Evolving the olfactory system with machine learning

An end-to-end model of active electrosensation

Weakly electric fish localize and identify objects by sensing distortions in a self-generated electric field. Fish can determine the resistance and capacitance of an object, for example, even though the field distortions being sensed are small and highly dependent on object distance and size. The neural computations underlying these remarkable behavioral capacities are poorly understood. We measured responses of electroreceptor afferents and constructed a filter-based model that accurately accounts for them. We also built models of the electric fields generated by the fish and of the distortions in these fields due to objects of different resistances and capacitances at different locations. Combining these models provides an accurate and efficient method for generating large artificial datasets simulating fish interacting with a wide variety of objects. Using these sets, we trained an artificial neural network (ANN), representing brain areas downstream of the electroreceptors, to extract the 3D location, size, and electrical properties of objects. Model performance is comparable to that of real fish in experimentally tested behavioral tasks. Performance is maximized if the ANN operates in two stages: first estimating object distance and size and then using this information to extract electrical properties. These results highlight the potential of end-to-end modeling for studies of electrosensation and suggest a specific form of modularity in electrosensory processing that can be tested experimentally.

Universality of representation in biological and artificial neural networks

Many artificial neural networks (ANNs) trained with ecologically plausible objectives on naturalistic data align with behavior and neural representations in biological systems. Here, we show that this alignment is a consequence of convergence onto the same representations by high-performing ANNs and by brains. We developed a method to identify stimuli that systematically vary the degree of inter-model representation agreement. Across language and vision, we then showed that stimuli from high- and low-agreement sets predictably modulated model-to-brain alignment. We also examined which stimulus features distinguish high- from low-agreement sentences and images. Our results establish representation universality as a core component in the model-to-brain alignment and provide a new approach for using ANNs to uncover the structure of biological representations and computations.

Parallel development of social behavior in biological and artificial fish

Our algorithmic understanding of vision has been revolutionized by a reverse engineering paradigm that involves building artificial systems that perform the same tasks as biological systems. Here, we extend this paradigm to social behavior. We embodied artificial neural networks in artificial fish and raised the artificial fish in virtual fish tanks that mimicked the rearing conditions of biological fish. When artificial fish had deep reinforcement learning and curiosity-derived rewards, they spontaneously developed fish-like social behaviors, including collective behavior and social preferences (favoring in-group over out-group members). The artificial fish also developed social behavior in naturalistic ocean worlds, showing that these embodied models generalize to real-world learning contexts. Thus, animal-like social behaviors can develop from generic learning algorithms (reinforcement learning and intrinsic motivation). Our study provides a foundation for reverse-engineering the development of social behavior using image-computable models from artificial intelligence, bridging the divide between high-dimensional sensory inputs and collective action.

A Drosophila-inspired intelligent olfactory biomimetic sensing system for gas recognition in complex environments

The olfactory sensory system of Drosophila has several advantages, including low power consumption, high rapidity and high accuracy. Here, we present a biomimetic intelligent olfactory sensing system based on the integration of an 18-channel microelectromechanical system (MEMS) sensor array (16 gas sensors, 1 humidity sensor and 1 temperature sensor), a complementary metal‒oxide‒semiconductor (CMOS) circuit and an olfactory lightweight machine-learning algorithm inspired by Drosophila. This system is an artificial version of the biological olfactory perception system with the capabilities of environmental sensing, multi-signal processing, and odor recognition. The olfactory data are processed and reconstructed by the combination of a shallow neural network and a residual neural network, with the aim to determine the noxious gas information in challenging environments such as high humidity scenarios and partially damaged sensor units. As a result, our electronic olfactory sensing system is capable of achieving comprehensive gas recognition by qualitatively identifying 7 types of gases with an accuracy of 98.5%, reducing the number of parameters and the difficulty of calculation, and quantitatively predicting each gas of 3-5 concentration gradients with an accuracy of 93.2%; thus, these results show superiority of our system in supporting alarm systems in emergency rescue scenarios.

Odor representation and coding by the mitral/tufted cells in the olfactory bulb

The olfactory bulb (OB) is the first relay station in the olfactory system and functions as a crucial hub. It can represent odor information precisely and accurately in an ever-changing environment. As the only output neurons in the OB, mitral/tufted cells encode information such as odor identity and concentration. Recently, the neural strategies and mechanisms underlying odor representation and encoding in the OB have been investigated extensively. Here we review the main progress on this topic. We first review the neurons and circuits involved in odor representation, including the different cell types in the OB and the neural circuits within and beyond the OB. We will then discuss how two different coding strategies-spatial coding and temporal coding-work in the rodent OB. Finally, we discuss potential future directions for this research topic. Overall, this review provides a comprehensive description of our current understanding of how odor information is represented and encoded by mitral/tufted cells in the OB.

