes between metastable minima such as conformers and isomers on the multidimensional potential energy surface. However, TPS has historically suffered from the computational cost <em>vs.</em> accuracy trade-off between affordable force-field simulations and expensive high-fidelity quantum mechanical calculations. ML interatomic potentials combined with TPS offer a new approach for the exploration of transition pathways at near-quantum mechanical accuracy, while keeping the computational cost comparable to classical force fields. In this study, we employ the HIP-NN-TS and ANI-1x neural network-based ML potentials, both trained on the ANI-1x dataset of 5 million HCNO structures. We first verify the correctness of our approach by applying it to alanine dipeptide and compare the resulting energy surface and transition paths to the literature. Our findings suggest that proposed approach holds promise for conformational searches, as evidenced by the chemical accuracy (errors ≲1 kcal mol<small>⁻¹</small>) for thermal molecular dynamics trajectories of alanine dipeptide. While we were able to successfully reconstruct alanine dipeptide's potential energy landscape using both HIP-NN-TS and ANI-1x frameworks, we observed that ML models with a lower accuracy may still locate additional important conformations. We also find that manual active learning, augmenting the training data by structures taken from TPS trajectories, improved the accuracy by ∼30% with small amounts of additional data. Finally, we evaluated a more intricate case, azobenzene, and observed that seemingly simple torsions may bear a challenge for ML potentials and limit their applications in TPS. Inability of HIP-NN-TS to correctly describe the energetics of major rotational pathway in azobenzene isomerization highlights deficiencies of the reference method in describing the electronic degrees of freedom. Our study underscores the importance of domain expertise in selecting physically meaningful pathways for benchmarking ML potentials, especially considering the intricacies of electronic structure in chemical dynamics and non-equilibrium processes.</p>","PeriodicalId":72816,"journal":{"name":"Digital discovery","volume":" 5","pages":" 1158-1175"},"PeriodicalIF":6.2,"publicationDate":"2025-04-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://pubs.rsc.org/en/content/articlepdf/2025/dd/d4dd00265b?page=search","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143943995","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Machine learning-driven antiviral libraries targeting respiratory viruses†","authors":"Gabriela Valle-Núñez, Raziel Cedillo-González, Juan F. Avellaneda-Tamayo, Fernanda I. Saldívar-González, Diana L. Prado-Romero and José L. Medina-Franco","doi":"10.1039/D5DD00037H","DOIUrl":"https://doi.org/10.1039/D5DD00037H","url":null,"abstract":"<p >Viral infections represent a significant global health concern. Viral diseases can range from mild symptoms to life-threatening conditions, and the impact of these infections has grown due to increased contagious rates driven by globalization. A prime example is the SARS-CoV-2 pandemic, which emphasized the urgent need to design and develop new antiviral drugs. This study aimed to generate a curated data set of compounds relevant to respiratory infections, focusing on predicting their antiviral activity. Specifically, the study leverages ML classification models to evaluate focused and on-demand compound libraries targeting pathways associated with viral respiratory infections. ML models were trained based on the antiviral biological activity related to respiratory diseases deposited on a major public compound database annotated with biological activity. The models were validated and retrained to classify and design antiviral-focused libraries on seven respiratory targets.</p>","PeriodicalId":72816,"journal":{"name":"Digital discovery","volume":" 5","pages":" 1239-1258"},"PeriodicalIF":6.2,"publicationDate":"2025-04-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://pubs.rsc.org/en/content/articlepdf/2025/dd/d5dd00037h?page=search","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143944050","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Unravelling cyclic peptide membrane permeability prediction: a study on data augmentation, architecture choices, and representation schemes","authors":"Alfonso Cabezón, Erik Otović, Daniela Kalafatovic, Ángel Piñeiro, Rebeca García-Fandiño and Goran Mauša","doi":"10.1039/D4DD00375F","DOIUrl":"https://doi.org/10.1039/D4DD00375F","url":null,"abstract":"<p >Cyclic peptides have emerged as promising candidates for drug development due to their unique structural properties and potential therapeutic benefits. However, clinical applications are limited by their low membrane permeability, which is difficult to predict. This study explores the impact of data augmentation and the inclusion of cyclic structure information in ML modeling to enhance the prediction of membrane permeability of cyclic peptides from their amino acid sequence. Various peptide representation strategies in combination with data augmentation techniques based on amino acid mutations and cyclic permutations were investigated to address the limited availability of experimental data. Moreover, cyclic convolutional layers were explored