Nathansen, A. | bioRxiv Science

bioRxiv2025

Structure-Function Mapping of Olfactory Bulb Circuits with Synchrotron X-ray Nanotomography

Information is routed between brain areas via parallel streams. Neurons may share common inputs yet convey distinct information to different downstream targets. Here, we leverage the anatomical organisation of the mouse olfactory bulb (OB), where dozens of projection neurons (mitral and tufted cells, M/TCs) affiliate with a single input unit, a glomerulus1-3. To link functional properties of M/TCs to their anatomical glomerular association at scale, we combine in vivo two-photon (2P) imaging with synchrotron {micro}CT4-6 anatomical analysis and targeted X- ray nano-holotomography (XNH)7,8. Improving XNH resolution for mm3 volumes enables us to reliably identify subcellular features, automatically segment >80,000 cell nuclei in individual experiments, and delineate several hundred functionally imaged projection neurons and their detailed morphology, including up to 20 M/TCs per individual glomerulus ("sister" cells). In over 2400 sister cell pairs, we consistently find that odour response amplitudes to a panel of 47 monomolecular odours are conserved between sister cells, with, however, distinct responses to individual odours. Responses correlated with anatomical features such as cell body position and lateral dendritic arborisation. Thus, sister cells neither simply relay glomerular inputs nor are they dominated by network activity. Instead, they show a "balanced diversity" in their responses, enabling efficient encoding of odour stimuli whilst retaining the overall structure of odour space. Thus, synchrotron X-ray tomography can reliably link subcellular anatomy to function in a non-destructive way across the mm3 scale. With recent advances in X-ray optics9 and the emergence of 4th generation synchrotrons10,11, it becomes conceivable to extend this highly accessible approach to entire brain regions with increasing resolution.

neuroscience↗

bioRxiv2023

Evaluating Prompt Tuning for Conditional Protein Sequence Generation

Designing artificial proteins with specialized functions promises new solutions for biological, medical, and environmental use cases. This field benefits from advances in natural language processing, with state-of-the-art text generation models already being successfully applied to protein sequences. Openly available pre-trained protein language models are able to generate artificial protein sequences and can be finetuned on very specific tasks. Considering the high computational cost of finetuning a model exclusively for one downstream task, prompt tuning has been proposed as a more cost-efficient alternative that shares one model across different tasks. However, no openly available implementation of this approach compatible with protein language models has been previously published. Thus, we adapt an open-source codebase designed for NLP models to build a pipeline for prompt tuning on protein sequence data, supporting the protein language models ProtGPT2 and RITA. We benchmark this implementation for generating proteins of a specific family and evaluate the approach using text processing metrics as well as family membership prediction and protein activity prediction of generated sequences. Our results confirm the advantages of prompt tuning in resource usage, especially storage, encouraging further research and expansion of this technique to related use cases. For our evaluated use case, prompt tuning does not reach up to finetuning in terms of the quality of generated protein sequences, indicating the need for more extensive optimization. Lastly, we observe discrepancies between results of similar evaluation tools, highlighting open problems for principled assessment of protein sequence generation quality.

bioinformatics↗