Biology subjects

Leoni, G.

Publications and source records attributed to Leoni, G..

5 recordsLinked to original sources

bioRxiv2026

Evo 2 as a classification machine: evidence from in-context learning and mechanistic interpretability

In-context learning (iCL) is an emergent capability of Large Language Models (LLMs), allowing them to perform new tasks at inference time using prompt-injected examples. While extensively studied in LLMs, the boundaries of its capabilities and underlying mechanisms remain poorly characterised in genomic Language Models (gLMs). Here, we map the operating regime of Evo 2, a nucleotide-level foundation gLM, across five binary classification tasks spanning biological and artificial sequences. We observe robust iCL on shorter natural sequences (F1=0.902 for miRNA, 0.785 for Toxins), degrading with sequence length, and collapsing at kilobase scale. Strikingly, we find no benefit in model scaling, as the 7B model systematically outperforms the 40B variant. We further find that perplexity, a widely used gLM performance proxy, poorly predicts accuracy. Mechanistic interpretability reconciles these observations: the logit-lens profiling suggests a prediction-generalisation trade-off, while Jacobian Scope indicates models might track prompts' structure rather than the signal-carrying content.

genomics↗

bioRxiv2026

Evo 2's Perception of Single Nucleotide Substitutions in the Genes of Two Plant Model Organisms

Although DNA Large Language Models (DNA-LLMs) offer a path to decoding genetic complexity, our ability to evaluate these models is constrained by our incomplete understanding of the very same genetic syntax and functional logic that these models are trained to learn. In this study we use single nucleotide substitutions that have or have not been observed in living organisms, to evaluate how the DNA-LLM Evo 2 interprets gene sequences from two plant model organisms, Arabidopsis thaliana and Oryza sativa japonica. Using perplexity as a measure of the models confidence, we observe that alleles containing simulated substitutions are perceived, on average, as less likely than those observed in vivo. Although the size of the effect is modest, the effect is statistically significant and robust, suggesting that Evo 2 is aligned with our current understanding of evolutionary selective constraints. This approach is designed to be model-agnostic and species-agnostic and could serve as a generic framework for evaluating the performance of DNA-LLMs.

genomics↗

bioRxiv2024

The RNA binding protein LIN28A mediates chromatin dynamics during neuronal differentiation

The transition of embryonic stem cells (ESCs) from pluripotency to lineage commitment is regulated by multiple mechanisms, including chromatin dynamics and both transcriptional and post-transcriptional processes. Recent advances have highlighted that these mechanisms often interact, forming intricate multi-layered regulatory networks that require detailed characterization. In this study, we demonstrate that the RNA-binding protein LIN28A plays a pivotal role in neuronal differentiation by mediating RNA-dependent interactions with the Polycomb repressive complex 2 (PRC2). This interaction facilitates the eviction of PRC2 from chromatin, thereby activating a neuronal lineage-specific transcriptional program. Proteomic analyses revealed that the LIN28A interactome undergoes substantial remodeling during differentiation, corresponding to changes in LIN28A localization. In ESCs, LIN28A is predominantly nuclear and interacts with several components of the PRC2 complex in an RNA-dependent manner, assisting in chromatin dynamics. Our findings show that in the absence of LIN28A, PRC2 remains associated with chromatin, impairing the expression of genes critical for neuronal differentiation in ESCs. Chromatin immunoprecipitation sequencing (ChIP-seq) further confirmed that loss of LIN28A results in preferential PRC2 occupancy at the promoters of differentiation-associated genes. This study uncovers a novel role for LIN28A in epigenetic remodeling, which is essential for the proper differentiation of ESCs into the neuronal lineage.

molecular biology↗

bioRxiv2024

Broadly neutralizing antibodies against emerging delta-coronaviruses

Porcine deltacoronavirus (PDCoV) spillovers were recently detected in children with acute undifferentiated febrile illness, underscoring recurrent zoonoses of divergent coronaviruses. To date, no vaccines or specific therapeutics are approved for use in humans against PDCoV. To prepare for possible future PDCoV epidemics, we isolated human spike (S)-directed monoclonal antibodies from transgenic mice and found that two of them, designated PD33 and PD41, broadly neutralized a panel of PDCoV variants. Cryo-electron microscopy structures of PD33 and PD41 in complex with the PDCoV receptor-binding domain and S ectodomain trimer provide a blueprint of the epitopes recognized by these mAbs, rationalizing their broad inhibitory activity. We show that both mAbs inhibit PDCoV by competitively interfering with host APN binding to the PDCoV receptor-binding loops, explaining the mechanism of viral neutralization. PD33 and PD41 are candidates for clinical advancement, which could be stockpiled to prepare for possible future PDCoV outbreaks.

biochemistry↗

bioRxiv2023

Therapeutic and vaccine-induced cross-reactive antibodies with effector function against emerging Omicron variants

Currently circulating SARS-CoV-2 variants acquired convergent mutations at receptor-binding domain (RBD) hot spots1. Their impact on viral infection, transmission, and efficacy of vaccines and therapeutics remains poorly understood. Here, we demonstrate that recently emerged BQ.1.1. and XBB.1 variants bind ACE2 with high affinity and promote membrane fusion more efficiently than earlier Omicron variants. Structures of the BQ.1.1 and XBB.1 RBDs bound to human ACE2 and S309 Fab (sotrovimab parent) explain the altered ACE2 recognition and preserved antibody binding through conformational selection. We show that sotrovimab binds avidly to all Omicron variants, promotes Fc-dependent effector functions and protects mice challenged with BQ.1.1, the variant displaying the greatest loss of neutralization. Moreover, in several donors vaccine-elicited plasma antibodies cross-react with and trigger effector functions against Omicron variants despite reduced neutralizing activity. Cross-reactive RBD-directed human memory B cells remained dominant even after two exposures to Omicron spikes, underscoring persistent immune imprinting. Our findings suggest that this previously overlooked class of cross-reactive antibodies, exemplified by S309, may contribute to protection against disease caused by emerging variants through elicitation of effector functions.

immunology↗