Biology subjects

Ferrari, U.

Publications and source records attributed to Ferrari, U..

At least 19 recordsLinked to original sources

bioRxiv2026

The multi-scale geometry of neural encoding

As recorded neural populations become larger and artificial networks more powerful, there is growing interest in understanding and comparing what different populations encode. Representational geometry offers a powerful way to characterise this: it remaps stimulus space to stretch well-encoded stimulus differences while shrinking poorly encoded ones. However, this remapping can be constructed in many ways from neural responses. Here, we introduce a multi-scale representational geometry, derived from first principles, that captures how neurons encode stimulus differences at every scale. It is exactly related to mutual information, so independent of decoding assumptions. The more a stimulus difference contributes to information, the more the geometry stretches it. Applied to spatial, motor and visual codes, it shows how neurons combine to encode information, how areas differ, and which features are best encoded. A single framework thus maps the structure of encoded information, uniting geometric and information-theoretic views of neural coding.

neuroscience↗

bioRxiv2026

Biologically grounded locality priors close the data gap for vision transformers in neural prediction

For datasets with thousands of neurons and images, vision transformers have proven successful at predicting neural responses to stimuli. However, they are expected to underperform in low-data regimes, where CNNs and Gaussian processes are considered more effective. We ask whether transformers can be made competitive for small-scale neural prediction, and show that underperformance in this regime can be overturned with the right inductive bias. We equip a vision transformer with a differentiable per-neuron circular crop in feature space. The crop is centered on each neuron's receptive field, with a radius selected per neuron, so the model only sees the small image region that drives that neuron instead of the whole image. This makes the cost of attention scale with the size of the receptive field rather than with the size of the image. A single transformer stack is shared across all neurons: each neuron's specificity resides in the crop, not in the architecture. We call this model circular receptive-field vision transformer (CiRF-ViT). We evaluate it on small multi-electrode-array recordings of mouse and salamander retinas: a mouse preparation of 41 ganglion cells and two salamander preparations totaling 49 ganglion cells, each with only a few thousand stimulus-response pairs, two orders of magnitude below the scale at which transformers are typically trained. Against CNN and Gaussian-process baselines, CiRF-ViT reaches the highest mean explained variance on both datasets (0.94 on mouse, 0.95 on salamander). Probed with the local spike-triggered average (LSTA), a zero-shot test of context-dependent, nonlinear stimulus sensitivity, CiRF-ViT reproduces the qualitative polarity inversion that a linear model cannot capture by construction. A biologically grounded, per-neuron locality prior is therefore enough to make transformers competitive for neural prediction well below their usual data scale, while matching the reference models on an established functional signature of the retina's nonlinear computation.

neuroscience↗

bioRxiv2026

Viability of engineered AAVs via protein language models

Capsid engineering has greatly improved the performance of recombinant AAV vectors used for gene therapy. One commonly used strategy is the insertion of a short, 7-mer, peptide into surface-exposed loops to modify receptor interactions and enhance cell entry. While effective in receptor retargeting and improved transduction, these insertions might destabilize the capsid protein, hinder assembly, and thus limit production. While previous attempts have used deep mutational scanning and AI to predict which insertions are viable, there is lack in understanding the structural consequences of these peptide insertions at the amino-acid level. Here we combined experiments, deep sequencing and large protein language models to gain insight on the impact of 7-mer insertions on the VR-VIII region. We first characterize the biochemical properties of viable insertions, thus identifying which residues are well tolerated, and which should instead be avoided. We then focus on the nearby context of those insertions, by studying the effect of the linkers, either for highly diverse libraries or for individual variants known for their efficiency. Next, we study the broader context, by extending our analysis to the whole capsid sequence, and identifying regions that can tolerate insertions without long-ranged structural deformations that could affect capsid functionality. We conclude with a cross-serotype comparison and a viability analysis of tens of previously engineered variants. Our work showcases how AI can uncover structure-function rules governing the success of engineered AAV capsids.

bioinformatics↗

bioRxiv2026

Promfusion: a synthetic fusion promoter enabling enhanced and balanced photoreceptor transgene expression