Shaping dynamical neural computations using spatiotemporal constraints

Dynamics play a critical role in computation. The principled evolution of states over time enables both biological and artificial networks to represent and integrate information to make decisions. In the past few decades, significant multidisciplinary progress has been made in bridging the gap between how we understand biological versus artificial computation, including how insights gained from one can translate to the other. Research has revealed that neurobiology is a key determinant of brain network architecture, which gives rise to spatiotemporally constrained patterns of activity that underlie computation. Here, we discuss how neural systems use dynamics for computation, and claim that the biological constraints that shape brain networks may be leveraged to improve the implementation of artificial neural networks. To formalize this discussion, we consider a natural artificial analog of the brain that has been used extensively to model neural computation: the recurrent neural network (RNN). In both the brain and the RNN, we emphasize the common computational substrate atop which dynamics occur-the connectivity between neurons-and we explore the unique computational advantages offered by biophysical constraints such as resource efficiency, spatial embedding, and neurodevelopment.

Digital Twin Studies for Reverse Engineering the Origins of Visual Intelligence

What are the core learning algorithms in brains? Nativists propose that intelligence emerges from innate domain-specific knowledge systems, whereas empiricists propose that intelligence emerges from domain-general systems that learn domain-specific knowledge from experience. We address this debate by reviewing digital twin studies designed to reverse engineer the learning algorithms in newborn brains. In digital twin studies, newborn animals and artificial agents are raised in the same environments and tested with the same tasks, permitting direct comparison of their learning abilities. Supporting empiricism, digital twin studies show that domain-general algorithms learn animal-like object perception when trained on the first-person visual experiences of newborn animals. Supporting nativism, digital twin studies show that domain-general algorithms produce innate domain-specific knowledge when trained on prenatal experiences (retinal waves). We argue that learning across humans, animals, and machines can be explained by a universal principle, which we call space-time fitting. Space-time fitting explains both empiricist and nativist phenomena, providing a unified framework for understanding the origins of intelligence.

Overcoming the Limits of Cross-Sensitivity: Pattern Recognition Methods for Chemiresistive Gas Sensor Array

As information acquisition terminals for artificial olfaction, chemiresistive gas sensors are often troubled by their cross-sensitivity, and reducing their cross-response to ambient gases has always been a difficult and important point in the gas sensing area. Pattern recognition based on sensor array is the most conspicuous way to overcome the cross-sensitivity of gas sensors. It is crucial to choose an appropriate pattern recognition method for enhancing data analysis, reducing errors and improving system reliability, obtaining better classification or gas concentration prediction results. In this review, we analyze the sensing mechanism of cross-sensitivity for chemiresistive gas sensors. We further examine the types, working principles, characteristics, and applicable gas detection range of pattern recognition algorithms utilized in gas-sensing arrays. Additionally, we report, summarize, and evaluate the outstanding and novel advancements in pattern recognition methods for gas identification. At the same time, this work showcases the recent advancements in utilizing these methods for gas identification, particularly within three crucial domains: ensuring food safety, monitoring the environment, and aiding in medical diagnosis. In conclusion, this study anticipates future research prospects by considering the existing landscape and challenges. It is hoped that this work will make a positive contribution towards mitigating cross-sensitivity in gas-sensitive devices and offer valuable insights for algorithm selection in gas recognition applications.

A virtual rodent predicts the structure of neural activity across behaviours

Animals have exquisite control of their bodies, allowing them to perform a diverse range of behaviours. How such control is implemented by the brain, however, remains unclear. Advancing our understanding requires models that can relate principles of control to the structure of neural activity in behaving animals. Here, to facilitate this, we built a 'virtual rodent', in which an artificial neural network actuates a biomechanically realistic model of the rat1 in a physics simulator2. We used deep reinforcement learning3-5 to train the virtual agent to imitate the behaviour of freely moving rats, thus allowing us to compare neural activity recorded in real rats to the network activity of a virtual rodent mimicking their behaviour. We found that neural activity in the sensorimotor striatum and motor cortex was better predicted by the virtual rodent's network activity than by any features of the real rat's movements, consistent with both regions implementing inverse dynamics6. Furthermore, the network's latent variability predicted the structure of neural variability across behaviours and afforded robustness in a way consistent with the minimal intervention principle of optimal feedback control7. These results demonstrate how physical simulation of biomechanically realistic virtual animals can help interpret the structure of neural activity across behaviour and relate it to theoretical principles of motor control.