to explicitly model the cyclic nature of the peptides. The results indicated that combining sequential and peptide properties demonstrated superior performance across multiple metrics. The model performance is highly sensitive to the number and degree of similarity of amino acids involved in mutations. Cyclic permutations improved model performance, particularly in a larger and more diverse dataset and standard architectures captured most of the relevant cyclic information. Highlighting the complexity of peptide-membrane interactions, these results lay a foundation for future improvements in computational methods for the design of cyclic peptide drugs and offer practical guidelines for researchers in this field. The best-performing model was integrated into a user-friendly web-based tool, CYCLOPS: CYCLOpeptide Permeability Simulator (available at http://cyclopep.com/cyclops), to facilitate wider accessibility and application in drug discovery community. This tool allows for rapid predictions of the membrane permeability for cyclic peptides with a classification accuracy score of 0.824 and a regression mean absolute error of 0.477.</p>","PeriodicalId":72816,"journal":{"name":"Digital discovery","volume":" 5","pages":" 1259-1275"},"PeriodicalIF":6.2,"publicationDate":"2025-04-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://pubs.rsc.org/en/content/articlepdf/2025/dd/d4dd00375f?page=search","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143944051","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A workflow to create a high-quality protein–ligand binding dataset for training, validation, and prediction tasks†","authors":"Yingze Wang, Kunyang Sun, Jie Li, Xingyi Guan, Oufan Zhang, Dorian Bagni, Yang Zhang, Heather A. Carlson and Teresa Head-Gordon","doi":"10.1039/D4DD00357H","DOIUrl":"10.1039/D4DD00357H","url":null,"abstract":"<p >Development of scoring functions (SFs) used to predict protein–ligand binding energies requires high-quality 3D structures and binding assay data for training and testing their parameters. In this work, we show that one of the widely-used datasets, PDBbind, suffers from several common structural artifacts of both proteins and ligands, which may compromise the accuracy, reliability, and generalizability of the resulting SFs. Therefore, we have developed a series of algorithms organized in a semi-automated workflow, HiQBind-WF, that curates non-covalent protein–ligand datasets to fix these problems. We also used this workflow to create an independent data set, HiQBind, by matching binding free energies from various sources including BioLiP, Binding MOAD and Binding DB with co-crystalized ligand–protein complexes from the PDB. The resulting HiQBind workflow and dataset are designed to ensure reproducibility and to minimize human intervention, while also being open-source to foster transparency in the improvements made to this important resource for the biology and drug discovery communities.</p>","PeriodicalId":72816,"journal":{"name":"Digital discovery","volume":" 5","pages":" 1209-1220"},"PeriodicalIF":6.2,"publicationDate":"2025-04-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11967698/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143796976","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Lightweight target detection for large-field ddPCR images based on improved YOLOv5†","authors":"Xingyu Jin, Jing Yang, Xiaorui Jiang, Zhenqing Li, Jinrong Shen, Zhiheng Yu, Cunliang Yang, Fengli Huang, Dunlu Peng, Yoshinori Yamaguchi and Jijun Feng","doi":"10.1039/D5DD00006H","DOIUrl":"https://doi.org/10.1039/D5DD00006H","url":null,"abstract":"<p >The large-field rapid nucleic acid concentration measurement system is capable of achieving one-time gene chip imaging with high resolution. However, it encounters challenges in the precise detection of positive microchambers, which is caused by factors such as reagent residue, uneven lighting, and environmental noise. Herein we proposed an improved, lightweight algorithm based on You Only Look Once (YOLOv5) for detecting the positive microchambers. We determined appropriate detection scales based on the target size distribution and utilized the bidirectional feature pyramid network (BiFPN) for efficient multi-scale feature fusion. To reduce model size without sacrificing performance, GhostConv, C3Ghost, and a simple, parameter-free attention module (SimAM) were integrated into the network, followed by network pruning. The improved YOLOv5 model was trained on a self-built dataset, and employed a partitioned fusion prediction strategy to detect large-field ddPCR images by self-developed software. In contrast to single-stage lightweight object detection algorithms, our model features a mere 1.5MB size while achieving 99.5% precision, 99.5% recall, and a 78.1% mAP(0.5 : 0.95), significantly reducing the system's demand for computing resources without compromising efficiency and accuracy.