Achieving efficient and balanced transgene expression in both rods and cones remains a major challenge in retinal gene therapy. Current promoters either lack specificity or fail to provide sufficient cellular coverage and expression level. To address this limitation, we developed and evaluated two fusion promoters, Pikali and Nocchu, by combining PR1.7, a cone-specific promoter and GRK1, a promoter most active in rods. Here, we show that Pikali and Nocchu outperform their parental promoters, driving broader and more balanced GFP expression in rods and cones of human iPSC-derived retinal organoids. These constructs achieved transduction in 30% to 45% of photoreceptors, with higher expression levels than GRK1 and broader cellular coverage than PR1.7. Our findings establish Pikali and Nocchu as excellent candidates for retinal gene therapy, overcoming the limitations of existing promoters. By combining specificity, efficiency, and extensive photoreceptor targeting, these fusion constructs represent a novel and promising strategy for next-generation gene therapy vectors, addressing inherited retinal dystrophies and advancing clinical translation.

genetics↗

bioRxiv2026

VIP+ amacrine cells synchronize neural activity in the retina

In early sensory circuits, inhibitory interneurons are best known for mediating lateral inhibition, gain control, and feature selectivity, but their roles at the population level are less understood. Using two-photon digital holography, we found that in the retina, vasoactive intestinal peptide-expressing (VIP+) amacrine cells not only inhibit some retinal ganglion cell types via GABAergic synapses but also excite specific other types through gap junctions. Selective knockout of these junctions abolished synchronized firing among these ganglion cells, showing that VIP+ amacrine cells coordinate their activity. Thus, VIP+ interneurons have a dual, cell-type-specific role: synchronizing some ganglion cells electrically while inhibiting others chemically, thereby differentially shaping retinal output.

neuroscience↗

bioRxiv2025

Stimulus sensitivity in noisy neural systems

Understanding how neural populations encode sensory information requires a precise definition of neuronal sensitivity to stimuli. While tuning curves and firing rates offer intuitive insights, theoretical frameworks based on signal-to-noise ratio and decoding efficiency identify Fisher information as the canonical measure of sensitivity, due to its relation with decoding performance. However, this relation holds only under restrictive conditions--when many neurons encode the same stimulus feature or when neural noise is weak. In realistic settings, such as when complex or high-dimensional stimuli are represented by a small ensemble of neurons, Fisher information becomes ill-defined or misleading. To overcome these limitations, we investigate two complementary information-theoretic quantities--the stimulus-specific information (ISSI) and the local information (Iloc)--and propose them as robust alternatives for quantifying sensitivity. We show that ISSI and Iloc converge with Fisher information when signal to noise is large, yet remain meaningful and interpretable beyond that regime. Importantly, these measures capture distinct aspects of sensitivity: ISSI quantifies how observing a response reduces stimulus uncertainty, whereas Iloc reflects how small stimulus perturbations reshape the systems posterior beliefs. Together, they offer a unifying perspective linking information-theoretic and statistical notions of sensitivity, bridging theoretical analysis and experimental investigation of neural coding.

neuroscience↗

bioRxiv2025

Direction-selective retinal ganglion cells encode motion direction uniformly, despite having discretely distributed cardinal preferences

On-Off direction-selective retinal ganglion cells (DS RGCs) exhibit broad tuning curves, responding robustly to motion aligned with or near their preferred direction. These cells comprise four major subtypes, each tuned to motion along one of the four cardinal axes: nasal, superior, temporal, or inferior. However, natural stimuli can move in any direction, and it remains unclear whether intermediate directions are encoded less effectively, or whether this cardinal organization nevertheless supports uniform direction encoding. Here, we combined previous electrophysiological recordings with an information-theoretic measure, the Stimulus Specific Information, to estimate the directional sensitivity of small populations of recorded neurons. This analysis revealed that DS RGC populations are uniformly sensitive across all directions of motion. We then asked whether the observed homogeneous sensitivity was a consequence of DS RGCs maximizing the average stimulus information. Simulations with artificially modified tuning curve widths revealed that DS cells prioritize avoiding pronounced drops in sensitivity over maximizing the average transmitted Information. Maximizing this minimal sensitivity may therefore be a principle to understand the organization of sensory systems.

