Multi-Task Cascade Forest Framework for Predicting Acute Toxicity across Species

Unexpected toxicity has become a significant obstacle to drug candidate development, accounting for 30% of drug discovery failures. Traditional toxicity assessment through animal testing is costly and time‐consuming. Big data and artificial intelligence (AI), especially machine learning (ML), are robustly contributing to innovation and progress in toxicology research. However, the optimal AI model for different types of toxicity usually varies, making it essential to conduct comparative analyses of AI methods across toxicity domains. The diverse data sources also pose challenges for researchers focusing on specific toxicity studies. In this review, 10 categories of drug‐induced toxicity is examined, summarizing the characteristics and applicable ML models, including both predictive and interpretable algorithms, striking a balance between breadth and depth. Key databases and tools used in toxicity prediction are also highlighted, including toxicology, chemical, multi‐omics, and benchmark databases, organized by their focus and function to clarify their roles in drug‐induced toxicity prediction. Finally, strategies to turn challenges into opportunities are analyzed and discussed. This review may provide researchers with a valuable reference for understanding and utilizing the available resources to bridge prediction and mechanistic insights, and further advance the application of ML in drugs‐induced toxicity prediction.

Large-scale wind power grid connection increases the uncertainty of the power system, which reduces the economy and security of power system operations. Wind power prediction technology provides the wind power sequence for a period of time in the future, which provides key technical support for the reasonable development of the power generation plan and the arrangement of spare capacity. For large-scale wind farm groups, we propose a cluster model of wind power prediction based on multi-task learning, which can directly output the power prediction results of multiple wind farms. Firstly, the spatial and temporal feature matrix is constructed based on the meteorological forecast data provided by eight wind farms, and the dimensionality of each attribute is reduced by the principal component analysis algorithm to form the spatial fusion feature set. Then, a network structure with bidirectional gated cycle units is constructed, and a multi-output network structure is designed based on the Multi-gate Mixture-of-Experts (MMoE) framework to design the wind power group prediction model. Finally, the data provided by eight wind farms in Jilin, China, was used for experimental analysis, and the predicted average normalized root mean square error is 0.1754, meaning the prediction precision meets the scheduling requirement, which verifies the validity of the wind power prediction model.

<p><b>Dynamic flexible job shop scheduling (DFJSS) has received widespread attention from academia and industry due to its reflection in real-world scheduling applications such as order picking in the warehouse and the manufacturing industry. It requires complex routing and sequencing decisions under unpredicted dynamic events. Genetic programming, as ahyper-heuristic approach (GPHH), has been successfully applied to evolve scheduling heuristics for DFJSS automatically due to its flexible representation. Although GPHH has achieved certain success in solving the DFJSS problems, there are still some limitations for applying GPHH to DFJSS, particularly in terms of its training efficiency, large search space, search mechanism, and multitask solving ability. The overall goal of this thesis is to develop effective GPHH algorithmsto evolve scheduling heuristics for DFJSS efficiently. Different machine learning techniques, i.e., surrogate, feature selection, specialised genetic operator, and multitask learning, are incorporated in this thesis to tackle the limitations.</b></p> <p>First, this thesis develops a novel multi-fidelity based surrogate-assisted GPHH for DFJSS to improve the training efficiency of GPHH. Specifically, multi-fidelity based surrogate models are first designed by simplifying the problem to be solved. Then, an effective collaboration mechanism with knowledge transfer is proposed for utilising the advantages of the multifidelity based surrogate models to solve the problem. The results show that the proposed algorithm can dramatically reduce the computational cost of GPHH without sacrificing the performance in all the test scenarios. With the same training time, the proposed algorithm can achieve significantly better performance than its counterparts in most scenarios while no worse in others.</p> <p>Second, this thesis designs a novel two-stage GPHH framework with feature selection to evolve scheduling heuristics for DFJSS automatically. Based on this framework, this thesis further proposes to evolve scheduling heuristics with only the selected features by eliminating the unselected features properly. Specifically, individual adaptation strategies are proposed to generate individuals with only the selected features by utilising the information of both the selected features and the investigated individuals during the feature selection process. The results show that the proposed algorithm can successfully achieve scheduling heuristics with fewer unique features and smaller sizes, which tends to be more interpretable. In addition, the proposed algorithm can evolve comparable scheduling heuristic with that obtained by the traditional GPHH within a much shorter training time.</p> <p>Third, this thesis proposes a novel recombinative mechanism to provide guidance for GPHH based on the importance of subtrees to realise effective and adaptive recombination for parents to produce offspring. Two measures are proposed to measure the importance of all the subtrees of an individual. The first one is based on the frequency of features, and the second is based on the correlation between the behaviour of subtrees and the whole tree (i.e., an individual). The importance information is utilised to decide the crossover points for the parents. The proposed recombinative guidance mechanism attempts to improve the quality of offspring by preserving the promising building-blocks of one parent and incorporating good building-blocks from the other. The results show that the proposed algorithm based on the correlation importance measure performs better than the proposed algorithm based on the feature frequency importance measure. In addition, the proposed algorithm based on the correlation importance measure between the behaviour of subtrees significantly also outperforms the state-of-the-art algorithms on most tested scenarios.</p> <p>Last, this thesis proposes a multitask GPHH approach and a surrogate-assisted multitask GPHH approach to solving multiple DFJSS tasks simultaneously. First, an effective hyper-heuristic multitask algorithm is proposed by adapting the traditional evolutionary multitask algorithms based on the characteristics of GPHH. Second, this thesis develops a novel surrogate-assisted multitask GPHH approach to solving multiple DFJSS tasks by sharing useful knowledge between different DFJSS scheduling tasks. Specifically, the surrogate-assisted multitask GPHH algorithm employs the phenotypic characterisation technique to measure the behaviours of scheduling rules to build a surrogate for each task accordingly. The built surrogates are not only used to improve the efficiency of solving each single DFJSS task but also utilised for knowledge sharing between multiple DFJSS tasks in multitask learning. The results show that the proposed algorithm can significantly improve the quality of scheduling heuristics for all the test scenarios. The results also observe that the proposed algorithms manage to solve multiple tasks collaboratively in terms of the evolved scheduling heuristics for different tasks in a multitask scenario.</p>

