Abstract Aim The bioconcentration factor (BCF) indicates the potential of chemicals to accumulate in organisms. Greater BCFs imply higher bioaccumulation and health risks [1]. To date, BCF prediction models based on molecular structure have mainly been evaluated for accuracy [2,3], with limited systematic assessment of their reliability or identification of outliers. In this study, we aimed to improve model practicality. Methods We developed a regression model based on physicochemical descriptors from SMILES strings and compared predicted BCF values with experimental data. We then constructed a classifier to identify chemicals with large prediction deviations (overestimation and underestimation groups) and analyzed the contribution of individual features. Results In the underestimation group, electrical properties (qnmax), hydrophobicity (LogP), and the presence or absence of specific functional groups strongly influenced predictions. In contrast, no interpretable patterns in important molecular descriptors were identified in the overestimated group. Conclusion These findings provide insights into the chemicals for which BCF predictions should be interpreted with caution and may contribute to more efficient environmental risk assessments. References 1. Arnot J.A., Gobas F.A.P.C. ‘A review of bioconcentration factor (BCF) and bioaccumulation factor (BAF) assessments for organic chemicals in aquatic organisms.’ Environmental Reviews 2006:14(4). 2. Kobayashi Y., Yoshida K. ‘Development of QSAR models for prediction of fish bioconcentration factors using physicochemical properties and molecular descriptors with machine learning algorithms.’ Ecological Informatics 2021: 63:101285. 3. Pore S. et al. ‘Machine learning-based q-RASAR predictions of the bioconcentration factor of organic molecules estimated following the organisation for economic co-operation and development guideline 305.’ Journal of Hazardous Materials 2024: 479:135725.

Abstract Aim Water and fishery resources are essential for human life and regional economies. However, marine ecosystems face increasing pressures from human activities such as urbanization, agriculture, and climate change. Although the effects of rising sea temperatures on fish growth have been well documented [1], the impacts of chemical pollutants, including pesticides and heavy metals, remain less understood. This study aimed to identify environmental pollutants and oceanographic factors influencing Atlantic cod (Gadus morhua) growth using interpretable machine-learning methods. Methods Comprehensive datasets of environmental pollutants and oceanographic parameters from the North Atlantic were compiled. Random forest regression combined with SHAP (SHapley Additive exPlanations) was used to quantify variable importance. Results Zinc (Zn), lead (Pb), and chromium (Cr) exhibited significant negative contributions to predicted body length, suggesting potential growth inhibition associated with these metals. In contrast, mercury (Hg) showed a positive association with body size, suggesting bioaccumulation consistent with previous findings. Furthermore, dissolved oxygen, salinity, and temperature were also identified as important predictors. Conclusion These findings indicate that both chemical pollutants and oceanographic parameters influence cod growth. Recognizing the importance of these two qualitatively different types of environmental factors can contribute to more comprehensive approaches in ecosystem-based fisheries management and in the environmental risk assessment and management of marine pollutants. References [1] Pörtner H.O., Knust R. ‘Oxygen- and capacity-limited thermal tolerance: Linking ecology and physiology.’ Annual Review of Physiology 2010; 72:61–90.

Abstract Aim The bioconcentration factor (BCF) is a key parameter for evaluating the bioaccumulation potential of chemicals. However, experimental measurement of BCF is time-consuming, costly, and ethically challenging due to animal testing. Many existing computational models prioritize predictive accuracy over interpretability. Yet, clarifying the molecular features that govern bioaccumulation is increasingly important.1 This study aimed to develop an interpretable regression model for BCF prediction based on physicochemical descriptors. Methods Molecular descriptors were computed from SMILES. Several machine learning algorithms were benchmarked, including random forest, and XGBoost. SHapley Additive exPlanations (SHAP) values were used for feature selection and model interpretation, identifying key physicochemical and structural contributors to BCF. Results The optimized Gradient Boosting Decision Trees (GBDT) model achieved an R2 of 0.57 with 10 selected features and outperformed random forest and linear models. SHAP analysis revealed that Topological Polar Surface Area (TPSA), hydrophobicity (log P), and molecular volume were the most influential descriptors. Higher log P and molecular volume increased BCF, whereas greater TPSA was associated with lower bioaccumulation. Conclusions This study highlights an interpretable machine learning framework for predicting BCF based on physicochemical descriptors. By linking molecular features with bioaccumulation potential, the model supports transparent and reliable in silico assessment, contributing to ethical and cost-effective chemical risk evaluation. References 1. OECD, ‘(Q)SAR Assessment Framework: Guidance for the regulatory assessment of (Quantitative) Structure Activity Relationship models and predictions - Second edition, OECD Series on Testing and Assessment No. 405,’ OECD Publishing 2024.

MOTIVATION: Disease states are distinguished from each other in terms of differing clinical phenotypes, but characteristic molecular features are often common to various diseases. Similarities between diseases can be explained by characteristic gene expression patterns. However, most disease-disease relationships remain uncharacterized. RESULTS: In this study, we proposed a novel approach for network-based characterization of disease-disease relationships in terms of drugs and therapeutic targets. We performed large-scale analyses of omics data and molecular interaction networks for 79 diseases, including adrenoleukodystrophy, leukaemia, Alzheimer's disease, asthma, atopic dermatitis, breast cancer, cystic fibrosis and inflammatory bowel disease. We quantified disease-disease similarities based on proximities of abnormally expressed genes in various molecular networks, and showed that similarities between diseases could be explained by characteristic molecular network topologies. Furthermore, we developed a kernel matrix regression algorithm to predict the commonalities of drugs and therapeutic targets among diseases. Our comprehensive prediction strategy indicated many new associations among phenotypically diverse diseases. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.