neuroscience↗

bioRxiv2025

Accurate spatiotemporal retinal responses require a color intensity balance fine-tuned to natural conditions

Color vision is vital for animal survival, essential for foraging and predator detection. In mice, as in other mammals, color vision originates in the retina, where photoreceptor signals are processed by neural circuits. However, retinal responses to stimuli involving multiple colors are still not well understood. One possible explanation of this knowledge gap is that previous studies have not thoroughly examined how neuronal activity adapts to a 30 seconds to a few minutes timescale when exposed to multiple color sources. To address this, we systematically varied the UV-to-green light balance with a custom-built stimulator targeting mice opsins spectra while recording retinal ganglion cell responses across the dorso-ventral axis of the retina using multielectrode arrays. Responses to full-field chirp and checkerboard stimulations with alternating UV and green light revealed that more than one order of magnitude of intensity difference favoring green M-opsin over UV S-opsin is needed for a balanced reliability in retinal ganglion cell responses in the ventral retina. An incorrect balance, with slightly increased UV light, silenced responses to green illumination. To determine if these values are consistent with natural conditions, we analyzed isomerisation rates in the mouse retina across different times of the day. We found that the M- to S-opsin activation ratio remains constant through the mesopic-photopic range, and that our empirically determined values in the ventral retina align well with these natural conditions. These lie far from a simple equalization of M- and S-opsin isomerisation rates, which we found only balances ganglion cell responses in the dorsal retina. In conclusion, a finely tuned color intensity balance matching natural light spectrum is essential for accurately measuring both fast temporal responses and detailed spatial receptive fields in the ventral retina.

neuroscience↗

bioRxiv2024

The Role of Thermal Stability in AAV Titration of Engineered Variants

Determining the concentration of recombinant adeno-associated virus (AAV) productions, also known as titering, is crucial not only for quality control purposes but also for comparative studies of preclinical and clinical gene therapy trials. Recently, several AAVs were engineered by inserting seven amino acids at the outermost tip of the capsids protruding VR-VIII loop. These variants have demonstrated increased transduction capabilities over naturally occurring AAV serotypes in several studies. However, they have also been shown to produce lower yields when titered using standard techniques, raising questions about their adequacy for clinical development and use. Here, we investigated why peptide insertion onto AAV capsids reduces their titer by examining viral stocks using electron microscopy and PCR-based titering. We reveal that the DNAse digestion step, performed to eliminate free-floating DNA prior to qPCR or ddPCR, adversely impacts engineered capsid stability due to exposure to heat, artificially lowering viral titers of engineered serotypes. Titering without heating yields significantly higher titers for these variants which have melting temperatures (Tm) close to the DNAse inactivation temperature, while titers for parental serotypes with higher Tm remain unchanged. Our findings provide an important new perspective for titering engineered variants with lower thermostability, especially when comparing their effectiveness to their parental serotypes.

molecular biology↗

bioRxiv2024

Strong, but not weak, noise correlations are beneficial for population coding

Neural correlations play a critical role in sensory information coding. They are of two kinds: signal correlations, when neurons have overlapping sensitivities, and noise correlations from network effects and shared noise. It is commonly thought that stimulus and noise correlations should have opposite signs to improve coding. However, experiments from early sensory systems and cortex typically show the opposite effect, with many pairs of neurons showing both types of correlations to be positive and large. Here, we develop a theory of information coding by correlated neurons which resolves this paradox. We show that noise correlations are always beneficial if they are strong enough. Extensive tests on retinal recordings under different visual stimuli confirm our predictions. Finally, using neuronal recordings and modeling, we show that for high dimensional stimuli noise correlation benefits the encoding of fine-grained details of visual stimuli, at the expense of large-scale features, which are already well encoded.

neuroscience↗

bioRxiv2024

Optimal sequencing depth for measuring the concentrations of molecular barcodes

In combinatorial genetic engineering experiments, next-generation sequencing (NGS) allows for measuring the concentrations of barcoded or mutated genes within highly diverse libraries. When designing and interpreting these experiments, sequencing depths are thus important parameters to take into account. Service providers follow established guidelines to determine NGS depth depending on the type of experiment, such as RNA sequencing or whole genome sequencing. However, guidelines specifically tailored for measuring barcode concentrations have not yet reached an accepted consensus. To address this issue, we combine the analysis of NGS datasets from barcoded libraries with a mathematical model taking into account the PCR amplification in library preparation. We demonstrate on several datasets that noise in the NGS counts increases with the sequencing depth; consequently, beyond certain limits, deeper sequencing does not improve the precision of measuring barcode concentrations. We propose, as rule of thumb, that the optimal sequencing depth should be about ten times the initial amount of barcoded DNA molecules before any amplification step.

