Splitting Strategies Research Articles

SS-DNN: A hybrid strang splitting deep neural network approach for solving the Allen–Cahn equation

The Allen–Cahn equation is a fundamental partial differential equation that describes phase separation and interface motion in materials science, physics, and various other scientific domains. The presence of interfacial width (ϵ) between two stable phases, associated with a nonlinear term, is a small positive parameter which makes the problem more challenging to solve as ϵ approaches zero. This paper proposes a novel hybrid deep splitting method to efficiently and accurately solve the Allen–Cahn equation in a convex polygonal domain in Rd(d=1,2,3). The method combines the benefits of deep learning and splitting strategies, leveraging the strengths of both approaches. Essentially, a second-order splitting method is employed to split the Allen–Cahn equation into two simpler linear and non-linear sub-problems. While the nonlinear sub-problem can be solved analytically, the deep neural network is utilized to approximate the linear sub-problem. By integrating deep learning into the splitting strategy, we achieve a more efficient and accurate solution for the Allen–Cahn equation, demonstrating promising results. We also derive an error estimate for the proposed hybrid method. Modified space adaptivity and transform learning techniques are employed to enhance the efficiency of the neural network.

A path splitting and pruning strategy on list decoder for PAC codes

AbstractRecently, Arıkan proposed a polarization‐adjusted convolutional (PAC) codes, demonstrating their superior error correction performance over polar codes at short block lengths. It was confirmed that PAC codes approached the optimal performance achievable with limited code length. This paper proposes a novel low‐complexity list decoding algorithm for PAC codes, incorporating path splitting and pruning strategies based on a set of highly reliable information bits. Simulation results reveal that the proposed algorithm significantly reduces sorting complexity and average list size, all while incurring negligible performance loss. Unlike previous pruning algorithms designed for polar codes, the proposed strategy eliminates the need to individually assess the reliability of decoding paths in each decoding process. Instead, the algorithm minimizes redundant decoding paths through a high‐reliability information bit set, constructed using Monte Carlo experiments.

Approaching Optimal pH Enzyme Prediction with Large Language Models.

Enzymes are widely used in biotechnology due to their ability to catalyze chemical reactions: food making, laundry, pharmaceutics, textile, brewing─all these areas benefit from utilizing various enzymes. Proton concentration (pH) is one of the key factors that define the enzyme functioning and efficiency. Usually there is only a narrow range of pH values where the enzyme is active. This is a common problem in biotechnology to design an enzyme with optimal activity in a given pH range. A large part of this task can be completed in silico, by predicting the optimal pH of designed candidates. The success of such computational methods critically depends on the available data. In this study, we developed a language-model-based approach to predict the optimal pH range from the enzyme sequence. We used different splitting strategies based on sequence similarity, protein family annotation, and enzyme classification to validate the robustness of the proposed approach. The derived machine-learning models demonstrated high accuracy across proteins from different protein families and proteins with lower sequence similarities compared with the training set. The proposed method is fast enough for the high-throughput virtual exploration of protein space for the search for sequences with desired optimal pH levels.

Reinforcement learning for distributed hybrid flowshop scheduling problem with variable task splitting towards mass personalized manufacturing

Mass personalization manufacturing (MPM), an emerging production pattern, aims to improve enterprise profit in modern industries. However, the processing of heterogeneous orders from the consumers complicates such production scheduling problem. In addition, different scale tasks should adopt different splitting strategies in practical manufacturing, which makes the task splitting method more worthy of investigation. Towards MPM, this paper presents a distributed hybrid flowshop scheduling problem with variable task splitting (DHFSP-VTS) to minimize the makespan and total energy consumption simultaneously. Meanwhile, the VTS allows the tasks to be split into different sublots so they can save setup and transfer time. To solve these problems, we present an order modularization processing method that can categorize multiple types of orders into specific generation tasks, and a highly effective reinforcement learning-multiple objective evolutionary algorithm based on decomposition (RL-MOEA/D) is designed. In RL-MOEA/D, there are three features: (1) three initial rules are used for initialization based on the current splitting scheme that can increase the diversity of solutions; (2) the reinforcement learning agent uses the Q-learning mechanism to dynamically select the scheme of task splitting as action; (3) a neighborhood search strategy improves the exploitation ability and expand the solution space. To verify the effectiveness of RL-MOEA/D, the MOEA/Ds based on four splitting schemes and four RL combined meta-heuristics are compared on 18 instances. The results show that RL-MOEA/D can obtain the best optimization and stability of all the other comparison algorithms. Therefore, it’s a new technique to solve DHFSP with large-scale tasks, especially for the problem of MPM.

