Biology subjects

Koulakov, A.

Publications and source records attributed to Koulakov, A..

7 recordsLinked to original sources

bioRxiv2026

Grounding olfactory perception in language: Benchmarks and models for generating natural language odor descriptions

Recent advances in deep learning have enabled prediction of odorant perception from molecular structure, opening new avenues for odor classification. However, most existing models are limited to predicting percepts from fixed vocabularies and fail to capture the full richness of olfactory experience. Progress is further limited by the scarcity of large-scale olfactory datasets and the lack of standardized metrics for evaluating free-form natural-language odor descriptions. To address these challenges, we introduce Odor Description and Inference Evaluation Understudy (ODIEU), a benchmark which includes perceptual descriptions of over 10,000 molecules paired with a model-based metric for evaluating free-form odor text descriptions. The model-based metric uses Sentence-BERT (SBERT) models which are fine-tuned on olfactory descriptions to allow better evaluation of human-generated odor texts. Using the fine-tuned SBERT models, we show that free-form text odor descriptions contain additional perceptual information in their syntactic structure compared to semantic labels. We further introduce CIRANO (Chemical Information Recognition and Annotation Network for Odors), a transformer-based model that generates free-form odor descriptions directly from molecular structure, thus implementing the molecular structure-to-text (S2T) prediction. CIRANO achieves performance comparable to humans. Finally, we generate human-like descriptions from mouse olfactory bulb neural data using an invertible SBERT model, yielding neural-to-text (N2T) predictions highly aligned with human descriptions. Together, CIRANO and ODIEU establish a standardized framework for generating natural language olfactory descriptions and evaluating their alignment with human perception. Code is available at https://github.com/KoulakovLab/ODIEU

animal behavior and cognition↗

bioRxiv2026

How the olfactory bulb maintains stable odor manifolds amid adaptation and representational drift

Adaptive coding in sensory circuits enables stable perception while accommodating experience-dependent changes. In the olfactory bulb (OB), repeated odor exposure reshapes population activity even without explicit behavioral feedback, but the underlying circuit mechanisms remain unclear. By analyzing longitudinal two-photon calcium imaging datasets from the mouse OB, we identified three concurrent forms of representational change: gain adaptation, similarity-dependent pattern separation or convergence, and a rotation of encoding subspace resulting in the representational drift. Using a computational model of the mitral cell-granule cell circuit, we showed that Hebbian plasticity and structural connectivity constraints are sufficient to reproduce these transformations. Despite global representational drift, the relative geometry of odor response vectors remained stable, preserving a low-dimensional odor manifold. Together, our results reveal how local plasticity and network structure jointly enable both stability and flexibility in early sensory coding.

neuroscience↗

bioRxiv2025

Order code in the olfactory system.

The ability to recognize odor identity across a wide range of concentrations is essential for natural olfactory behaviors. However, how odor identity is represented in the early olfactory system remains an open question. One theory proposes that glomeruli in the olfactory bulb, along with their associated odorant receptors (ORs), are activated by odorants in a temporal order that conveys information about odor identity. This order code is relatively robust to concentration changes, as the rank order of glomerular activation by a given odorant remains similar across concentrations. Alternatively, the primacy coding theory suggests that odor identity is defined by the most sensitive ORs, those that respond first, collectively referred to as the primacy set. To test these theories, we measured glomerular responses to a large set of odorants in the mouse olfactory bulb using calcium imaging. We found that receptor affinities can be embedded in a low-dimensional space (D = 10) with minimal loss of information. Within this space, we identified two clusters of glomeruli with distinct tuning properties that form independent odor representations. These clusters may correspond to the two phylogenetic classes of ORs, as revealed by both their functional characteristics and anatomical locations. In the OR affinity space, odorants evoke orderly activation waves whose directions can be used to define odor identities in the order coding model. Finally, we compared the order code and primacy model in predicting odor identity both across concentrations and across animals. Despite containing less information overall, the primacy model performed comparably to the order code. We confirmed the primacy models prediction that each receptor tends to contribute to the primacy set of at least one odorant. Analysis of binary odor mixtures revealed that mixture responses lie along geodesic lines connecting their component odors. Together, these findings suggest that odor information in the olfactory bulb may be represented through two complementary coding strategies: the primacy and order codes.