MotivationMachine learning models for predicting cell-type-specific transcription factor (TF) binding sites have become increasingly more accurate thanks to the increased availability of next-generation sequencing data and more standardized model evaluation criteria. However, knowledge transfer from data-rich to data-limited TFs and cell types remains crucial for improving TF binding prediction models because available binding labels are highly skewed towards a small collection of TFs and cell types. Transfer prediction of TF binding sites can potentially benefit from a multitask learning approach; however, existing methods typically use shallow single-task models to generate low-resolution predictions. Here, we propose NetTIME, a multitask learning framework for predicting cell-type-specific TF binding sites with base-pair resolution.ResultsWe show that the multitask learning strategy for TF binding prediction is more efficient than the single-task approach due to the increased data availability. NetTIME trains high-dimensional embedding vectors to distinguish TF and cell-type identities. We show that this approach is critical for the success of the multitask learning strategy and allows our model to make accurate transfer predictions within and beyond the training panels of TFs and cell types. We additionally train a linear-chain conditional random field (CRF) to classify binding predictions and show that this CRF eliminates the need for setting a probability threshold and reduces classification noise. We compare our method’s predictive performance with two state-of-the-art methods, Catchitt and Leopard, and show that our method outperforms previous methods under both supervised and transfer learning settings.Availability and implementationNetTIME is freely available at https://github.com/ryi06/NetTIME and the code is also archived at https://doi.org/10.5281/zenodo.6994897.Supplementary informationSupplementary data are available at Bioinformatics online.

Inertial navigation systems (INSs) are crucial for autonomous vehicles, yet their accuracy is significantly compromised when using low-cost IMUs. This article proposes a novel multitask learning (MTL) framework, named MTNet, to enhance the accuracy and robustness of pure inertial navigation. Traditional approaches struggle with sensor noise, bias, and dynamic motion complexities and often rely on single-task optimization or simplistic sensor fusion. MTNet innovatively addresses these limitations by simultaneously tackling three interrelated tasks through tensor normal (TN) priors and optimizing feature extraction and knowledge transfer across tasks, improving training efficiency and generalization. Experiments on the KAIST dataset demonstrate that MTNet achieves approximately 25% lower relative trajectory error (RTE) and about 20% lower absolute trajectory error (ATE) compared to state-of-the-art methods. At the same time, we also demonstrate good performance in real-world experiments. The proposed approach highlights the potential of using MTL in addressing complex inertial navigation challenges, offering a robust solution for the localization of autonomous systems.

