GDspike: An accurate spike estimation algorithm from noisy calcium fluorescence signals
07 Mar 2017-pp 1043-1047
TL;DR: A non-model-based approach for spike train inference using group delay (GD) analysis, inspired by GD-based high-resolution processing of the Ca2+ fluorescence signal, which is found to be unaffected when tested with five different GCaMP indicators and scanning rate varying from 15Hz to 60Hz.
Abstract: Accurate estimation of spike train from calcium (Ca2+) fluorescence signals is challenging owing to significant fluctuations of fluorescence level. This paper proposes a non-model-based approach for spike train inference using group delay (GD) analysis. It primarily exploits the property that change in Ca2+ fluorescence corresponding to a spike has a notable onset location followed by a decaying transient. The proposed algorithm, GDspike, is compared with state-of-the-art systems on five datasets. F-measure is best for GDspike (41%) followed by STM (40%), MLspike (39%), and Vogelstein (35%). While existing methods are inspired by the physiology of neuronal responses, the proposed approach is inspired by GD-based high-resolution processing of the Ca2+ fluorescence signal. GDspike is a fast and unsupervised algorithm. It is found to be unaffected when tested with five different GCaMP indicators and scanning rate varying from 15Hz to 60Hz.
Citations
More filters
TL;DR: This work compared multiple approaches using multiple datasets with ground truth electrophysiology and found that simple non-negative deconvolution (NND) outperformed all other algorithms on out-of-sample test data, and recommended that spikes be inferred from calcium traces using simple NND because of its simplicity, efficiency, and accuracy.
Abstract: Calcium imaging is a powerful method to record the activity of neural populations in many species, but inferring spike times from calcium signals is a challenging problem. We compared multiple approaches using multiple datasets with ground truth electrophysiology and found that simple non-negative deconvolution (NND) outperformed all other algorithms on out-of-sample test data. We introduce a novel benchmark applicable to recordings without electrophysiological ground truth, based on the correlation of responses to two stimulus repeats, and used this to show that unconstrained NND also outperformed the other algorithms when run on "zoomed out" datasets of ∼10,000 cell recordings from the visual cortex of mice of either sex. Finally, we show that NND-based methods match the performance of a supervised method based on convolutional neural networks while avoiding some of the biases of such methods, and at much faster running times. We therefore recommend that spikes be inferred from calcium traces using simple NND because of its simplicity, efficiency, and accuracy.SIGNIFICANCE STATEMENT The experimental method that currently allows for recordings of the largest numbers of cells simultaneously is two-photon calcium imaging. However, use of this powerful method requires that neuronal firing times be inferred correctly from the large resulting datasets. Previous studies have claimed that complex supervised learning algorithms outperform simple deconvolution methods at this task. Unfortunately, these studies suffered from several problems and biases. When we repeated the analysis, using the same data and correcting these problems, we found that simpler spike inference methods perform better. Even more importantly, we found that supervised learning methods can introduce artifactual structure into spike trains, which can in turn lead to erroneous scientific conclusions. Of the algorithms we evaluated, we found that an extremely simple method performed best in all circumstances tested, was much faster to run, and was insensitive to parameter choices, making incorrect scientific conclusions much less likely.
130 citations
Cites background or methods from "GDspike: An accurate spike estimati..."
...Simple NND outperforms the state-of-the-art results Many calcium deconvolution algorithms have recently been described (Vogelstein et al., 2010; Andilla and Hamprecht, 2014; Reynolds et al., 2015; Deneux et al., 2016; Theis et al., 2016; Friedrich et al., 2017; Jewell and Witten, 2017; Sebastian et al., 2017; Kazemipour et al., 2018), some of which have provided their code publicly....
[...]
...…data processing methods will be required to make optimal use of this activity (Vogelstein et al., 2010; Andilla and Hamprecht, 2014; Reynolds et al., 2015; Deneux et al., 2016; Theis et al., 2016; Friedrich et al., 2017; Jewell and Witten, 2017; Sebastian et al., 2017; Kazemipour et al., 2018)....
[...]