Benchmarking variational AutoEncoders on cancer transcriptomics data

Deep generative models, such as variational autoencoders (VAE), have gained increasing attention in computational biology due to their ability to capture complex data manifolds which subsequently can be used to achieve better performance in downstream tasks, such as cancer type prediction or subtyping of cancer. However, these models are difficult to train due to the large number of hyperparameters that need to be tuned. To get a better understanding of the importance of the different hyperparameters, we examined six different VAE models when trained on TCGA transcriptomics data and evaluated on the downstream tasks of cluster agreement with cancer subtypes and survival analysis. We studied the effect of the latent space dimensionality, learning rate, optimizer, initialization and activation function on the quality of subsequent downstream tasks on the TCGA samples. We found β-TCVAE and DIP-VAE to have a good performance, on average, despite being more sensitive to hyperparameters selection. Based on these experiments, we derived recommendations for selecting the different hyperparameters settings. To ensure generalization, we tested all hyperparameter configurations on the GTEx dataset. We found a significant correlation (ρ = 0.7) between the hyperparameter effects on clustering performance in the TCGA and GTEx datasets. This highlights the robustness and generalizability of our recommendations. In addition, we examined whether the learned latent spaces capture biologically relevant information. Hereto, we measured the correlation and mutual information of the different representations with various data characteristics such as gender, age, days to metastasis, immune infiltration, and mutation signatures. We found that for all models the latent factors, in general, do not uniquely correlate with one of the data characteristics nor capture separable information in the latent factors even for models specifically designed for disentanglement.

XGBoost odor prediction model: finding the structure-odor relationship of odorant molecules using the extreme gradient boosting algorithm

Determining the structure-odor relationship has always been a very challenging task. The main challenge in investigating the correlation between the molecular structure and its associated odor is the ambiguous and obscure nature of verbally defined odor descriptors, particularly when the odorant molecules are from different sources. With the recent developments in machine learning (ML) technology, ML and data analytic techniques are significantly being used for quantitative structure-activity relationship (QSAR) in the chemistry domain toward knowledge discovery where the traditional Edisonian methods have not been useful. The smell perception of odorant molecules is one of the aforementioned tasks, as olfaction is one of the least understood senses as compared to other senses. In this study, the XGBoost odor prediction model was generated to classify smells of odorant molecules from their SMILES strings. We first collected the dataset of 1278 odorant molecules with seven basic odor descriptors, and then 1875 physicochemical properties of odorant molecules were calculated. To obtain relevant physicochemical features, a feature reduction algorithm called PCA was also employed. The ML model developed in this study was able to predict all seven basic smells with high precision (>99%) and high sensitivity (>99%) when tested on an independent test dataset. The results of the proposed study were also compared with three recently conducted studies. The results indicate that the XGBoost-PCA model performed better than the other models for predicting common odor descriptors. The methodology and ML model developed in this study may be helpful in understanding the structure-odor relationship.Communicated by Ramaswamy H. Sarma.

Normative and mechanistic model of an adaptive circuit for efficient encoding and feature extraction

One major question in neuroscience is how to relate connectomes to neural activity, circuit function, and learning. We offer an answer in the peripheral olfactory circuit of the Drosophila larva, composed of olfactory receptor neurons (ORNs) connected through feedback loops with interconnected inhibitory local neurons (LNs). We combine structural and activity data and, using a holistic normative framework based on similarity-matching, we formulate biologically plausible mechanistic models of the circuit. In particular, we consider a linear circuit model, for which we derive an exact theoretical solution, and a nonnegative circuit model, which we examine through simulations. The latter largely predicts the ORN [Formula: see text] LN synaptic weights found in the connectome and demonstrates that they reflect correlations in ORN activity patterns. Furthermore, this model accounts for the relationship between ORN [Formula: see text] LN and LN-LN synaptic counts and the emergence of different LN types. Functionally, we propose that LNs encode soft cluster memberships of ORN activity, and partially whiten and normalize the stimulus representations in ORNs through inhibitory feedback. Such a synaptic organization could, in principle, autonomously arise through Hebbian plasticity and would allow the circuit to adapt to different environments in an unsupervised manner. We thus uncover a general and potent circuit motif that can learn and extract significant input features and render stimulus representations more efficient. Finally, our study provides a unified framework for relating structure, activity, function, and learning in neural circuits and supports the conjecture that similarity-matching shapes the transformation of neural representations.