Recently, data-driven decision-making has attracted great interest; this requires high-quality datasets. However, real-world datasets often feature missing values for unknown or intentional reasons, rendering data-driven decision-making inaccurate. If a machine learning model is trained using incomplete datasets with missing values, the inferred results may be biased. In this case, a commonly used technique is the missing value imputation (MVI), which fills missing data with possible values estimated based on observed values. Various data imputation methods using machine learning, statistical inference, and relational database theories have been developed. Among them, conventional machine learning based imputation methods that handle tabular data can deal with only numerical columns or are time-consuming and cumbersome because they create an individualized predictive model for each column. Therefore, we have developed a novel imputational neural network that we term the Denoising Self-Attention Network (DSAN). Our proposed DSAN can deal with tabular datasets containing both numerical and categorical columns; it considers discretized numerical values as categorical values for embedding and self-attention layers. Furthermore, the DSAN learns robust feature expression vectors by combining self-attention and denoising techniques, and can predict multiple, appropriate substituted values simultaneously (via multi-task learning). To verify the validity of the method, we performed data imputation experiments after arbitrarily generating missing values for several real-world tabular datasets. We evaluated both imputational and downstream task performances, and we have seen that the DSAN outperformed the other models, especially in terms of category variable imputation.

Toxicity assessment is an essential step in the development of drugs, agrochemicals and other bioactive substances. In silico toxicity prediction, or computational toxicology, has been growing fast during the last years mainly due to advantageous cost and labor reduction, and avoidance of experiments using animals. Considering the vast number of software available for toxicity prediction, one should be aware to pick up the right software for the right task. Herein, we aim to describe the main software as well as corresponding methodologies behind them, that are able to provide reliable and accurate results for diverse toxicological endpoints. We give special emphasis to software based on quantitative structure–activity relationships, expert systems and machine learning methodologies. With this in mind, we hope that this chapter content may serve as a valuable guide to select toxicity prediction software, and also to reinforce their usefulness towards researches concerning the development of bioactive substances.

Propolis, has emerged as a promising candidate in this regard due to its active compounds, such as quercetin and caffeic acid phenethyl ester (CAPE).7,8 These compounds exhibit anti-inflammatory, making propolis a potential alternative to conventional nonsteroidal anti-inflammatory drugs (NSAID). This study evaluates the anti-inflammatory potential of quercetin and CAPE, compounds from propolis, through in silico analysis focusing on cyclooxygenase-2 (COX-2) inhibition. Toxicity predictions, molecular docking, and molecular dynamics simulations were used to assess their safety and efficacy. The COX-2 structure (PDB ID: 4PH9) was retrieved from the Protein Data Bank, and ligands (quercetin, CAPE, ibuprofen, aspirin, and vanillin) were prepared for docking. Toxicity assessments were conducted using PROTOX II and pkCSM, including AMES, LD50, skin sensitization, and hepatotoxicity tests. Molecular docking was performed using Dock 6.8, and protein-ligand interactions were validated by RMSD. Molecular dynamics simulations were carried out with Amber22, including energy minimization, solvation, and stability evaluation via RMSD, RMSF, and MM-GBSA calculations. Toxicity predictions indicated that quercetin and CAPE are safe, with negative results in mutagenicity, skin sensitization, and hepatotoxicity tests, and LD50 values supporting their therapeutic potential. Molecular docking showed strong binding of quercetin (-51.4 kcal/mol) and CAPE (-53.3 kcal/mol) to COX-2, outperforming ibuprofen and aspirin. Molecular dynamics simulations revealed stable protein-ligand interactions, with low RMSD (0.3–0.4 Å) and RMSF values, indicating high stability. MM-GBSA calculations confirmed the strong binding of quercetin (-35.4 kcal/mol) and CAPE (-33.0 kcal/mol). Quercetin and CAPE exhibit strong COX-2 binding and stability, alongside favorable safety profiles, supporting their potential as anti-inflammatory agents for further clinical investigation.

