Volume 22, Issue 2 - Special Issue: Selected extended papers of the 18th International Conference on Practical Applications of Computational Biology and Bioinformatics (PACBB); Guest Editor: Florentino Fernández Riverola

Plants produce specialized metabolites, which play critical roles in defending against biotic and abiotic stresses. Due to their chemical diversity and bioactivity, these compounds have significant economic implications, particularly in the pharmaceutical and agrotechnology sectors. Despite their importance, the biosynthetic pathways of these metabolites remain largely unresolved. Automating the prediction of their precursors, derived from primary metabolism, is essential for accelerating pathway discovery. Using DeepMol’s automated machine learning engine, we found that regularized linear classifiers offer optimal, accurate, and interpretable models for this task, outperforming state-of-the-art models while providing chemical insights into their predictions. The pipeline and models are available at the repository: https://github.com/jcapels/SMPrecursorPredictor.

Family structures play a crucial role in personal development, social dynamics, and mental health. Traditional systems for encoding genealogical data, such as Ahnentafel and the Register System, offer methods to document lineage but face limitations, particularly in accommodating horizontal relationships or handling changes in family datasets. Modern computational systems like LINKAGE and PED, while powerful for genetic analysis, lack human readability and are challenging to apply in fields where unstructured, narrative data is common, such as sociology or psychiatry. This paper aims to bridge this gap by enhancing Fcodes, a flexible and intuitive algorithm for encoding kinship relationships that is suited for both manual and computational use. Building on our previous work, we present improvements to the Fcodes core algorithm and command-line interface (CLI), as well as the development of F-Tree, a new graphical user interface (GUI) to streamline the encoding process. Additionally, we introduce a method for estimating the coefficient of inbreeding using Fcodes and explore the application of artificial intelligence, namely large language models (LLMs), to automatically infer family relationships from narrative text. These advancements highlight the potential of Fcodes in a wide range of research contexts, from social studies to genetics and mental health research.

The characterization of protein-protein interactions (PPIs) is fundamental to understand cellular functions. Although machine learning methods in this task have historically reported prediction accuracies up to 95 %, including those only using raw protein sequences, it has been highlighted that this could be overestimated due to the use of random splits and metrics that do not take into account potential biases in the datasets. Here, we propose a per-protein utility metric, pp_MCC, able to show a drop in the performance in both random and unseen-protein splits scenarios. We tested ML models based on sequence embeddings. The pp_MCC metric evidences a reduced performance even in a random split, reaching levels similar to those shown by the raw MCC metric computed over an unseen protein split, and drops even further when the pp_MCC is used in an unseen protein split scenario. Thus, the metric is able to give a more realistic performance estimation while allowing to use random splits, which could be interesting for more protein-centric studies. Given the low adjusted performance obtained, there seems to be room for improvement when using only primary sequence information, suggesting the need of inclusion of complementary protein data, accompanied with the use of the pp_MCC metric.

The rapid advancement of Next-Generation Sequencing (NGS) technologies has revolutionized the field of genomics, producing large volumes of data that necessitate sophisticated analytical techniques. This paper introduces a Deep Learning model designed to predict the pathogenicity of genetic variants, a vital component in advancing personalized medicine. The model is trained on a dataset derived from the analysis of NGS outputs, containing a combination of well-defined and ambiguous genetic variants. By employing a semi-supervised learning approach, the model efficiently utilizes both confidently labeled and less certain data. At the core of the methodology is the Feature Tokenizer Transformer architecture, which processes both numerical and categorical genomic information. The preprocessing pipeline includes key steps such as data imputation, scaling, and encoding to ensure high data quality. The results highlight the model’s impressive accuracy, particularly in detecting confidently labeled variants, while also addressing the impact of its predictions on less certain (soft-labeled) data.

This paper investigates the application of Survival Analysis (SA) techniques to forecast outcomes after autologous Hematopoietic Stem Cell Transplantation (aHSCT) for Multiple Myeloma (MM). By leveraging six SA models, we examine their predictive capabilities, measured through the Concordance Index (C-index) metric. Beyond evaluating model performance, we analyze feature importance using permutation and SHAP methods, highlighting key clinical factors such as treatment history, disease stage, and prior disease progression or relapse as critical predictors of survival. The findings suggest that while all models performed well based on the C-index, a detailed examination revealed variations in how each model processed data. Specifically, the Coxnet and Random Survival Forest models exhibited a more thorough use of clinical variables, whereas the gradient boosting models appeared to rely on a narrower range of features, potentially limiting their ability to differentiate between patients with comparable profiles. Risk predictions categorized patients into low, moderate, and high-risk levels. For lower-risk patients, the procedure showed positive outcomes, while higher-risk individuals were predicted to have limited survival benefits, recommending alternative treatments. Lastly, we propose future research to expand these models into time-to-event estimations, offering additional support for decision-making by predicting patient life expectancy post-transplant, considering their pre-transplant clinical attributes.

