Search bioRxiv⌕ Search

Biology subjects

Puhger, K.

Publications and source records attributed to Puhger, K..

3 recordsLinked to original sources

Life as a Function: Why Transformer Architectures Struggle to Gain Genome-Level Foundational Capabilities

Recent advances in generative models for nucleotide sequences have shown promise, but their practical utility remains limited. In this study, we explore DNA as a complex functional representation of evolutionary processes and assess the ability of transformer-based models to capture this complexity. Through experiments with both synthetic and real DNA sequences, we demonstrate that current transformer architectures, particularly auto-regressive models relying on next-token prediction, struggle to effectively learn the underlying biological functions. Our findings suggest that these models face inherent limitations, that cannot be overcome with scale, highlighting the need for alternative approaches that incorporate evolutionary constraints and structural information. We propose potential future directions, including the integration of topological methods or the switch of modelling paradigms, to enhance the generation of genomic sequences.

synthetic biology↗

Phasic locus coeruleus activity facilitates hippocampus-dependent trace fear memory formation

Locus coeruleus (LC) projections to the hippocampus play a critical role in learning and memory. However, the precise timing of LC-hippocampus communication during learning and which LC-derived neurotransmitters are important for memory formation in the hippocampus are currently unknown. Although the LC is typically thought to modulate neural activity via the release of norepinephrine, several recent studies have suggested that it may also release dopamine into the hippocampus and other cortical regions. In some cases, it appears that dopamine release from LC into the hippocampus may be more important for memory than norepinephrine. Here, we extend these data by characterizing the phasic responses of the LC and its projections to the dorsal hippocampus during trace fear conditioning. We find that the LC and its projections to the hippocampus respond to task-relevant stimuli and that amplifying these responses with optogenetic stimulation can enhance long-term memory formation. We also demonstrate that LC activity increases both norepinephrine and dopamine content in the dorsal hippocampus and that the timing of hippocampal dopamine release during trace fear conditioning is similar to the timing of LC activity. Finally, we show that hippocampal dopamine is important for trace fear memory formation, while norepinephrine is not.

animal behavior and cognition↗

The hippocampus contributes to retroactive stimulus associations in trace fear conditioning

Binding events that occur at different times is essential for memory formation. In trace fear conditioning, animals associate a tone and footshock despite no temporal overlap. The hippocampus is thought to mediate this learning by maintaining a memory of the tone until shock occurrence, however evidence for sustained hippocampal tone representations is lacking. Here we demonstrate a retrospective role for the hippocampus in trace fear conditioning. Bulk calcium imaging revealed sustained increases in CA1 activity after footshock that were not observed after tone termination. Optogenetic silencing of CA1 immediately after footshock impaired subsequent memory. Additionally, footshock increased the number of sharp wave-ripples compared to baseline during conditioning. Therefore, post-shock hippocampal activity likely supports learning by reactivating and linking latent tone and shock representations. These findings highlight an underappreciated function of post-trial hippocampal activity in enabling retroactive temporal associations during new learning, as opposed to persistent maintenance of stimulus representations.

animal behavior and cognition↗