Simple Behavioral Analysis (SimBA) as a platform for explainable machine learning in behavioral neuroscience.

The study of complex behaviors is often challenging when using manual annotation due to the absence of quantifiable behavioral definitions and the subjective nature of behavioral annotation. Integration of supervised machine learning approaches mitigates some of these issues through the inclusion of accessible and explainable model interpretation. To decrease barriers to access, and with an emphasis on accessible model explainability, we developed the open-source Simple Behavioral Analysis (SimBA) platform for behavioral neuroscientists. SimBA introduces several machine learning interpretability tools, including SHapley Additive exPlanation (SHAP) scores, that aid in creating explainable and transparent behavioral classifiers. Here we show how the addition of explainability metrics allows for quantifiable comparisons of aggressive social behavior across research groups and species, reconceptualizing behavior as a sharable reagent and providing an open-source framework. We provide an open-source, graphical user interface (GUI)-driven, well-documented package to facilitate the movement toward improved automation and sharing of behavioral classification tools across laboratories.

Lateral line ablation by ototoxic compounds results in distinct rheotaxis profiles in larval zebrafish.

The zebrafish lateral line is an established model for hair cell organ damage, yet few studies link mechanistic disruptions to changes in biologically relevant behavior. We used larval zebrafish to determine how damage via ototoxic compounds impact rheotaxis. Larvae were treated with CuSO4 or neomycin to disrupt lateral line function then exposed to water flow stimuli. Their swimming behavior was recorded on video then DeepLabCut and SimBA software were used to track movements and classify rheotaxis behavior, respectively. Lateral line-disrupted fish performed rheotaxis, but they swam greater distances, for shorter durations, and with greater angular variance than controls. Furthermore, spectral decomposition analyses confirmed that lesioned fish exhibited ototoxic compound-specific behavioral profiles with distinct changes in the magnitude, frequency, and cross-correlation between fluctuations in linear and angular movements. Our observations demonstrate that lateral line input is needed for fish to hold their station in flow efficiently and reveals that commonly used lesion methods have unique effects on rheotaxis behavior.

Toward the explainability, transparency, and universality of machine learning for behavioral classification in neuroscience.

The use of rigorous ethological observation via machine learning techniques to understand brain function (computational neuroethology) is a rapidly growing approach that is poised to significantly change how behavioral neuroscience is commonly performed. With the development of open-source platforms for automated tracking and behavioral recognition, these approaches are now accessible to a wide array of neuroscientists despite variations in budget and computational experience. Importantly, this adoption has moved the field toward a common understanding of behavior and brain function through the removal of manual bias and the identification of previously unknown behavioral repertoires. Although less apparent, another consequence of this movement is the introduction of analytical tools that increase the explainabilty, transparency, and universality of the machine-based behavioral classifications both within and between research groups. Here, we focus on three main applications of such machine model explainabilty tools and metrics in the drive toward behavioral (i) standardization, (ii) specialization, and (iii) explainability. We provide a perspective on the use of explainability tools in computational neuroethology, and detail why this is a necessary next step in the expansion of the field. Specifically, as a possible solution in behavioral neuroscience, we propose the use of Shapley values via Shapley Additive Explanations (SHAP) as a diagnostic resource toward explainability of human annotation, as well as supervised and unsupervised behavioral machine learning analysis.

Rage Against the Machine: Advancing the study of aggression ethology via machine learning.

RATIONALE: Aggression, comorbid with neuropsychiatric disorders, exhibits with diverse clinical presentations and places a significant burden on patients, caregivers, and society. This diversity is observed because aggression is a complex behavior that can be ethologically demarcated as either appetitive (rewarding) or reactive (defensive), each with its own behavioral characteristics, functionality, and neural basis that may transition from adaptive to maladaptive depending on genetic and environmental factors. There has been a recent surge in the development of preclinical animal models for studying appetitive aggression-related behaviors and identifying the neural mechanisms guiding their progression and expression. However, adoption of these procedures is often impeded by the arduous task of manually scoring complex social interactions. Manual observations are generally susceptible to observer drift, long analysis times, and poor inter-rater reliability, and are further incompatible with the sampling frequencies required of modern neuroscience methods. OBJECTIVES: In this review, we discuss recent advances in the preclinical study of appetitive aggression in mice, paired with our perspective on the potential for machine learning techniques in producing automated, robust scoring of aggressive social behavior. We discuss critical considerations for implementing valid computer classifications within behavioral pharmacological studies. KEY RESULTS: Open-source automated classification platforms can match or exceed the performance of human observers while removing the confounds of observer drift, bias, and inter-rater reliability. Furthermore, unsupervised approaches can identify previously uncharacterized aggression-related behavioral repertoires in model species. DISCUSSION AND CONCLUSIONS: Advances in open-source computational approaches hold promise for overcoming current manual annotation caveats while also introducing and generalizing computational neuroethology to the greater behavioral neuroscience community. We propose that currently available open-source approaches are sufficient for overcoming the main limitations preventing wide adoption of machine learning within the context of preclinical aggression behavioral research.

