Humayun Kabir

Motivation: Single-cell RNA sequencing (scRNA-seq) enables characterization of cellular heterogeneity at unprecedented resolution. However, accurate clustering of scRNA-seq data remains challenging due to high dimensionality, dropout-induced sparsity, and technical noise. Existing contrastive learning methods usually create augmented views by perturbing gene expression values, so the uncertainty in cell-cell relationships caused by dropout events is not properly modeled. Results: We propose scTGCL, a novel Transformer-based graph contrastive learning framework that learns robust and discriminative cell representations directly from gene expression data. scTGCL employs multi-head self-attention to adaptively construct weighted cell-cell graphs and a Transformer encoder to generate cell embeddings from them, capturing diverse biological relationships. For contrastive augmentation, we apply feature-level gene masking together with attention masking that randomly drops edges of the learned attention matrices, enabling the model to withstand dropout events and technical noise. A symmetric contrastive loss maximizes agreement between original and augmented representations, while joint optimization with reconstruction and imputation losses preserves biological interpretability. Extensive experiments on twenty real and simulated scRNA-seq datasets demonstrate that scTGCL achieves the highest average performance among nine state-of-the-art baseline methods in clustering accuracy, normalized mutual information, and adjusted Rand index, while also achieving superior imputation performance and substantially lower runtime on large-scale datasets. Furthermore, biological analysis confirms that the learned clusters align with known cell types and yield biologically interpretable pathway enrichment, validating the utility of scTGCL for downstream cell annotation and marker gene discovery. Availability and Implementation: The source code and test data are freely available at GitHub ( https://github.com/ShoaibAbdullahKhan/scTGCL ).

Identification of new drug and cell therapy targets for disease treatment will be facilitated by a detailed molecular understanding of normal and disease development. Human pluripotent stem cells can provide a large in vitro source of human cell types and, in a growing number of instances, also three-dimensional multicellular tissues called organoids. The application of stem cell technology to discovery and development of new therapies will be aided by detailed molecular characterisation of cell identity, cell signalling pathways and target gene networks. Big data or ‘omics’ techniques—particularly transcriptomics and proteomics—facilitate cell and tissue characterisation using thousands to tens-of-thousands of genes or proteins. These gene and protein profiles are analysed using existing and/or emergent bioinformatics methods, including a growing number of methods that compare sample profiles against compendia of reference samples. This review assesses how compendium-based analyses can aid the application of stem cell technology for new therapy development. This includes via robust definition of differentiated stem cell identity, as well as elucidation of complex signalling pathways and target gene networks involved in normal and diseased states.

Signaling pathways are the key biological mechanisms that transduce extracellular signals to affect transcription factor mediated gene regulation within cells. A number of computational methods have been developed to identify the topological structure of a specific signaling pathway using protein-protein interaction data, but they are not designed for identifying active signaling pathways in an unbiased manner. On the other hand, there are statistical methods based on gene sets or pathway data that can prioritize likely active signaling pathways, but they do not make full use of active pathway structure that link receptor, kinases and downstream transcription factors. Here, we present a method to simultaneously predict the set of active signaling pathways, together with their pathway structure, by integrating protein-protein interaction network and gene expression data. We evaluated the capacity for our method to predict active signaling pathways for dental epithelial cells, ocular lens epithelial cells, human pluripotent stem cell-derived lens epithelial cells, and lens fiber cells. This analysis showed our approach could identify all the known active pathways that are associated with tooth formation and lens development. The results suggest that SPAGI can be a useful approach to identify the potential active signaling pathways given a gene expression profile. Our method is implemented as an open source R package, available via https://github.com/VCCRI/SPAGI/ .

Cataracts cause vision loss and blindness by impairing the ability of the ocular lens to focus light onto the retina. Various cataract risk factors have been identified, including drug treatments, age, smoking and diabetes. However, the molecular events responsible for these different forms of cataract are ill-defined, and the advent of modern cataract surgery in the 1960s virtually eliminated access to human lenses for research. Here, we demonstrate large-scale production of light-focusing human micro-lenses from spheroidal masses of human lens epithelial cells purified from differentiating pluripotent stem cells. The purified lens cells and micro-lenses display similar morphology, cellular arrangement, mRNA expression and protein expression to human lens cells and lenses. Exposing the micro-lenses to the emergent cystic fibrosis drug Vx-770 reduces micro-lens transparency and focusing ability. These human micro-lenses provide a powerful and large-scale platform for defining molecular disease mechanisms caused by cataract risk factors, for anti-cataract drug screening and for clinically relevant toxicity assays.