bioinformatics↗

bioRxiv2023

In-silico monitoring of directed evolution convergence to unveil best performing variants with credibility score

Directed evolution (DE) is a versatile protein-engineering strategy, successfully applied to a range of proteins, including enzymes, antibodies, and viral vectors. However, DE can be time-consuming and costly, as it typically requires many rounds of selection to identify desired mutants. Next-generation sequencing allows monitoring of millions of variants during DE and can be leveraged to reduce the number of selection rounds. Unfortunately the noisy nature of the sequencing data impedes the estimation of the performance of individual variants. Here, we propose ACIDES that combines statistical inference and in-silico simulations to improve performance estimation in DE by providing accurate statistical scores. We tested ACIDES first on a novel random-peptide-insertion experiment and then on several public datasets from DE of viral vectors and phage-display. ACIDES allows experimentalists to reliably estimate variant performance on the fly and can aid protein engineering pipelines in a range of applications, including gene therapy.

bioinformatics↗

bioRxiv2022

A small-correlation expansion to quantify information in noisy sensory systems

Neural networks encode information through their collective spiking activity in response to external stimuli. This population response is noisy and strongly correlated, with complex interplay between correlations induced by the stimulus, and correlations caused by shared noise. Understanding how these correlations affect information transmission has so far been limited to pairs or small groups of neurons, because the curse of dimensionality impedes the evaluation of mutual information in larger populations. Here we develop a small-correlation expansion to compute the stimulus information carried by a large population of neurons, yielding interpretable analytical expressions in terms of the neurons firing rates and pairwise correlations. We validate the approximation on synthetic data and demonstrate its applicability to electrophysiological recordings in the vertebrate retina, allowing us to quantify the effects of noise correlations between neurons and of memory in single neurons.

neuroscience↗

bioRxiv2022

Time-dependent maximum entropy model for populations of retinal ganglion cells

The inverse Ising model is used in computational neuroscience to infer probability distributions of the synchronous activity of large neuronal populations. This method allows for finding the Boltzmann distribution with single neuron biases and pairwise interactions that maximizes the entropy and reproduces the empirical statistics of the recorded neuronal activity. Here we apply this strategy to large populations of retinal output neurons (ganglion cells) of different types, stimulated by multiple visual stimuli with their own statistics. The activity of retinal output neurons is driven by both the inputs from upstream neurons, which encode the visual information and reflect stimulus statistics, and the recurrent connections, which induce network effects. We first apply the standard inverse Ising model approach, and show that it accounts well for the systems collective behavior when the input visual stimulus has short-ranged spatial correlations, but fails for long-ranged ones. This happens because stimuli with long-ranged spatial correlations synchronize the activity of neurons over long distances. This effect cannot be accounted for by pairwise interactions, and so by the pairwise Ising model. To solve this issue, we apply a previously proposed framework that includes a temporal dependence in the single neurons biases to model how neurons are driven in time by the stimulus. Thanks to this addition, the stimulus effects are taken into account by the biases, and the pairwise interactions allow for characterizing the network effect in the population activity and for reproducing the structure of the recurrent functional connections in the retinal architecture. In particular, the inferred interactions are strong and positive only for nearby neurons of the same type. Inter-type connections are instead small and slightly negative. Therefore, the retinal architecture splits into weakly interacting subpopulations composed of strongly interacting neurons. Overall, this temporal framework fixes the problems of the standard, static, inverse Ising model and accounts for the systems collective behavior, for stimuli with either short or long-range correlations.