...…have recently been described (Vogelstein et al., 2010; Andilla and Hamprecht, 2014; Reynolds et al., 2015; Deneux et al., 2016; Theis et al., 2016; Friedrich et al., 2017; Jewell and Witten, 2017; Sebastian et al., 2017; Kazemipour et al., 2018), some of which have provided their code publicly....
[...]
TL;DR: This work introduces a novel benchmark applicable to recordings without electrophysiological ground truth, based on the correlation of responses to two stimulus repeats, and shows that unconstrained NND also outperformed the other algorithms when run on “zoomed out” datasets of ~10,000 cell recordings.
Abstract: Calcium imaging is a powerful method to record the activity of neural populations, but inferring spike times from calcium signals is a challenging problem. We compared multiple approaches using multiple datasets with ground truth electrophysiology, and found that simple non-negative deconvolution (NND) outperformed all other algorithms. We introduce a novel benchmark applicable to recordings without electrophysiological ground truth, based on the correlation of responses to two stimulus repeats, and used this to show that unconstrained NND also outperformed the other algorithms when run on 9zoomed out9 datasets of ~10,000 cell recordings. Finally, we show that NND-based methods match the performance of a supervised method based on convolutional neural networks, while avoiding some of the biases of such methods, and at much faster running times. We therefore recommend that spikes be inferred from calcium traces using simple NND, due to its simplicity, efficiency and accuracy.
23 citations
TL;DR: A hyperacuity algorithm (HA_time) based on an approach that combines a generative model and machine learning to improve spike detection and the precision of spike time inference is developed and is a useful tool for spike reconstruction from two-photon imaging.
Abstract: Two-photon imaging is a major recording technique used in neuroscience. However, it suffers from several limitations, including a low sampling rate, the nonlinearity of calcium responses, the slow dynamics of calcium dyes and a low SNR, all of which severely limit the potential of two-photon imaging to elucidate neuronal dynamics with high temporal resolution. We developed a hyperacuity algorithm (HA_time) based on an approach that combines a generative model and machine learning to improve spike detection and the precision of spike time inference. Bayesian inference was performed to estimate the calcium spike model, assuming constant spike shape and size. A support vector machine using this information and a jittering method maximizing the likelihood of estimated spike times enhanced spike time estimation precision approximately fourfold (range, 2-7; mean, 3.5-4.0; 2SEM, 0.1-0.25) compared to the sampling interval. Benchmark scores of HA_time for biological data from three different brain regions were among the best of the benchmark algorithms. Simulation of broader data conditions indicated that our algorithm performed better than others with high firing rate conditions. Furthermore, HA_time exhibited comparable performance for conditions with and without ground truths. Thus HA_time is a useful tool for spike reconstruction from two-photon imaging.
18 citations
TL;DR: In this paper, the role of different inhibitory interneurons (INs) in reliable coding in the primary visual cortex (V1) was investigated. But, the same neurons can also respond highly reliably.
Abstract: Intrinsic neuronal variability significantly limits information encoding in the primary visual cortex (V1). However, under certain conditions, neurons can respond reliably with highly precise responses to the same visual stimuli from trial to trial. This suggests that there exists intrinsic neural circuit mechanisms that dynamically modulate the intertrial variability of visual cortical neurons. Here, we sought to elucidate the role of different inhibitory interneurons (INs) in reliable coding in mouse V1. To study the interactions between somatostatin-expressing interneurons (SST-INs) and parvalbumin-expressing interneurons (PV-INs), we used a dual-color calcium imaging technique that allowed us to simultaneously monitor these two neural ensembles while awake mice, of both sexes, passively viewed natural movies. SST neurons were more active during epochs of reliable pyramidal neuron firing, whereas PV neurons were more active during epochs of unreliable firing. SST neuron activity lagged that of PV neurons, consistent with a feedback inhibitory SST→PV circuit. To dissect the role of this circuit in pyramidal neuron activity, we used temporally limited optogenetic activation and inactivation of SST and PV interneurons during periods of reliable and unreliable pyramidal cell firing. Transient firing of SST neurons increased pyramidal neuron reliability by actively suppressing PV neurons, a proposal that was supported by a rate-based model of V1 neurons. These results identify a cooperative functional role for the SST→PV circuit in modulating the reliability of pyramidal neuron activity.SIGNIFICANCE STATEMENT Cortical neurons often respond to identical sensory stimuli with large variability. However, under certain conditions, the same neurons can also respond highly reliably. The circuit mechanisms that contribute to this modulation remain unknown. Here, we used novel dual-wavelength calcium imaging and temporally selective optical perturbation to identify an inhibitory neural circuit in visual cortex that can modulate the reliability of pyramidal neurons to naturalistic visual stimuli. Our results, supported by computational models, suggest that somatostatin interneurons increase pyramidal neuron reliability by suppressing parvalbumin interneurons via the inhibitory SST→PV circuit. These findings reveal a novel role of the SST→PV circuit in modulating the fidelity of neural coding critical for visual perception.