Knowledge transfer aims to apply existing knowledge to different tasks or new data, and it has extensive applications in multi-domain and Multi-Task Learning. The key to this task is quickly identifying a fine-grained object for knowledge sharing and efficiently transferring knowledge. Current methods, such as fine-tuning, layer-wise parameter sharing, and task-specific adapters, only offer coarse-grained sharing solutions and struggle to effectively search for shared parameters, thus hindering the performance and efficiency of knowledge transfer. To address these issues, we propose Channel-Wise Parameter Sharing (CWPS), a novel fine-grained parameter-sharing method for knowledge transfer, which is efficient for parameter sharing, comprehensive, and plug-and-play. For the coarse-grained problem, we first achieve fine-grained parameter sharing by refining the granularity of shared parameters from the level of layers to the level of neurons. The knowledge learned from previous tasks can be utilized through the explicit composition of the model neurons. Besides, we promote an effective search strategy to minimize computational costs, simplifying the selection of shared weights. In addition, our CWPS has strong composability and generalization ability, which theoretically can be applied to any network consisting of linear and convolution layers. We introduce several datasets in both Incremental Learning and Multi-Task Learning scenarios. Our method has achieved state-of-the-art precision-to-parameter ratio performance with various backbones, demonstrating its efficiency and versatility.

Despite therapeutic advancements, cervical cancer caused by high-risk subtypes of the human papillomavirus (HPV) remains a leading cause of cancer-related deaths among women worldwide. This study aimed to discover potential drug candidates from the Asian medicinal plant Andrographis paniculata, demonstrating efficacy against the E6 protein of high-risk HPV-16 subtype through an in-silico computational approach. The 3D structures of 32 compounds (selected from 42) derived from A. paniculata, exhibiting higher binding affinity, were obtained from the PubChem database. These structures underwent subsequent analysis and screening based on criteria including binding energy, molecular docking, drug likeness and toxicity prediction using computational techniques. Considering the spectrometry, pharmacokinetic properties, docking results, drug likeliness, and toxicological effects, five compounds—stigmasterol, 1H-Indole-3-carboxylic acid, 5-methoxy-, methyl ester (AP7), andrographolide, apigenin and wogonin—were selected as the potential inhibitors against the E6 protein of HPV-16. We also performed 200 ns molecular dynamics simulations of the compounds to analyze their stability and interactions as protein–ligand complexes using imiquimod (CID-57469) as a control. Screened compounds showed favorable characteristics, including stable root mean square deviation values, minimal root mean square fluctuations and consistent radius of gyration values. Intermolecular interactions, such as hydrogen bonds and hydrophobic contacts, were sustained throughout the simulations. The compounds displayed potential affinity, as indicated by negative binding free energy values. Overall, findings of this study suggest that the selected compounds have the potential to act as inhibitors against the E6 protein of HPV-16, offering promising prospects for the treatment and management of CC.

Network pharmacology and molecular docking are widely applied in drug discovery, yet fragmented workflows and inconsistent operations frequently undermine result reproducibility. Here, a standardized protocol integrating these approaches into a reproducible framework for drug screening and mechanism exploration is described, with the workflow organized into three sequential phases: data preparation, computational analysis, and validation. In the preparation phase, compound libraries from public databases are filtered via Absorption, Distribution, Metabolism, Excretion, and Toxicity (ADMET) criteria, including oral bioavailability, drug-likeness, and toxicity prediction, while potential therapeutic targets are obtained through target prediction and integration of disease-related databases to comprehensively identify drug-disease interaction candidates. In the computational analysis phase, overlapping targets undergo Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses and protein-protein interaction network topological analysis to identify core targets; molecular docking is configured with two standardized optional strategies with distinct advantages. The two-step progressive strategy employs AutoDock Vina for high-throughput preliminary screening of the compound library, followed by precise re-docking with YASARA, which eliminates false positives from high-throughput screening and generates protein-ligand complexes natively compatible with subsequent YASARA molecular dynamics (MD) simulations to avoid structural deviations caused by cross-software format conversion. The one-step strategy completes the full docking process via YASARA alone, which simplifies the operation workflow, improves experimental efficiency, and is fully applicable for specific research goals. In the validation phase, standardized MD simulations evaluate ligand-protein complex stability via core metrics of root mean square deviation (RMSD) and root mean square fluctuation (RMSF). This unified, reproducible pipeline enhances the reliability of network pharmacology and docking studies and facilitates cross-study comparisons in computational drug discovery.