Integrated optimization of worker assignment, batch splitting and scheduling for a hybrid assembly line-seru production system

This paper studies an integrated optimization problem of worker assignment, batch splitting and scheduling for a hybrid assembly line-seru production system, which includes multiple serus and a short assembly line. The worker assignment involves assigning each multi-skilled worker to a seru or the assembly line, and assigning the worker tasks to perform. The batch splitting determines the number of subbatches split for each batch, the size of each subbatch, and in which seru the production of each subbatch is performed. The batch scheduling aims to optimize the process sequences of batches in serus and on the assembly line. For the integrated optimization problem, we propose a mixed-integer nonlinear programming model that minimizes the makespan. To solve large-scale instances of the problem, we develop a tailored hybrid variable neighborhood search algorithm with a four-layer solution coding scheme and four novel neighborhood structures. We validate the effectiveness of our model and solution algorithm on two sets of instances. The experimental results indicate that the proposed hybrid metaheuristic achieves a significant reduction in makespan, with an improvement of up to 20.94% compared to the solutions obtained by the Gurobi optimization solver within one hour. In addition, we demonstrate the advantage of optimized batch splitting by comparing it with no batch splitting and equal-size batch splitting strategies in the literature.

Machine learning potentials for global multi-timescale diffuse irradiance estimation: Synthesizing ground observations, time-series, and environmental features

Separating diffuse horizontal irradiance (DHI) from ground-based global horizontal irradiance observations is critical owing to lack of direct DHI observations. Existing models are often site-specific and time-bound, thereby limiting their universal applicability. To address this, this study thoroughly explores machine learning (ML) for constructing global separation models. We develop 36 models using three ML algorithms—light gradient boosting machine (LightGBM), generalized additive model (GAM), and geographically weighted artificial neural network (GWANN)—alongside three data splitting strategies for minutely, 10-minutely, hourly, and daily timescales. LightGBM and GAM outperform GWANN in accuracy, with LightGBM excelling in interpretability, efficiency, and handling missing values, while GWANN exhibits superior stability. Compared with completely random splitting, model accuracy decreases by 2.26 % and 2.16 % for station- and date-based splitting, respectively. Prediction accuracy varies across timescales, with minutely models typically considered the best. Key predictors are Kt, solar zenith angle, and altitude, complemented by the significant influence of time-series Kt, aerosol, and cloud features. The influence of various factors on model accuracy differs and is scale-dependent. Among them, time-series Kt variability factors notably enhance ML-based predictions, especially at minutely scales. The LightGBM model outperforms classic models in overall and site-specific accuracy and adaptability, but its computational efficiency declines, especially with time-series variability factors. The findings highlight that ML-based separation models necessitate careful selection of algorithm, data splitting strategy, and input variables, fully considering study region, data situation, and temporal and spatial scales. This research offers a potential solution for constructing global separation models and insights for model improvement.

Efficient Heuristic Management of Cloud Data Lakes for Enhanced Big Data Handling in Cloud Computing Environments

Efficient control of cloud data lakes is important for improving big data dealing in cloud computing settings. Cloud data lakes serve as stores for vast amounts of organized, semi-structured, and unstructured data from varied sources, offering scaling and freedom in data storage and processing. However, the sheer amount and range of data offer difficulties in improving storage, access, and analysis processes. This study suggests a heuristic method to handle these issues by improving the order and recovery of data within cloud data lakes. The heuristic method focuses on automatically handling data splitting, storage, and caching strategies based on data access trends and task factors. By leveraging heuristic principles, such as adaptable splitting and predictive caching, the suggested method aims to reduce delay and improve total query speed. Key components of the heuristic management include adaptable splitting methods that actively change data partition sizes based on access frequency and data distribution patterns. This flexible method ensures that frequently accessed data is easily available, while less accessed data is efficiently saved to improve storage usage. Furthermore, predictive storing methods expect future data access patterns using machine learning models or previous data trends. By strategically saving relevant datasets in memory or high-speed storage, the system reduces delay for repeated questions, thereby improving real-time data processing capabilities. The success of the suggested heuristic management method is tested through models and benchmarks using real big data tasks. Comparative research against standard static splitting and caching methods shows significant changes in query response times and resource utilization efficiency.