14 citations
TL;DR: The proposed approach, GD spike, is compared with other spike estimation methods, including MLspike, Vogelstein de-convolution algorithm, and data-driven spike-triggered mixture model and shows superior results.
Abstract: Spike estimation from calcium (Ca $^{2+}$ ) fluorescence signals is a fundamental and challenging problem in neuroscience. Several models and algorithms have been proposed for this task over the past decade. Nevertheless, it is still hard to achieve accurate spike positions from the Ca $^{2+}$ fluorescence signals. While existing methods rely on data-driven methods and the physiology of neurons for modeling the spiking process, this paper exploits the nature of the fluorescence responses to spikes using signal processing. We first motivate the problem by a novel analysis of the high-resolution property of minimum-phase group delay (GD) functions for multi-pole resonators. The resonators could be connected either in series or in parallel. The Ca $^{2+}$ indicator responds to a spike with a sudden rise, that is followed by an exponential decay. We interpret the Ca $^{2+}$ signal as the response of an impulse train to the change in Ca $^{2+}$ concentration, where the Ca $^{2+}$ response corresponds to a resonator. We perform minimum-phase GD-based filtering of the Ca $^{2+}$ signal for resolving spike locations. The performance of the proposed algorithm is evaluated on nine datasets spanning various indicators, sampling rates, and mouse brain regions. The proposed approach, GDspike, is compared with other spike estimation methods, including MLspike, Vogelstein de-convolution algorithm, and data-driven spike-triggered mixture model. The performance of GDspike is superior to that of the Vogelstein algorithm and is comparable to that of MLspike. It can also be used to post-process the output of MLspike, which further enhances the performance.
7 citations
Cites methods from "GDspike: An accurate spike estimati..."
...In [20], we proposed a nonmodel based approach that is inspired by the high-resolving capability of the group delay function....
[...]
References
More filters
TL;DR: A family of ultrasensitive protein calcium sensors (GCaMP6) that outperformed other sensors in cultured neurons and in zebrafish, flies and mice in vivo are developed and provide new windows into the organization and dynamics of neural circuits over multiple spatial and temporal scales.
Abstract: Fluorescent calcium sensors are widely used to image neural activity. Using structure-based mutagenesis and neuron-based screening, we developed a family of ultrasensitive protein calcium sensors (GCaMP6) that outperformed other sensors in cultured neurons and in zebrafish, flies and mice in vivo. In layer 2/3 pyramidal neurons of the mouse visual cortex, GCaMP6 reliably detected single action potentials in neuronal somata and orientation-tuned synaptic calcium transients in individual dendritic spines. The orientation tuning of structurally persistent spines was largely stable over timescales of weeks. Orientation tuning averaged across spine populations predicted the tuning of their parent cell. Although the somata of GABAergic neurons showed little orientation tuning, their dendrites included highly tuned dendritic segments (5-40-µm long). GCaMP6 sensors thus provide new windows into the organization and dynamics of neural circuits over multiple spatial and temporal scales.
5,365 citations
"GDspike: An accurate spike estimati..." refers background or methods in this paper
...1 9 GCaMP5k 50 2735 [21] 2 11 GCaMP6f 60 4536 [20] 3 9 GCaMP6s 60 2123 [20] 4 11 jRGECO1a 25 9080 [22] 5 10 jRCaMP1a 15 3624 [22]...
[...]
...Figure 3(a) shows a segment of normalized fluorescence signal taken from the 1 neuron of GCaMP6f dataset [20]....
[...]