Predicting Drug-Drug interaction (DDI)-induced adverse drug reactions (ADRs) using computational methods is challenging due to the availability of limited data samples, data sparsity, and high dimensionality. The issue of class imbalance further increases the intricacy of prediction. Different computational techniques have been presented for predicting DDI-induced ADRs in the general population; however, the area of DDI-induced pregnancy and neonatal ADRs has been underexplored. In the present work, a sparse ensemble-based computational approach is proposed that leverages SMILES strings as features, addresses high-dimensional and sparse data using Sparse Principal Component Analysis (SPCA), mitigates class imbalance with the Multilabel Synthetic Minority Oversampling Technique (MLSMOTE), and predicts pregnancy and neonatal ADRs through a stacking ensemble model. The SPCA has been evaluated for handling sparse data and shown 2.67 %–5.45 % improvement compared to PCA. The proposed stacking ensemble model has outperformed six state-of-the-art predictors regarding micro and macro scores for True Positive Rate ( TPR ), F1 Score, False Positive Rate ( FPR ), Precision, Hamming Loss, and ROC-AUC Score with 1.16 %–14.94 %.

Left ventricular non-compaction is a cardiac condition marked by excessive trabeculae in the left ventricle’s inner wall. Although various methods exist to measure these structures, the medical community still lacks consensus on the best approach. Previously, we developed DL-LVTQ, a tool based on a UNet neural network, to quantify trabeculae in this region. In this study, we expand the dataset to include new patients with Titin cardiomyopathy and healthy individuals with fewer trabeculae, requiring retraining of our models to enhance predictions. We also propose ViTUNeT, a neural network architecture combining U-Net and Vision Transformers to segment the left ventricle more accurately. Additionally, we train a YOLOv8 model to detect the ventricle and integrate it with ViTUNeT model to focus on the region of interest. Results from ViTUNet and YOLOv8 are similar to DL-LVTQ, suggesting dataset quality limits further accuracy improvements. To test this, we analyze MRI images and develop a method using two YOLOv8 models to identify and remove problematic images, leading to better results. Combining YOLOv8 with deep learning networks offers a promising approach for improving cardiac image analysis and segmentation.

Accurate mitosis detection is essential for cancer diagnosis and treatment. Traditional manual counting by pathologists is time-consuming and may cause errors. This research investigates automated mitosis detection in stained histopathological images using Deep Learning (DL) techniques, particularly object detection models. We propose a two-stage object detection model based on Faster R-CNN to effectively detect mitosis within histopathological images. The stain augmentation and normalization techniques are also applied to address the significant challenge of domain shift in histopathological image analysis. The experiments are conducted using the MIDOG++ dataset, the most recent dataset from the MIDOG challenge. This research builds on our previous work, in which two one-stage frameworks, in particular on RetinaNet using fastai and PyTorch, are proposed. Our results indicate favorable F1-scores across various scenarios and tumor types, demonstrating the effectiveness of the object detection models. In addition, Faster R-CNN with stain techniques provides the most accurate and reliable mitosis detection, while RetinaNet models exhibit faster performance. Our results highlight the importance of handling domain shifts and the number of mitotic figures for robust diagnostic tools.

This study investigates the application of deep learning techniques for segmenting glands in histopathological images of colorectal cancer. We trained two convolutional neural network models, U-Net and DCAN, on a combination of the GlaS and CRAG datasets to enhance generalization across diverse histological appearances, selecting DCAN for its superior accuracy in delineating gland boundaries. The goal was to achieve robust gland segmentation applicable to whole slide images (WSIs) from The Cancer Genome Atlas (TCGA). Using the segmented glands, we extracted patient-level morphological features and used them to predict survival outcomes. A Cox proportional hazards model was trained on these features and achieved a high concordance index, indicating strong predictive performance. Patients were then stratified into high- and low-risk groups, with significant differences in survival distributions (log-rank p -value: 0.01317). In addition, we benchmarked our models against state-of-the-art gland segmentation methods on GlaS and CRAG, highlighting the trade-off between domain-specific accuracy and cross-dataset robustness.

This study explores the use of deep learning to analyze genetic data and predict phenotypic traits associated with schizophrenia, a complex psychiatric disorder with a strong hereditary component yet incomplete genetic characterization. We applied Convolutional Neural Networks models to a large-scale case-control exome sequencing dataset from the Swedish population to identify genetic patterns linked to schizophrenia. To enhance model performance and reduce overfitting, we employed advanced optimization techniques, including dropout layers, learning rate scheduling, batch normalization, and early stopping. Following systematic refinements in data preprocessing, model architecture, and hyperparameter tuning, the final model achieved an accuracy of 80 %. These results demonstrate the potential of deep learning approaches to uncover intricate genotype-phenotype relationships and support their future integration into precision medicine and genetic diagnostics for psychiatric disorders such as schizophrenia.