Deep learning algorithms for detecting freezing of gait in Parkinson’s disease: A cross-dataset study

Freezing of gait is a complex and disabling symptom of Parkinson’s disease, which has a significant impact on the patients’ quality of life and increases the risk of falls and related injuries. This study aims to evaluate the generalization capability of deep learning algorithms in freezing of gait detection. To address this task, various machine learning and deep learning algorithms were implemented, fine-tuned, and evaluated using diverse data splitting and validation strategies. The experiments performed yielded mixed results. Although the implementations demonstrated competitive performance in single-dataset settings (area under the curve ranging from 0.77 to 0.94), all approaches showed limited robustness in cross-dataset tests and suboptimal generalization across different datasets (area under the curve ranging from 0.65 to 0.84). These results highlight the importance of standardized data collection procedures to ensure uniformity. The specification of sensor settings and predefined sensor locations can foster homogeneity in datasets, even when dealing with diverse subjects and environments. Such standardization efforts are crucial for advancing generalized methodologies in the detection of freezing of gait, applicable to both research and clinical applications.

A colour vision test assisted by neural networks

Aims/Purpose: Traditional computerized colour vision tests resort to the magnitude of chromatic thresholds to determine the quality of the colour vision of an observer. The limits of perceived colours are tested against a background that will be conspicuous to normal colour vision observers, but inconspicuous to colour vision deficient (CVD) observers. The purpose of this work was to implement a Neural Network capable of performing independent and automatic classification of the colour vision of an observer, using as inputs the sRGB chromaticity coordinates, response times, and eye movements obtained using a colour vision test on a computer screen calibrated in colour and luminance (Display++).Methods: Chromatic discrimination limits were estimated using a staircase procedure that tested colour vision close to the dichromatic confusion lines. Only the last 8 responses were considered in the analysis. 50 normal colour vision observers and 8 CVD observers performed the test on Display++. CVD participants were diagnosed using the results of the anomaloscope and the CAD colour vision tests. A Sequential Neural Network was implemented with a dense structure (256–128–64‐16‐1 nodes), with relu and sigmoid activation functions, dropout commands to prevent overfitting, the Adam optimizer, and the binary crossentropy loss function. The performance of the neural network was estimated by computing its accuracy, where variables are related to the results obtained after training the Neural Network and inspecting the test results. The database was divided into a training/test following different splitting strategies: T1 observers' responses considering sRGB coordinates, response times, and eye movements. T2 observers' responses considering only sRGB coordinates.Results: The averages and standard deviations of the estimated accuracy for five consecutive trials with a random division between training/test for conditions T1 and T2 was 91% ± 7% and 93% ± 5% respectively. It was found that the implemented neural network had an accuracy of more than 90% across all conditions.Conclusions: The accuracies found across conditions considered seem to indicate that the implemented Neural Network can analyse and classify the colour vision of human observers, with an accuracy greater than 90% of success.

An endeavor from the Glowinski-Le Tallec splitting for approximating the solution of Kawarada equation

This paper studies an extended application of the Glowinski-Le Tallec splitting for approximating solutions of linear and nonlinear partial differential equations. It is shown that the three-level, six-component operator decomposition, originally designed for Lagrangian optimizations, provides a stable second-order operator splitting approximation for the solutions of evolutional partial differential equations. It is also found that the Glowinski-Le Tallec formula not only provides an effective enhancement to conventional two-level, four-component ADI and LOD methods, but also introduces a flexible way for constructing multi-parameter operator splitting strategies in respective spaces where broad spectrums of mathematical models may exist for important natural phenomena and applications. The extended operator splitting is utilized for solving a singular and nonlinear Kawarada problem satisfactorily. Multiple simulation results are presented.

SiSGC: A Drug Repositioning Prediction Model Based on Heterogeneous Simplifying Graph Convolution.

Drug repositioning plays a key role in disease treatment. With the large-scale chemical data increasing, many computational methods are utilized for drug-disease association prediction. However, most of the existing models neglect the positive influence of non-Euclidean data and multisource information, and there is still a critical issue for graph neural networks regarding how to set the feature diffuse distance. To solve the problems, we proposed SiSGC, which makes full use of the biological knowledge information as initial features and learns the structure information from the constructed heterogeneous graph with the adaptive selection of the information diffuse distance. Then, the structural features are fused with the denoised similarity information and fed to the advanced classifier of CatBoost to make predictions. Three different data sets are used to confirm the robustness and generalization of SiSGC under two splitting strategies. Experiment results demonstrate that the proposed model achieves superior performance compared with the six leading methods and four variants. Our case study on breast neoplasms further indicates that SiSGC is trustworthy and robust yet simple. We also present four drugs for breast cancer treatment with high confidence and further give an explanation for demonstrating the rationality. There is no doubt that SiSGC can be used as a beneficial supplement for drug repositioning.