Model of Neuromorphic Odorant-Recognition Network

We propose a new model for a neuromorphic olfactory analyzer based on memristive synapses. The model comprises a layer of receptive neurons that perceive various odors and a layer of "decoder" neurons that recognize these odors. It is demonstrated that connecting these layers with memristive synapses enables the training of the "decoder" layer to recognize two types of odorants of varying concentrations. In the absence of such synapses, the layer of "decoder" neurons does not exhibit specificity in recognizing odorants. The recognition of the 'odorant' occurs through the neural activity of a group of decoder neurons that have acquired specificity for the odorant in the learning process. The proposed phenomenological model showcases the potential use of a memristive synapse in practical odorant recognition applications.

Metabolic activity organizes olfactory representations

Hearing and vision sensory systems are tuned to the natural statistics of acoustic and electromagnetic energy on earth and are evolved to be sensitive in ethologically relevant ranges. But what are the natural statistics of odors, and how do olfactory systems exploit them? Dissecting an accurate machine learning model (Lee et al., 2022) for human odor perception, we find a computable representation for odor at the molecular level that can predict the odor-evoked receptor, neural, and behavioral responses of nearly all terrestrial organisms studied in olfactory neuroscience. Using this olfactory representation (principal odor map [POM]), we find that odorous compounds with similar POM representations are more likely to co-occur within a substance and be metabolically closely related; metabolic reaction sequences (Caspi et al., 2014) also follow smooth paths in POM despite large jumps in molecular structure. Just as the brain's visual representations have evolved around the natural statistics of light and shapes, the natural statistics of metabolism appear to shape the brain's representation of the olfactory world.

Neuroscience: Seq-ing maps in the olfactory cortex

Many cortical brain regions are spatially organized to optimize sensory representation. Such topographic maps have so far been elusive in the olfactory cortex. A high-throughput tracing study reveals that the neural circuits connecting olfactory regions are indeed topographically organized.

The human odorant receptor OR10A6 is tuned to the pheromone of the commensal fruit fly Drosophila melanogaster

All living things speak chemistry. The challenge is to reveal the vocabulary, the odorants that enable communication across phylogenies and to translate them to physiological, behavioral, and ecological function. Olfactory receptors (ORs) interface animals with airborne odorants. Expression in heterologous cells makes it possible to interrogate single ORs and to identify cognate ligands. The cosmopolitan, anthropophilic strain of the vinegar fly Drosophila melanogaster depends on human resources and housing for survival. Curiously, humans sense the pheromone (Z)-4-undecenal (Z4-11Al) released by single fly females. A screening of all human ORs shows that the most highly expressed OR10A6 is tuned to Z4-11Al. Females of an ancestral African fly strain release a blend of Z4-11Al and Z4-9Al that produces a different aroma, which is how we distinguish these fly strains by nose. That flies and humans sense Z4-11Al via dedicated ORs shows how convergent evolution shapes communication channels between vertebrate and invertebrate animals.

•

Humans sense the sex pheromone Z411-Al released by single Drosophila melanogaster females

•

The most highly expressed human olfactory receptor OR10A6 is tuned to Z411-Al

•

An African fly strain emits two aldehydes, which we distinguish from Z411-Al by nose

•

Convergent evolution shapes chemical communication between phylogenies

Artificial olfactory sensor technology that mimics the olfactory mechanism: a comprehensive review

Artificial olfactory sensors that recognize patterns transmitted by olfactory receptors are emerging as a technology for monitoring volatile organic compounds. Advances in statistical processing methods and data processing technology have made it possible to classify patterns in sensor arrays. Moreover, biomimetic olfactory recognition sensors in the form of pattern recognition have been developed. Deep learning and artificial intelligence technologies have enabled the classification of pattern data from more sensor arrays, and improved artificial olfactory sensor technology is being developed with the introduction of artificial neural networks. An example of an artificial olfactory sensor is the electronic nose. It is an array of various types of sensors, such as metal oxides, electrochemical sensors, surface acoustic waves, quartz crystal microbalances, organic dyes, colorimetric sensors, conductive polymers, and mass spectrometers. It can be tailored depending on the operating environment and the performance requirements of the artificial olfactory sensor. This review compiles artificial olfactory sensor technology based on olfactory mechanisms. We introduce the mechanisms of artificial olfactory sensors and examples used in food quality and stability assessment, environmental monitoring, and diagnostics. Although current artificial olfactory sensor technology has several limitations and there is limited commercialization owing to reliability and standardization issues, there is considerable potential for developing this technology. Artificial olfactory sensors are expected to be widely used in advanced pattern recognition and learning technologies, along with advanced sensor technology in the future.