The development of novel therapeutic approaches to cancers requires significant time and labour, especially for uncommon cancers (e.g. of mesenchymal origin). In addition, these rare cancers typically receive less research attention but still need sufficient analysis. For example, a trustworthy analysis of tissue samples to identify cancer progression and cell response to candidate therapies is a task to develop certain treatments and drugs. Thus, this research work develops a new explainable Deep Learning pipeline for accurate, high-throughput and autonomous recognition of metastasising cells from large-scale fluorescence microscopy images, which can be used to evaluate cell response with low cost and high trustworthiness, and facilitate the therapeutic approaches development. The pipeline includes pre-trained and domain-generalised segmentation to identify single cells, training the latest self-supervised transformers: DINO-based models to provide normal or metastasising classification, and leveraging Explainable AI methods to produce biologically meaningful model explanations towards the trustworthiness of model predictions; this is crucial in future clinical applications. Compared to supervised learning-based models, DINO achieves the best predictability (over 97
The marked heterogeneity of cancer poses a substantial challenge to precision drug therapy, resulting in considerable variability in patient responses to identical treatments. Accurately predicting drug sensitivity thus represents an urgent and critical challenge in the field of pharmacology and personalized medicine. Existing methods have limitations in addressing the complexity of multi-modal biological data and heterogeneous graphs. Additionally, their heavy reliance on labeled data hinders model generalization. In this work, we propose the Heterogeneous Graph Attention Dual-Perturbation Contrastive Learning Network for Drug Response Prediction (HGACL-DRP). This framework constructs an optimized heterogeneous graph structure by integrating multi-omics features and implementing a robust neighbor filtering mechanism. HGACL-DRP innovatively introduces a Dual-Perturbation contrastive learning paradigm that incorporates both feature and structural perturbations specifically designed for drug response prediction, thereby enabling the self-supervised learning of more robust heterogeneous graph node embeddings. On two major public datasets, the Genomics of Drug Sensitivity in Cancer (GDSC) and the Cancer Cell Line Encyclopedia (CCLE), HGACL-DRP exhibits exceptional performance across multiple classification metrics, achieving a mean AUC of 95.48
Accurate clinical genetic variant classification is becoming increasingly challenging as data volume and knowledge complexity increase. Large Language Models (LLMs) could potentially automate evidence retrieval and assist variant interpretation. To systematically evaluate the capacity of general-purpose LLMs (Claude, ChatGPT, Gemini and Perplexity) to support clinical genetic variant classification. We analyzed 475 variants across 185 genes, with reference classifications established by expert variant scientists. Classification was performed by the four LLMs using a detailed structured prompting strategy and compared with a commercial non-LLM specialized platform. Performance was evaluated at the classification and criteria levels and stratified by various variant characteristics. Detailed prompting generally improved performance. Mean LLM classification concordance was 69.0
One of the most important unresolved issues in computational biology is systematic protein function annotation. Despite decades of curation by the Gene Ontology Consortium and UniProt Swiss-Prot, most sequenced proteins lack experimentally validated functional annotations. The emergence of Large Language Models (LLMs) trained on scientific literature raises the question of whether they can map natural-language descriptions of experimentally characterized protein function to the correct Gene Ontology (GO) terms without task-specific training. Answering this question requires a benchmark with gold-standard experimental annotations and multi-organism coverage, which existing benchmarks lack. We introduce PROBE (PRotein functiOn Benchmark Evaluation): 1,000 UniProt Swiss-Prot proteins from five organisms, experimental-evidence-only annotations. We evaluate 10 open-source LLMs across five prompt formats, zero-shot and inference-only, yielding 150,000 predictions. The best model (Mistral Large 123B) achieves macro-F1 of 9.3
Gut microbiome dysbiosis has emerged as a non-genetic risk factor in breast cancer (BC), yet the taxonomic, predicted functional, and microbial association patterns of the gut microbiome in BC remain incompletely understood. This study aimed to provide a combined taxonomic, predicted functional, and network-based characterization of the gut microbiome in BC using publicly available 16 S rRNA sequencing data. Publicly available 16 S rRNA V4 sequencing data from PRJNA658160 were analyzed with 804 fecal samples, including 386 BC samples and 418 healthy controls. Taxonomic profiling was performed, and predicted pathway profiling was conducted with PICRUSt2. Diversity, differential abundance, and network analyses were performed at both the genus and predicted pathway levels. Taxonomic differential abundance was evaluated using ANCOM-BC2, pathway differences using LinDA, taxonomic networks using SPIEC-EASI, and predicted pathway association networks using Spearman correlation. BC samples exhibited significantly reduced taxonomic alpha diversity, whereas genus-level beta diversity showed no strong global separation between groups. 13 genera were differentially abundant; Bacteroides, Erysipelatoclostridium, Parabacteroides, and Fusobacterium were enriched in BC, whereas Prevotella, Alloprevotella, and Coprococcus were more abundant in healthy controls. Exploratory network analysis identified Bacteroides as a prominent hub-like taxon and Erysipelatoclostridium as a bridge-like taxon in the BC inferred association network. Predicted functional analysis indicated reduced pathway alpha diversity and identified 130 differentially abundant predicted pathways. BC-enriched predicted pathways were mainly associated with carbohydrate degradation, the glyoxylate cycle, fatty acid β-oxidation, and stress-related metabolism, whereas control-enriched predicted pathways were associated with menaquinone biosynthesis, biotin biosynthesis, fermentation, and broader biosynthetic functions. Predicted pathway beta diversity also differed significantly between groups (PERMANOVA R² = 0.013, p = 0.001); however, beta-dispersion was significantly greater in BC samples, indicating that this difference may partly reflect increased within-group heterogeneity rather than only a shift in group centroids. Because PICRUSt2 infers pathway abundance from 16 S rRNA gene profiles, these pathway-level findings represent predicted functional potential rather than confirmed functional activity. BC-associated gut microbiome dysbiosis appears to involve selective genus-level alterations, descriptive differences in inferred microbial association patterns, and broader predicted pathway-level shifts rather than complete community replacement. These findings highlight candidate microbial and pathway-level signatures that require validation using shotgun metagenomics, metabolomics, and independent cohorts.