DeepFPlearn +: enhancing toxicity prediction across the chemical universe using graph neural networks.

Sophisticated approaches for the in silico prediction of toxicity are required to support the risk assessment of chemicals. The number of chemicals on the global chemical market and the speed of chemical innovation stand in massive contrast to the capacity for regularizing chemical use. We recently proved our ready-to-use application deepFPlearn as a suitable approach for this task. Here, we present its extension deepFPlearn+ incorporating (i) a graph neural network to feed our AI with a more sophisticated molecular structure representation and (ii) alternative train-test splitting strategies that involve scaffold structures and the molecular weights of chemicals. We show that the GNNs outperform the previous model substantially and that our models can generalize on unseen data even with a more robust and challenging test set. Therefore, we highly recommend the application of deepFPlearn+ on the chemical inventory to prioritize chemicals for experimental testing or any chemical subset of interest in monitoring studies. The software is compatible with python 3.6 or higher, and the source code can be found on our GitHub repository: https://github.com/yigbt/deepFPlearn. The data underlying this article are available in Zenodo, and can be accessed with the link below: https://zenodo.org/record/8146252. Detailed installation guides via Docker, Singularity, and Conda are provided within the repository for operability across all operating systems.

Preconditioned weighted full orothogonalization method for solving singular linear systems from PageRank problems

AbstractThe PageRank model, which was first proposed by Google for its web search engine application, has since become a popular computational tool in a wide range of scientific fields, including chemistry, bioinformatics, neuroscience, bibliometrics, social networks, and others. PageRank calculations necessitate the use of fast computational techniques with low algorithmic and memory complexity. In recent years, much attention has been paid to Krylov subspace algorithms for solving difficult PageRank linear systems, such as those with large damping parameters close to one. In this article, we examine the full orthogonalization method (FOM). We present a convergence study of the method that extends and clarifies part of the conclusions reached in Zhang et al. (J Comput Appl Math. 2016; 296:397–409.). Furthermore, we demonstrate that FOM is breakdown free when solving singular PageRank linear systems with index one and we investigate the effect of using weighted inner‐products instead of conventional inner‐products in the orthonormalization procedure on FOM convergence. Finally, we develop a shifted polynomial preconditioner that takes advantage of the special structure of the PageRank linear system and has a good ability to cluster most of the eigenvalues, making it a good choice for an iterative method like FOM or GMRES. Numerical experiments are presented to support the theoretical findings and to evaluate the performance of the new weighted preconditioned FOM PageRank solver in comparison to other established solvers for this class of problem, including conventional stationary methods, hybrid combinations of stationary and Krylov subspace methods, and multi‐step splitting strategies.

SplitPlace: AI Augmented Splitting and Placement of Large-Scale Neural Networks in Mobile Edge Environments

In recent years, deep learning models have become ubiquitous in industry and academia alike. Deep neural networks can solve some of the most complex pattern-recognition problems today, but come with the price of massive compute and memory requirements. This makes the problem of deploying such large-scale neural networks challenging in resource-constrained mobile edge computing platforms. To solve this, a promising solution is to split resource-hungry neural networks into lightweight disjoint smaller components for pipelined distributed processing. At present, there are two main approaches to do this: semantic and layer-wise splitting. The former partitions a neural network into parallel disjoint models that produce a part of the result, whereas the latter partitions into sequential models that produce intermediate results. However, there is no intelligent algorithm that decides which splitting strategy to use and places such modular splits to edge nodes for optimal performance. To combat this, this work proposes a novel AI-driven online policy, SplitPlace, that uses Multi-Armed-Bandits to intelligently decide between layer and semantic splitting strategies based on the input task's service deadline demands. SplitPlace places such neural network split fragments on mobile edge devices using decision-aware reinforcement learning for efficient and scalable computing. Moreover, SplitPlace fine-tunes its placement engine to adapt to volatile environments. Our experiments on physical mobile-edge environments with real-world workloads show that SplitPlace can significantly improve the state-of-the-art in terms of average response time, deadline violation rate, inference accuracy, and total reward by up to 46, 69, 3 and 12 percent respectively.