neuroscience↗

bioRxiv2025

A quantitative framework for predicting odor intensity across molecule and mixtures

In vision and hearing, standardized units such as lumens (for brightness) and decibels (for loudness) allow consistent quantification of stimulus intensity, enabling precise control of sensory experiences. Olfaction, by contrast, currently lacks a robust quantitative framework linking physical stimulus properties directly to perceived odor intensity, complicating efforts to accurately characterize and manipulate aromas. To bridge this gap, we used a precisely controlled odor delivery system combined with deep learning models to predict the intensity of both single molecules and mixtures from physical properties. These models allowed us to develop an automated, quantitative method that accurately identifies which volatile components meaningfully contribute to aroma perception, overcoming the limitations of traditional heuristic approaches such as odor activity values and demonstrating practical utility in complex naturalistic odors.

neuroscience↗

bioRxiv2024

Weight Transfer in the Reinforcement Learning Model of Songbird Acquisition

Song acquisition behavior observed in the songbird system provides a notable example of learning through trial- and-error which parallels human speech acquisition. Studying songbird vocal learning can offer insights into mechanisms underlying human language. We present a computational model of song learning that integrates reinforcement learning (RL) and Hebbian learning and agrees with known songbird circuitry. The song circuit outputs activity from nucleus RA, which receives two primary inputs: timing information from area HVC and stochastic activity from nucleus LMAN. Additionally, song learning relies on Area X, a basal ganglia area that receives dopaminergic inputs from VTA. In our model, song is first acquired in the HVC-to-Area X connectivity, employing an RL mechanism that involves node perturbation. This information is then consolidated into HVC-to-RA synapses through a Hebbian mechanism. The transfer of weights from Area X to RA takes place via the thalamus, utilizing a specific form of spike-timing-dependent plasticity (STDP). Thus, we present a computational model grounded in songbird circuitry in which the optimal policy is initially guided by RL and subsequently transferred to another circuit through Hebbian plasticity.

neuroscience↗

bioRxiv2021

Encoding innate ability through a genomic bottleneck

AO_SCPLOWBSTRACTC_SCPLOWAnimals are born with extensive innate behavioral capabilities, which arise from neural circuits encoded in the genome. However, the information capacity of the genome is orders of magnitude smaller than that needed to specify the connectivity of an arbitrary brain circuit, indicating that the rules encoding circuit formation must fit through a "genomic bottleneck" as they pass from one generation to the next. Here we formulate the problem of innate behavioral capacity in the context of artificial neural networks in terms of lossy compression of the weight matrix. We find that several standard network architectures can be compressed by several orders of magnitude, yielding pre-training performance that can approach that of the fully-trained network. Interestingly, for complex but not for simple test problems, the genomic bottleneck algorithm also captures essential features of the circuit, leading to enhanced transfer learning to novel tasks and datasets. Our results suggest that compressing a neural circuit through the genomic bottleneck serves as a regularizer, enabling evolution to select simple circuits that can be readily adapted to important real-world tasks. The genomic bottleneck also suggests how innate priors can complement conventional approaches to learning in designing algorithms for artificial intelligence.

neuroscience↗

bioRxiv2018

Mosaic representations of odors in the input and output layers of the mouse olfactory bulb

The elementary stimulus features encoded by the olfactory system remain poorly understood. We examined the relationship between 1,666 physical-chemical descriptors of odors and the activity of olfactory bulb inputs as well as outputs in awake mice. Glomerular and M/T cell responses were sparse and locally heterogeneous, with only a coarse dependence of glomerular positions on physical-chemical properties. Odor features represented by ensembles of M/T cells were overlapping, but distinct from those represented in glomeruli, consistent with extensive interplay between feedforward and feedback inputs to the bulb. This reformatting was well-described as a rotation in odor space. The descriptors accounted for a small fraction in response variance, and the similarity of odors in physical-chemical space was a poor predictor of similarity in neuronal representations. Our results suggest that commonly used physical-chemical properties are not systematically represented in bulbar activity and encourage further search for better descriptors of odor space.

neuroscience↗