DefensePredictor: A machine learning model to discover prokaryotic immune systems

Bacteria deploy diverse immune systems to fend off viral attack, and cataloguing these systems has driven major biotechnological breakthroughs. Most discovery efforts rely on the tendency of defense genes to cluster together in the genome, but this misses systems located elsewhere. Researchers developed DefensePredictor, a machine learning classifier that uses protein language model embeddings to identify defensive proteins regardless of their genomic location. Applied to 69 Escherichia coli strains, the tool predicted hundreds of novel defense systems, and 42 were experimentally confirmed. Scaling the approach to 1,000 prokaryotic genomes turned up nearly 3,000 protein clusters with no similarity to any known defense system, pointing to a large unexplored immune repertoire. By moving beyond genomic neighborhood as the sole discovery signal, DefensePredictor offers a broader path toward understanding how bacterial immunity relates to eukaryotic innate immunity and generating new tools for biotechnology.

Read the original article →

Targeting modulated vascular smooth muscle cells in atherosclerosis via FAP-directed immunotherapy

Researchers investigated how vascular smooth muscle cells (VSMCs) change identity during atherosclerotic coronary artery disease (CAD) and whether these transformed cells could be therapeutically targeted. Using single-cell multiomic profiling, epitope mapping, and spatial transcriptomics across 27 human coronary arteries, they identified fibroblast activation protein (FAP) as a marker of VSMCs that have shifted to a disease-promoting state. Mouse lineage tracing confirmed these FAP-positive cells derive from VSMCs, and FAP-targeted PET imaging in CAD patients revealed uptake in plaques. These FAP-positive cells clustered within macrophage-rich neointimal regions. The team then engineered a bispecific T cell engager antibody directed at FAP, which reduced plaque burden in preclinical models and restructured the local immune environment through T cell expansion. The work provides a detailed cellular atlas of human CAD and opens a potential immunotherapy avenue targeting disease-relevant cell states independently of lipid-lowering strategies.

Read the original article →

Protein and genomic language models uncover the unexplored diversity of bacterial immunity

Bacteria possess a vast arsenal of defenses against phages, but most remain undiscovered. Researchers developed several machine learning approaches—drawing on genomic context, protein sequences, or both—to predict which bacterial genes encode antiphage functions. The models performed with up to 99% precision and 92% recall, and experimental validation in Escherichia and Streptomyces confirmed 12 previously unknown defense systems. Applying these models across more than 32,000 bacterial genomes revealed that roughly 1.5% of each genome is dedicated to immunity, with over 85% of predicted protein families still functionally uncharacterized. The work produces an interactive catalog of more than 19,000 candidate defense operon families awaiting experimental follow-up. The central takeaway is that the great majority of bacterial immune diversity has yet to be explored, and these tools offer a practical roadmap for doing so systematically.

Read the original article →

Concurrent L1 retrotransposition events promote reciprocal translocations in human tumorigenesis

Researchers examined how LINE-1 (L1) jumping DNA elements reshape cancer genomes, a question previously hampered by short-read sequencing technologies. Using long-read sequencing on ten tumors with unusually high rates of L1 activity—together totaling over 6,000 somatic insertion events—they discovered that L1 elements frequently trigger reciprocal chromosomal translocations. These rearrangements typically arise when two L1 retrotransposition events occur simultaneously on different chromosomes. Extending the analysis to a broader tumor cohort with varying L1 activity levels, the team estimated that roughly one chromosomal rearrangement emerges for every 60 somatic retrotransposition events. Molecular timing analyses further showed that these structural changes appear early in tumor development, implicating L1 activity as an initiating force in chromosomal instability rather than a late consequence of cancer progression. The findings establish L1 retrotransposition as a meaningful contributor to cancer genome evolution in tumors with high retrotransposon activity.

Read the original article →

Untangling the LINEs

Mobile genetic elements called LINEs (Long Interspersed Nuclear Elements) have long been suspected of reshaping cancer genomes, but short-read sequencing technology made it difficult to fully characterize their activity. Using long-read sequencing, researchers examined cancer genomes at much higher resolution and uncovered a broader and more complex landscape of LINE-driven genomic rearrangements than previously appreciated. The approach allowed them to detect structural variants and insertions that were invisible to conventional methods, revealing how these jumping DNA sequences actively contribute to chromosomal instability in tumors. The findings suggest that mobile element activity is a more significant driver of cancer genome evolution than earlier estimates indicated, with implications for understanding tumor heterogeneity and potentially identifying new genomic vulnerabilities. Better mapping of these rearrangements could improve our ability to interpret cancer genomes and design targeted therapeutic strategies.

Read the original article →