Speaker: Alfonso Rincón Pérez.
Abstract: Bridging chromatin architecture to transcriptional regulation is essential for uncovering the epigenetic mechanisms driving gene expression. While supervised predictive models provide valuable benchmarks for discriminating active from repressed regimes, rigid classifications often struggle with the statistical noise and overwhelming background density of non-significant genes. In this seminar, we briefly recap our initial predictive modeling findings and introduce an unsupervised analytical framework designed to discover latent chromatin signatures across promoter regions (2000 base-pairs before and after the center of the region). Using differential coverage signals (WT – KO) across multi-omic sequencing tracks (ATAC-seq and multi-mark ChIP-seq), we construct an orthogonal 700-dimensional multiscale feature space capturing both sharp transcription start site (TSS) dynamics and broader distal chromatin variations without collinear redundancy. To prevent the massive pool of neutral genes from acting as an artificial density attractor, our UMAP and HDBSCAN pipeline is trained strictly on differentially expressed genes (DEGs), isolating robust epigenetic states while explicitly segregating technical noise. Statistically non-significant (NS) genes are subsequently mapped onto this manifold via inductive, out-of-sample projection, contextualizing basal chromatin behavior against functional regulatory states. Finally, we demonstrate a high-performance interactive exploration platform operating at native base-pair resolution with real-time variance estimation (intervals of one standard deviation), setting the stage for downstream motif enrichment and biological pathway validation.