</p>","PeriodicalId":72816,"journal":{"name":"Digital discovery","volume":" 5","pages":" 1298-1305"},"PeriodicalIF":6.2,"publicationDate":"2025-04-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://pubs.rsc.org/en/content/articlepdf/2025/dd/d5dd00006h?page=search","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143944054","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Improving reaction prediction through chemically aware transfer learning†","authors":"Angus Keto, Taicheng Guo, Nils Gönnheimer, Xiangliang Zhang, Elizabeth H. Krenske and Olaf Wiest","doi":"10.1039/D4DD00412D","DOIUrl":"https://doi.org/10.1039/D4DD00412D","url":null,"abstract":"<p >Practical applications of machine learning (ML) to new chemical domains are often hindered by data scarcity. Here we show how data gaps can be circumvented by means of transfer learning that leverages chemically relevant pre-training data. Case studies are presented in which the outcomes of two classes of pericyclic reactions are predicted: [3,3] rearrangements (Cope and Claisen rearrangements) and [4 + 2] cycloadditions (Diels–Alder reactions). Using the graph-based generative algorithm NERF, we evaluate the data efficiencies achieved with different starting models that we pre-trained on datasets of different sizes and chemical scope. We show that the greatest data efficiency is obtained when the pre-training is performed on smaller datasets of mechanistically related reactions (Diels–Alder, Cope and Claisen, Ene, and Nazarov) rather than &gt;50× larger datasets of mechanistically unrelated reactions (USPTO-MIT). These small bespoke datasets were more efficient in both low re-training and low pre-training regimes, and are thus recommended alternatives to large diverse datasets for pre-training ML models.</p>","PeriodicalId":72816,"journal":{"name":"Digital discovery","volume":" 5","pages":" 1232-1238"},"PeriodicalIF":6.2,"publicationDate":"2025-03-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://pubs.rsc.org/en/content/articlepdf/2025/dd/d4dd00412d?page=search","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143944049","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Paddy: an evolutionary optimization algorithm for chemical systems and spaces†","authors":"Armen G. Beck, Sanjay Iyer, Jonathan Fine and Gaurav Chopra","doi":"10.1039/D4DD00226A","DOIUrl":"https://doi.org/10.1039/D4DD00226A","url":null,"abstract":"<p >Optimization of chemical systems and processes have been enhanced and enabled by the development of new algorithms and analytical approaches. While several methods systematically investigate how underlying variables correlate with a given outcome, there is often a substantial number of experiments needed to accurately model such relationships. As chemical systems increase in complexity, algorithms are needed to propose experiments that efficiently optimize the underlying objective, while effectively sampling parameter space to avoid convergence on local minima. We have developed the Paddy software package based on the Paddy field algorithm, a biologically inspired evolutionary optimization algorithm that propagates parameters without direct inference of the underlying objective function. We benchmarked Paddy against several optimization approaches: the Tree of Parzen Estimator through the Hyperopt software library, Bayesian optimization with a Gaussian process <em>via</em> Meta's Ax framework, and two population-based methods from EvoTorch—an evolutionary algorithm with Gaussian mutation, and a genetic algorithm using both a Gaussian mutation and single-point crossover—all representing diverse approaches to optimization. Paddy's performance is benchmarked for mathematical and chemical optimization tasks including global optimization of a two-dimensional bimodal distribution, interpolation of an irregular sinusoidal function, hyperparameter optimization of an artificial neural network tasked with classification of solvent for reaction components, targeted molecule generation by optimizing input vectors for a decoder network, and sampling discrete experimental space for optimal experimental planning. Paddy demonstrates robust versatility by maintaining strong performance across all optimization benchmarks, compared to other algorithms with varying performance. Additionally, Paddy avoids early convergence with its ability to bypass local optima in search of global solutions. We anticipate that the facile, versatile, robust and open-source nature of Paddy will serve as a toolkit in chemical problem-solving tasks towards automated experimentation with high priority for exploratory sampling and innate resistance to early convergence to identify optimal solutions.</p>","PeriodicalId":72816,"journal":{"name":"Digital discovery","volume":" 5","pages":" 1352-1371"},"PeriodicalIF":6.2,"publicationDate":"2025-03-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://pubs.rsc.org/en/content/articlepdf/2025/dd/d4dd00226a?page=search","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143944076","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Improving structural plausibility in diffusion-based 3D molecule generation via property-conditioned training with distorted molecules†","authors":"Lucy Vost, Vijil Chenthamarakshan, Payel Das and Charlotte M. Deane","doi":"10.1039/D4DD00331D","DOIUrl":"https://doi.org/10.1039/D4DD00331D","url":null,"abstract":"<p >Traditional drug design methods are costly and time-consuming due to their reliance on trial-and-error processes. As a result, computational methods, including diffusion models, designed for molecule generation tasks have gained significant traction. Despite their potential, they have faced criticism for