neuroscience↗

bioRxiv2021

Context-dependent selectivity to natural scenes in the retina

Retina ganglion cells extract specific features from natural scenes and send this information to the brain. In particular, they respond to local light increase (ON responses), and/or decrease (OFF). However, it is unclear if this ON-OFF selectivity, characterized with synthetic stimuli, is maintained when they are stimulated with natural scenes. Here we recorded the responses of ganglion cells of mice and axolotls to stimuli composed of natural images slightly perturbed by patterns of random noise to determine their selectivity during natural stimulation. The ON-OFF selectivity strongly depended on the natural image. A single ganglion cell can signal luminance increase for one natural image, and luminance decrease for another. Modeling and experiments showed that this was due to the non-linear combination of different pathways of the retinal circuit. Despite the versatility of the ON-OFF selectivity, a systematic analysis demonstrated that contrast was reliably encoded in these responses. Our perturbative approach thus uncovers the selectivity of retinal ganglion cells to more complex features than initially thought during natural scene stimulation.

neuroscience↗

bioRxiv2021

Functional ultrasound imaging of the spreading activity following optogenetic stimulation of the rat visual cortex

Optogenetic stimulation of the primary visual cortex (V1) is a promising therapy for sight restoration, but it remains unclear what total cerebral volume is activated after surface stimulation. In this study, we expressed the red-shifted opsin ChrimsonR in excitatory neurons within V1 in rats, and used the fine spatial resolution provided by functional ultrasound imaging (fUS) over the whole depth of the brain to investigate the brain response to focal surface stimulation. We observed optogenetic activation of a high proportion of the volume of V1. Extracellular recordings confirmed the neuronal origin of this activation. Moreover, neuronal responses were even located in deep layers under conditions of low irradiance, spreading to the LGN and V2, consistent with a normal visual information process. This study paves the way for the use of optogenetics for cortical therapies, and highlights the value of coupling fUS with optogenetics.

neuroscience↗

bioRxiv2020

A new inference approach for training shallow and deep generalized linear models of noisy interacting neurons

Generalized linear models are one of the most efficient paradigms for predicting the correlated stochastic activity of neuronal networks in response to external stimuli, with applications in many brain areas. However, when dealing with complex stimuli, the inferred coupling parameters often do not generalize across different stimulus statistics, leading to degraded performance and blowup instabilities. Here, we develop a two-step inference strategy that allows us to train robust generalized linear models of interacting neurons, by explicitly separating the effects of correlations in the stimulus from network interactions in each training step. Applying this approach to the responses of retinal ganglion cells to complex visual stimuli, we show that, compared to classical methods, the models trained in this way exhibit improved performance, are more stable, yield robust interaction networks, and generalize well across complex visual statistics. The method can be extended to deep convolutional neural networks, leading to models with high predictive accuracy for both the neuron firing rates and their correlations.

neuroscience↗

bioRxiv2019

From serial to parallel: predicting synchronous firing of large neural populations from sequential recordings

A major goal in neuroscience is to understand how populations of neurons code for stimuli or actions. While the number of neurons that can be recorded simultaneously is increasing at a fast pace, in most cases these recordings cannot access a complete population: some neurons that carry relevant information remain unrecorded. In particular, it is hard to simultaneously record all the neurons of the same type in a given area. Recent progress has made possible to determine the type of each recorded neuron in a given area thanks to genetic and physiological tools. However, it is unclear how to infer the activity of a full population of neurons of the same type from sequential recordings across different experiments. Neural networks exhibit collective behaviour, e.g. noise correlations and synchronous activity, that are not directly captured by a conditionally-independent model that would just pool together the spike trains from sequential recordings. Here we present a method to build population activity from single cell responses taken from sequential recordings, which only requires pairwise recordings to train the model. Our method combines copula distributions and maximum entropy modeling. After training, the model allows us to predict the activity of large populations using only sequential recordings of single cells. We applied this method to a population of ganglion cells, the retinal output, all belonging to the same type. From just the spiking response of each cell to a repeated stimulus, we could predict the full activity of the population. We could then generalize to predict the population responses to different stimuli and even to different experiments. As a result, we were able to use our approach to construct a synthetic model of a very large neuronal population, which uses data combined from multiple experiments. We then predicted the extent of synchronous activity and showed it grew with the number of neurons. This approach is a promising way to infer population activity from sequential recordings in sensory areas.

neuroscience↗