Efficient SNN multi-cores MAC array acceleration on SpiNNaker 2.

The potential low-energy feature of the spiking neural network (SNN) engages the attention of the AI community. Only CPU-involved SNN processing inevitably results in an inherently long temporal span in the cases of large models and massive datasets. This study introduces the MAC array, a parallel architecture on each processing element (PE) of SpiNNaker 2, into the computational process of SNN inference. Based on the work of single-core optimization algorithms, we investigate the parallel acceleration algorithms for collaborating with multi-core MAC arrays. The proposed Echelon Reorder model information densification algorithm, along with the adapted multi-core two-stage splitting and authorization deployment strategies, achieves efficient spatio-temporal load balancing and optimization performance. We evaluate the performance by benchmarking a wide range of constructed SNN models to research on the influence degree of different factors. We also benchmark with two actual SNN models (the gesture recognition model of the real-world application and balanced random cortex-like network from neuroscience) on the neuromorphic multi-core hardware SpiNNaker 2. The echelon optimization algorithm with mixed processors realizes 74.28% and 85.78% memory footprint of the original MAC calculation on these two models, respectively. The execution time of echelon algorithms using only MAC or mixed processors accounts for ≤ 24.56% of the serial ARM baseline. Accelerating SNN inference with algorithms in this study is essentially the general sparse matrix-matrix multiplication (SpGEMM) problem. This article explicitly expands the application field of the SpGEMM issue to SNN, developing novel SpGEMM optimization algorithms fitting the SNN feature and MAC array.

Cost Splitting for Multi-Objective Conflict-Based Search

The Multi-Objective Multi-Agent Path Finding (MO-MAPF) problem is the problem of finding the Pareto-optimal frontier of collision-free paths for a team of agents while minimizing multiple cost metrics. Examples of such cost metrics include arrival times, travel distances, and energy consumption. In this paper, we focus on the Multi-Objective Conflict-Based Search (MO-CBS) algorithm, a state-of-the-art MO-MAPF algorithm. We show that the standard splitting strategy used by MO-CBS can lead to duplicate search nodes and hence can duplicate the search effort of MO-CBS. To address this issue, we propose two new splitting strategies for MO-CBS, namely cost splitting and disjoint cost splitting. Our theoretical results show that, when using either splitting strategy, MO-CBS maintains its completeness and optimality guarantees. Our experimental results show that disjoint cost splitting, our best splitting strategy, speeds up MO-CBS by up to two orders of magnitude and substantially improves its success rates in various settings.

An industrial evaluation of proteochemometric modelling: Predicting drug-target affinities for kinases

Deep learning proteochemometric (PCM) models have been reported to achieve excellent performances on public benchmarking datasets. Nevertheless, numerous papers have cast doubt on commonly used evaluation metrics, suggesting they do not reflect true prospective predictive abilities. The aim of this study is to provide a comprehensive assessment of performance of a state-of-the-art PCM model on proprietary data and evaluate its potential over other modelling approaches as a virtual screening tool for kinase inhibitors. Whilst the model has been shown to achieve an RMSE of 0.48 on a public benchmarking dataset, an impaired overall performance was observed for the proprietary dataset in this study, with an RMSE of 0.85 and a Pearson Correlation Coefficient of 0.65 using a temporal splitting strategy. We hypothesise that the more limited performance can be in part attributed to a shift in the chemical space observed over time in an industrial setting, which is not considered by the more lenient random ligand splitting strategy, more commonly used on benchmarking datasets. The overall performance of the PCM model was statistically similar to a multitask model and only slightly superior to a KNN and random forest PCM model. A comprehensive analysis of performance was performed to capture the key challenges faced in the design of competitive kinase inhibitors, which revealed the key limitations of PCM modelling. For example, the model showed poor predictive abilities for understudied targets, and a limited ability to assess ligand selectivity and promiscuity, with no improved performance over a multitask model or a random forest PCM model. Overall, these findings reveal that the PCM model assessed in this study does not provide significant benefits over less complex models such as multitask model or a random forest PCM model as a virtual screening tool for kinase inhibitors in an industrial setting. Taken together, this study highlights the need for more robust evaluations of PCM models by using stricter splitting strategies, more extensive benchmarking and more comprehensive performance analysis beyond traditional metrics.