...GDspike is evaluated the publicly available dataset provided by Svoboda lab, at Janelia Research Campus1 ([21, 20, 22])....
[...]
Howard Hughes Medical Institute1, University of California, Davis2, Champalimaud Foundation3, University of Puerto Rico4, Medical Research Council5, Yale University6, Princeton University7, Rockefeller University8, Harvard University9, University of California, San Francisco10, Tata Institute of Fundamental Research11, University of California, Los Angeles12
TL;DR: GCaMP5 fluorescence provides a more reliable measure of neuronal activity than its predecessor GCaMP3, which allows more sensitive detection of neural activity in vivo and may find widespread applications for cellular imaging in general.
Abstract: Genetically encoded calcium indicators (GECIs) are powerful tools for systems neuroscience. Recent efforts in protein engineering have significantly increased the performance of GECIs. The state-of-the art single-wavelength GECI, GCaMP3, has been deployed in a number of model organisms and can reliably detect three or more action potentials in short bursts in several systems in vivo. Through protein structure determination, targeted mutagenesis, high-throughput screening, and a battery of in vitro assays, we have increased the dynamic range of GCaMP3 by severalfold, creating a family of “GCaMP5” sensors. We tested GCaMP5s in several systems: cultured neurons and astrocytes, mouse retina, and in vivo in Caenorhabditis chemosensory neurons, Drosophila larval neuromuscular junction and adult antennal lobe, zebrafish retina and tectum, and mouse visual cortex. Signal-to-noise ratio was improved by at least 2- to 3-fold. In the visual cortex, two GCaMP5 variants detected twice as many visual stimulus-responsive cells as GCaMP3. By combining in vivo imaging with electrophysiology we show that GCaMP5 fluorescence provides a more reliable measure of neuronal activity than its predecessor GCaMP3. GCaMP5 allows more sensitive detection of neural activity in vivo and may find widespread applications for cellular imaging in general.
1,179 citations
"GDspike: An accurate spike estimati..." refers methods in this paper
...1 9 GCaMP5k 50 2735 [21] 2 11 GCaMP6f 60 4536 [20] 3 9 GCaMP6s 60 2123 [20] 4 11 jRGECO1a 25 9080 [22] 5 10 jRCaMP1a 15 3624 [22]...
[...]
...GDspike is evaluated the publicly available dataset provided by Svoboda lab, at Janelia Research Campus1 ([21, 20, 22])....
[...]
TL;DR: Improved red GECIs based on mRuby (jRCaMP1a, b) and mApple (jRGECO1a) are presented, with sensitivity comparable to GCaMP6, to facilitate deep-tissue imaging, dual-color imaging together with GFP-based reporters, and the use of optogenetics in combination with calcium imaging.
Abstract: Neurons encode information with brief electrical pulses called spikes. Monitoring spikes in large populations of neurons is a powerful method for studying how networks of neurons process information and produce behavior. This activity can be detected using fluorescent protein indicators, or “probes”, which light up when neurons are active. The best existing probes produce green fluorescence. However, red fluorescent probes would allow us to see deeper into the brain, and could also be used with green probes to image the activity and interactions of different neuron types simultaneously. However, existing red fluorescent probes are not as good at detecting neural activity as green probes. By optimizing two existing red fluorescent proteins, Dana et al. have now produced two new red fluorescent probes, each with different advantages. The new protein indicators detect neural activity with high sensitivity and allow researchers to image previously unseen brain activity. Tests showed that the probes work in cultured neurons and allow imaging of the activity of neurons in mice, flies, fish and worms. History has shown that enhancing the techniques used to study biological processes can lead to fundamentally new insights. In the future, Dana et al. would therefore like to make even more sensitive protein indicators that will allow larger networks of neurons deeper in the brain to be imaged.
762 citations
"GDspike: An accurate spike estimati..." refers methods in this paper
...1 9 GCaMP5k 50 2735 [21] 2 11 GCaMP6f 60 4536 [20] 3 9 GCaMP6s 60 2123 [20] 4 11 jRGECO1a 25 9080 [22] 5 10 jRCaMP1a 15 3624 [22]...
[...]
...GDspike is evaluated the publicly available dataset provided by Svoboda lab, at Janelia Research Campus1 ([21, 20, 22])....
[...]