The emergence of 2019 novel Coronavirus (COVID-19 or 2019-nCoV) has caused significant global morbidity and mortality with no consensus specific treatment. We tested the hypothesis that FDA-approved antiretrovirals, antibiotics, and antimalarials will effectively inhibit COVID-19 two major drug targets, coronavirus nucleocapsid protein (NP) and hemagglutinin-esterase (HE). To test this hypothesis, we carried out a phylogenic analysis of coronavirus genome to understand the origins of NP and HE, and also modeled the proteins before molecular docking, druglikeness, toxicity assessment, molecular dynamics simulation (MDS) and ligand-based pharmacophore modeling of the selected FDA-approved drugs. Our models for NP and HE had over 95% identity with templates 5EPW and 3CL5 respectively in the PDB database, with majority of the amino acids occupying acceptable regions. The active sites of the proteins contained conserved residues that were involved in ligand binding. Lopinavir and ritonavir possessed greater binding affinities for NP and HE relative to remdesivir, while levofloxacin and hydroxychloroquine were the most notable among the other classes of drugs. The Root Mean Square Deviation (RMSD), Root Mean Square Fluctuation (RMSF), Radius of gyration (Rg), and binding energy values obtained after 100 ns of MDS revealed good stability of these compounds in the binding sites of the proteins while important pharmacophore features were also identified. The study showed that COVID-19 likely originated from bat, owing to the over 90% genomic similarity observed, and that lopinavir, levofloxacin, and hydroxychloroquine might serve as potential anti-COVID-19 lead molecules for additional optimization and drug development for the treatment of COVID-19. Communicated by Ramaswamy H. Sarma

In order to effectively improve the accuracy of Consumer Price Index (CPI) prediction so as to more truly reflect the overall level of the country’s macroeconomic situation, a CPI big data prediction method based on wavelet twin support vector machine (SVM) is proposed. First, the historical CPI data are decomposed into high-frequency part and low-frequency part by wavelet transform. Then a more advanced twin SVM is used to build a prediction model to obtain two kinds of prediction results. Finally, the wavelet reconstruction method is used to fuse the two kinds of prediction results to obtain the final CPI prediction results. The wavelet twin SVM model is used to fit and predict CPI index. Experimental results show that compared with the similar prediction methods, the proposed prediction method has higher fitting accuracy and smaller root mean square error.

In the real-time tumor-tracking radiotherapy system, fluoroscopy is used to determine the real-time position of internal fiducial markers. The pattern recognition score (PRS) ranging from 0 to 100 is computed by a template pattern matching technique in order to determine the marker position on the fluoroscopic image. The PRS depends on the quality of the fluoroscopic image. However, the fluoroscopy parameters such as tube voltage, current and exposure duration are selected manually and empirically in the clinical situation. This may result in an unnecessary imaging dose from the fluoroscopy or loss of the marker because of too much or insufficient x-ray exposure. In this study, a novel optimization method is proposed in order to minimize the fluoroscopic dose while keeping the image quality usable for marker tracking. The PRS can be predicted in a region where the marker appears to move in the fluoroscopic image by the proposed method. The predicted PRS can be utilized to judge whether the marker can be tracked with accuracy. In this paper, experiments were performed to show the feasibility of the PRS prediction method under various conditions. The predicted PRS showed good agreement with the measured PRS. The root mean square error between the predicted PRS and the measured PRS was within 1.44. An experiment using a motion controller and an anthropomorphic chest phantom was also performed in order to imitate a clinical fluoroscopy situation. The result shows that the proposed prediction method is expected to be applicable in a real clinical situation.

Statistical models have been successful in accurately estimating the biochemical contents of vegetation from the reflectance spectra. However, their performance deteriorates when there is a scarcity of sizable amount of ground truth data for modeling the complex non-linear relationship occurring between the spectrum and the biochemical quantity. We propose a novel Gaussian process based multitask learning method for improving the prediction of a biochemical through the transfer of knowledge from the learned models for predicting related biochemicals. This method is most advantageous when there are few ground truth data for the biochemical of interest, but plenty of ground truth data for related biochemicals. The proposed multitask Gaussian process hypothesizes that the inter-relationship between the biochemical quantities is better modeled by using a combination of two or more covariance functions and inter-task correlation matrices. In the experiments, our method outperformed the current methods on two real-world datasets.