Degradation-oriented real-time power control of fuel cell hybrid vehicles under uncertain driving conditions

This work provides a real-time power allocation algorithm to address uncertain actual driving situations for fuel cell hybrid vehicles. To predict the vehicle speed under nondeterministic driving conditions, a fusion prediction model is developed based on the advantages of the Markov chain and neural network. The optimal power splitting decision in each receding horizon is then solved using the Pontryagin's minimum principle (PMP) method, considering fuel consumption, State of Charge (SOC), and performance degradation. A degradation model of electrochemical active surface area (ECSA) based on Pt catalyst dissolution was developed. Then the effect of the energy management algorithm on fuel cell degradation was evaluated using the degradation model. Compared with the two conventional real-time power splitting strategies, the approach suggested in this research can better reduce the fuel consumption and maintain the stability of battery SOC with a lower fluctuation while taking into account the degradation of the fuel cell.

Evaluating semi- and nonparametric regression algorithms in quantifying stem taper and volume with alternative test data selection strategies

Abstract Accurately quantifying stem taper is essential to predict diameter at any given height along the stem and to estimate tree volume for various sections of the stem. With increased computing power, semi- and nonparamatric methods have been proposed as alternative approaches for modelling tree taper. The main objective of this study was to assess the accuracy of stem taper predicted for four pine and four hardwood species by semi- and nonparametric models. Specifically, generalized additive models (GAM), random forests (RF) and regression-enhanced random forests (RERF) were compared with two widely-used parametric models: variable-exponent function (VAR) and segmented polynomial regression model (SEG). Test datasets selected from four different data splitting methods were used to examine the prediction accuracy of the models. Results showed that all examined models can be used to quantify stem taper and volume for all species when prediction is limited to be within the range of tree sizes used in model building. The nonparametric RF algorithm generally produced higher bias and lower precision than the semiparametric (GAM and RERF) and parametric models (VAR and SEG). For all species, GAM, VAR and SEG provided more robust predictions of stem taper than RF and RERF algorithms, especially when small or large trees were withheld for model testing. The data splitting strategies used in this work provide an efficient ‘stress test’ to evaluate model performance when collecting an independent test dataset is not feasible. The findings of this work provide additional insights for forest practitioners and resource managers to select appropriate methods in stem taper modelling.

Improved GNNs for Log D7.4 Prediction by Transferring Knowledge from Low-Fidelity Data

The n-octanol/buffer solution distribution coefficient at pH = 7.4 (log D7.4) is an indicator of lipophilicity, and it influences a wide variety of absorption, distribution, metabolism, excretion, and toxicity (ADMET) properties and druggability of compounds. In log D7.4 prediction, graph neural networks (GNNs) can uncover subtle structure-property relationships (SPRs) by automatically extracting features from molecular graphs that facilitate the learning of SPRs, but their performances are often limited by the small size of available datasets. Herein, we present a transfer learning strategy called pretraining on computational data and then fine-tuning on experimental data (PCFE) to fully exploit the predictive potential of GNNs. PCFE works by pretraining a GNN model on 1.71 million computational log D data (low-fidelity data) and then fine-tuning it on 19,155 experimental log D7.4 data (high-fidelity data). The experiments for three GNN architectures (graph convolutional network (GCN), graph attention network (GAT), and Attentive FP) demonstrated the effectiveness of PCFE in improving GNNs for log D7.4 predictions. Moreover, the optimal PCFE-trained GNN model (cx-Attentive FP, Rtest2 = 0.909) outperformed four excellent descriptor-based models (random forest (RF), gradient boosting (GB), support vector machine (SVM), and extreme gradient boosting (XGBoost)). The robustness of the cx-Attentive FP model was also confirmed by evaluating the models with different training data sizes and dataset splitting strategies. Therefore, we developed a webserver and defined the applicability domain for this model. The webserver (http://tools.scbdd.com/chemlogd/) provides free log D7.4 prediction services. In addition, the important descriptors for log D7.4 were detected by the Shapley additive explanations (SHAP) method, and the most relevant substructures of log D7.4 were identified by the attention mechanism. Finally, the matched molecular pair analysis (MMPA) was performed to summarize the contributions of common chemical substituents to log D7.4, including a variety of hydrocarbon groups, halogen groups, heteroatoms, and polar groups. In conclusion, we believe that the cx-Attentive FP model can serve as a reliable tool to predict log D7.4 and hope that pretraining on low-fidelity data can help GNNs make accurate predictions of other endpoints in drug discovery.