Abstract Synthetic lethality (SL) provides a powerful foundation for targeted oncology by identifying gene pairs whose simultaneous disruption is fatal to cancer cells while sparing healthy tissue. However, mapping these interactions experimentally across the human genome is restricted by massive combinatorial scale and high costs. To address this, computational approaches specifically graph neural networks (GNNs) have become the standard for modeling these complex biological relationships. Hence, by integrating heterogeneous knowledge graph (KG) and multi-omics data, GNNs can effectively capture the topological and molecular context required for accurate SL prediction. This review systematically evaluates the computational SL landscape across three core dimensions. First, we examine how diverse GNN architectures integrate multi-layered biological data. Second, we address model interpretability, contrasting intrinsically transparent architectures with post-hoc explanation methods, as a requirement for clinical trust. Finally, we explore the emerging role of Generative AI and large language models to leverage their zero-shot reasoning and broad biomedical knowledge.
Abstract Protein lysine acylation regulates protein function through changes in local chemistry, structure, and molecular interactions. Acetylation (Kac), succinylation (Ksucc), and crotonylation (Kcr) share the lysine centre but differ in acyl-group chemistry, motivating joint yet task-aware prediction. We developed KAT-Net, a multi-task framework integrating ESM-2 representations, physicochemical priors, and AlphaFold2-derived structural context. Predicted aligned error–aware graph modelling reduces uncertain structural relationships, while task-driven hyperbolic routing retrieves modification-specific information from shared representations. A masked asymmetric focal loss supports learning with unavailable labels and task-dependent class imbalance. Independent benchmark and external Kcr evaluations showed favourable performance, while ablation and single-task comparisons supported contributions from sequence–structure integration, structural modelling, hyperbolic routing, and multi-task learning. Sequence and structural analyses revealed modification-related positional, residue, and spatial patterns around candidate lysines. External functional analysis using an independent mouse lung acetylome and succinylome provided complementary evidence of functional plausibility, with Kac predictions showing partial concordance with experimentally reported metabolic processes rather than serving as a binary test of general Kac-site discrimination. For Kac prediction, model development and primary benchmarking were performed using the curated non-histone acetylation dataset from TransPTM and should be interpreted within this benchmark scope rather than as a comprehensive evaluation of the complete acetylome. KAT-Net therefore provides a unified framework for lysine-acylation prediction and candidate-site prioritization within the evaluated benchmark settings.
Abstract Bioinformatics plays a pivotal role in advancing healthcare, agriculture, and biotechnology. However, in Thailand, expectations regarding workforce demand and the core competencies of bioinformatics graduates remain insufficiently defined. Survey findings reveal that the demand for bioinformatics graduates is predominantly concentrated in the public sector, with comparatively limited demand from the private sector. Furthermore, there are differences between the public and private sectors in terms of the skill sets expected of bioinformaticians. To address these gaps, the Bioinformatics Association in Thailand, supported by Hub of Talents for Thailand Bioinformatics Research Network, National Research Council of Thailand, has developed a national competency framework for bioinformatics. This framework is intended to standardize graduate development and align educational outcomes with labor market needs. It encompasses four core domains: knowledge, skills, ethics, and character. This framework is informed by the Ministry of Higher Education, Science, Research and Innovation, together with expert consultations and public hearings involving key stakeholders. The proposed framework provides a structured foundation for cultivating a workforce equipped with both theoretical expertise and practical capabilities. Its implementation is expected to strengthen the national bioinformatics ecosystem and enhance the global competitiveness of Thai bioinformatics graduates.
Abstract The interpretation of genetic variants remains central to precision medicine, and machine-learning-based variant effect prediction tools have become indispensable for supporting decision-making processes. In this review we trace the progression from hand-crafted features and traditional machine learning classifiers, through supervised ensembles, to deep-learning architectures and foundation models that learn representations directly from biological sequences. We classify methods for genetic variant effect prediction by underlying machine learning model and variant type handled, from missense to non-coding and structural, as well as across feature categories, describing their biological rationale, adoption rate, and known limitations. We further examine three cross-cutting methodological challenges—feature orthogonality, data circularity, and training data bias—that affect reliability and clinical applicability, providing a critical overview of the current landscape and the challenges that lie ahead.
Abstract Large-scale population sequencing requires reference genomes that accurately represent the genetic background of target populations. Although high-quality linear references such as T2T-CHM13 have improved the completeness of human genome representation, a single reference genome remains insufficient to capture genetic diversity across populations. Pangenome graphs provide a richer representation of multi-source genomic variation, but translating pangenome diversity into practical reference genomes remains challenging. First, the direct application of pangenome graphs in routine large-scale sequencing analysis is limited by graph complexity and poor compatibility with existing linear-reference workflows. Second, existing pangenome-driven personalized references mainly target individual samples rather than populations, and current path recommendation strategies may overlook local source consistency during branch-node selection. In this study, we propose PGpath, a pangenome-based method for recommending population-adapted linear reference genomes. PGpath generates a reference from a selected pangenome subpath, preserving pangenome diversity while retaining a linear format compatible with standard downstream pipelines. In this study, we characterize the shared genetic background of samples using population-level k-mer frequency features and reformulate reference recommendation as branch-node prediction in the pangenome graph. To encourage local source consistency, we develop a topology-constrained multi-task neural network incorporating source-consistency relationships between adjacent branch nodes. Experiments on Human Pangenome Reference Consortium (HPRC) population datasets and HG002 variant detection benchmarks showed that PGpath-derived references improved read mapping performance and reduced downstream variant detection errors compared with commonly used references. PGpath provides a practical strategy for translating pangenome diversity into population-adapted linear references for large-scale genomic analysis.
