Cell tracking and analysis software for biological research. YeastTrack detects and follows individual cells across microscopy image sequences to support quantitative analysis.

End-to-end analysis of 10X Visium spatial transcriptomics data from human breast cancer tissue. The pipeline combines standard bioinformatics — QC, Leiden clustering, Cell2location deconvolution, spatially variable genes, and neighbourhood enrichment — with a graph convolutional network (PyTorch Geometric) that learns tissue domain labels directly from the spatial gene-expression graph. Built with Scanpy, Squidpy, and AnnData.

An end-to-end deep learning project that predicts whether a 251 bp human DNA sequence is a non-TATA promoter from sequence alone, using a convolutional neural network — then recovers the biological motifs the model learned as sequence logos (rediscovering known regulatory elements such as GC-boxes / Sp1 binding sites). Following the DeepBind / Basset / DeepSEA lineage, it reaches 0.905 accuracy and 0.969 AUROC on the human_nontata_promoters benchmark. Built with PyTorch.

A structure-based virtual screen identifying microbial enzyme candidates for the degradation of the lipofuscin components A2E and 7-ketocholesterol, in support of the SENS LysoSENS strategy. The pipeline mines UniProt for candidate enzymes, fetches AlphaFold structures and PubChem ligands, and runs 246 AutoDock Vina docking experiments across 123 bacterial enzymes × 2 substrates — surfacing 90 enzyme–ligand pairs with predicted binding free energies ≤ −9.0 kcal/mol.