producing physically implausible outputs. As a solution to this problem, we propose a conditional training framework resulting in a model capable of generating molecules of varying and controllable levels of structural plausibility. This framework consists of adding distorted molecules to training datasets, and then annotating each molecule with a label representing the extent of its distortion, and hence its quality. By training the model to distinguish between favourable and unfavourable molecular conformations alongside the standard molecule generation training process, we can selectively sample molecules from the high-quality region of learned space, resulting in improvements in the validity of generated molecules. In addition to the standard two datasets used by molecule generation methods (QM9 and GEOM), we also test our method on a druglike dataset derived from ZINC. We use our conditional method with EDM, the first E(3) equivariant diffusion model for molecule generation, as well as two further models—a more recent diffusion model and a flow matching model—which were built off EDM. We demonstrate improvements in validity as assessed by RDKit parsability and the PoseBusters test suite; more broadly, though, our findings highlight the effectiveness of conditioning methods on low-quality data to improve the sampling of high-quality data.</p>","PeriodicalId":72816,"journal":{"name":"Digital discovery","volume":" 4","pages":" 1092-1099"},"PeriodicalIF":6.2,"publicationDate":"2025-03-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://pubs.rsc.org/en/content/articlepdf/2025/dd/d4dd00331d?page=search","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143809074","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Efficient strategies for reducing sampling error in quantum Krylov subspace diagonalization","authors":"Gwonhak Lee, Seonghoon Choi, Joonsuk Huh and Artur F. Izmaylov","doi":"10.1039/D4DD00321G","DOIUrl":"https://doi.org/10.1039/D4DD00321G","url":null,"abstract":"<p >Within the realm of early fault-tolerant quantum computing (EFTQC), quantum Krylov subspace diagonalization (QKSD) has emerged as a promising quantum algorithm for the approximate Hamiltonian diagonalization <em>via</em> projection onto the quantum Krylov subspace. However, the algorithm often requires solving an ill-conditioned generalized eigenvalue problem (GEVP) involving erroneous matrix pairs, which can significantly distort the solution. Since EFTQC assumes limited-scale error correction, finite sampling error becomes a dominant source of error in these matrices. This work focuses on quantifying sampling errors during the measurement of matrix element in the projected Hamiltonian examining two measurement approaches based on the Hamiltonian decompositions: the linear combination of unitaries and diagonalizable fragments. To reduce sampling error within a fixed budget of quantum circuit repetitions, we propose two measurement strategies: the shifting technique and coefficient splitting. The shifting technique eliminates redundant Hamiltonian components that annihilate either the bra or ket states, while coefficient splitting optimizes the measurement of common terms across different circuits. Numerical experiments with electronic structures of small molecules demonstrate the effectiveness of these strategies, reducing sampling costs by a factor of 20–500.</p>","PeriodicalId":72816,"journal":{"name":"Digital discovery","volume":" 4","pages":" 954-969"},"PeriodicalIF":6.2,"publicationDate":"2025-03-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://pubs.rsc.org/en/content/articlepdf/2025/dd/d4dd00321g?page=search","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143809089","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Machine learning-driven optimization of the output force in photo-actuated organic crystals†","authors":"Kazuki Ishizaki, Toru Asahi and Takuya Taniguchi","doi":"10.1039/D4DD00380B","DOIUrl":"https://doi.org/10.1039/D4DD00380B","url":null,"abstract":"<p >Photo-actuated organic crystals that can be remotely controlled by light are gaining attention as next-generation actuator materials. In the practical application of actuator materials, the mode of deformation and the output force are important properties. Since the output force depends on the crystal properties and experimental conditions, it is necessary to explore the optimal conditions from a vast parameter space. In this study, we employed two types of machine learning for molecular design and experimental optimization to maximize the blocking force. Machine learning in molecular design led to the creation of a material pool of salicylideneamine derivatives. Bayesian optimization was used for efficient sampling from the material pool for force measurements in the real world, achieving a maximum blocking force of 37.0 mN. This method was at least 73 times more efficient than the grid search approach.</p>","PeriodicalId":72816,"journal":{"name":"Digital discovery","volume":" 5","pages":" 1199-1208"},"PeriodicalIF":6.2,"publicationDate":"2025-03-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://pubs.rsc.org/en/content/articlepdf/2025/dd/d4dd00380b?page=search","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143944047","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}