TL;DR: To examine the nature and precision of temporal coding, individual responses elicited by each set of stimuli were compared in terms of two families of metrics, which provided a possible mechanism for the simultaneous representation of multiple stimulus attributes in one spike train.
Abstract: 1. We recorded single-unit and multi-unit activity in response to transient presentation of texture and grating patterns at 25 sites within the parafoveal representation of V1, V2, and V3 of two awake monkeys trained to perform a fixation task. In grating experiments, stimuli varied in orientation, spatial frequency, or both. In texture experiments, stimuli varied in contrast, check size, texture type, or pairs of these attributes. 2. To examine the nature and precision of temporal coding, we compared individual responses elicited by each set of stimuli in terms of two families of metrics. One family of metrics, D(spike), was sensitive to the absolute spike time (following stimulus onset). The second family of metrics, D(interval), was sensitive to the pattern of interspike intervals. In each family, the metrics depend on a parameter q, which expresses the precision of temporal coding. For q = 0, both metrics collapse into the "spike count" metric D(Count), which is sensitive to the number of impulses but insensitive to their position in time. 3. Each of these metrics, with values of q ranging from 0 to 512/s, was used to calculate the distance between all pairs of spike trains within each dataset. The extent of stimulus-specific clustering manifest in these pairwise distances was quantified by an information measure. Chance clustering was estimated by applying the same procedure to synthetic data sets in which responses were assigned randomly to the input stimuli. 4. Of the 352 data sets, 170 showed evidence of tuning via the spike count (q = 0) metric, 294 showed evidence of tuning via the spike time metric, 272 showed evidence of tuning via the spike interval metric to the stimulus attribute (contrast, check size, orientation, spatial frequency, or texture type) under study. Across the entire dataset, the information not attributable to chance clustering averaged 0.042 bits for the spike count metric, 0.171 bits for the optimal spike time metric, and 0.107 bits for the optimal spike interval metric. 5. The reciprocal of the optimal cost q serves as a measure of the temporal precision of temporal coding. In V1 and V2, with both metrics, temporal precision was highest for contrast (ca. 10-30 ms) and lowest for texture type (ca. 100 ms). This systematic dependence of q on stimulus attribute provides a possible mechanism for the simultaneous representation of multiple stimulus attributes in one spike train. 6. Our findings are inconsistent with Poisson models of spike trains. Synthetic data sets in which firing rate was governed by a time-dependent Poisson process matched to the observed poststimulus time histogram (PSTH) overestimated clustering induced by D(count) and, for low values of q, D(spike)[q] and D(intervals)[q]. Synthetic data sets constructed from a modified Poisson process, which preserved not only the PSTH but also spike count statistics accounted for the clustering induced by D(count) but underestimated the clustering induced by D(spike)[q] and D(interval)[q].
639 citations
"GDspike: An accurate spike estimati..." refers methods in this paper
...The distance between the actual (ground truth) and estimated spikes are computed using a dynamic programming algorithm [23] which penalises the distance for spike deletions, insertions, and shifts....
[...]
TL;DR: In vivo imaging in mouse neocortex is reported with greatly improved temporal resolution using random-access scanning with acousto-optic deflectors, uncovering spatiotemporal trial-to-trial variability of sensory responses in barrel cortex and visual cortex.
Abstract: Two-photon calcium imaging of neuronal populations enables optical recording of spiking activity in living animals, but standard laser scanners are too slow to accurately determine spike times. Here we report in vivo imaging in mouse neocortex with greatly improved temporal resolution using random-access scanning with acousto-optic deflectors. We obtained fluorescence measurements from 34-91 layer 2/3 neurons at a 180-490 Hz sampling rate. We detected single action potential-evoked calcium transients with signal-to-noise ratios of 2-5 and determined spike times with near-millisecond precision and 5-15 ms confidence intervals. An automated 'peeling' algorithm enabled reconstruction of complex spike trains from fluorescence traces up to 20-30 Hz frequency, uncovering spatiotemporal trial-to-trial variability of sensory responses in barrel cortex and visual cortex. By revealing spike sequences in neuronal populations on a fast time scale, high-speed calcium imaging will facilitate optical studies of information processing in brain microcircuits.
471 citations