Dorsal and ventral striatal dopamine D1 and D2 receptors differentially modulate distinct phases of serial visual reversal learning.

Impaired cognitive flexibility in visual reversal-learning tasks has been observed in a wide range of neurological and neuropsychiatric disorders. Although both human and animal studies have implicated striatal D2-like and D1-like receptors (D2R; D1R) in this form of flexibility, less is known about the contribution they make within distinct sub-regions of the striatum and the different phases of visual reversal learning. The present study investigated the involvement of D2R and D1R during the early (perseverative) phase of reversal learning as well as in the intermediate and late stages (new learning) after microinfusions of D2R and D1R antagonists into the nucleus accumbens core and shell (NAcC; NAcS), the anterior and posterior dorsomedial striatum (DMS) and the dorsolateral striatum (DLS) on a touchscreen visual serial reversal-learning task. Reversal learning was improved after dopamine receptor blockade in the nucleus accumbens; the D1R antagonist, SCH23390, in the NAcS and the D2R antagonist, raclopride, in the NAcC selectively reduced early, perseverative errors. In contrast, reversal learning was impaired by D2R antagonism, but not D1R antagonism, in the dorsal striatum: raclopride increased errors in the intermediate phase after DMS infusions, and increased errors across phases after DLS infusions. These findings indicate that D1R and D2R modulate different stages of reversal learning through effects localised to different sub-regions of the striatum. Thus, deficits in behavioral flexibility observed in disorders linked to dopamine perturbations may be attributable to specific D1R and D2R dysfunction in distinct striatal sub-regions.

Dopamine D2-like receptor stimulation blocks negative feedback in visual and spatial reversal learning in the rat: behavioural and computational evidence.

RATIONALE: Dopamine D2-like receptors (D2R) are important drug targets in schizophrenia and Parkinson's disease, but D2R ligands also cause cognitive inflexibility such as poor reversal learning. The specific role of D2R in reversal learning remains unclear. OBJECTIVES: We tested the hypotheses that D2R agonism impairs reversal learning by blocking negative feedback and that antagonism of D1-like receptors (D1R) impairs learning from positive feedback. METHODS: Male Lister Hooded rats were trained on a novel visual reversal learning task. Performance on "probe trials", during which the correct or incorrect stimulus was presented with a third, probabilistically rewarded (50% of trials) and therefore intermediate stimulus, revealed individual learning curves for the processes of positive and negative feedback. The effects of D2R and D1R agonists and antagonists were evaluated. A separate cohort was tested on a spatial probabilistic reversal learning (PRL) task after D2R agonism. Computational reinforcement learning modelling was applied to choice data from the PRL task to evaluate the contribution of latent factors. RESULTS: D2R agonism with quinpirole dose-dependently impaired both visual reversal and PRL. Analysis of the probe trials on the visual task revealed a complete blockade of learning from negative feedback at the 0.25 mg/kg dose, while learning from positive feedback was intact. Estimated parameters from the model that best described the PRL choice data revealed a steep and selective decrease in learning rate from losses. D1R antagonism had a transient effect on the positive probe trials. CONCLUSIONS: D2R stimulation impairs reversal learning by blocking the impact of negative feedback.

Continuous performance test impairment in a 22q11.2 microdeletion mouse model: improvement by amphetamine.

The 22q11.2 deletion syndrome (22q11.2DS) confers high risk of neurodevelopmental disorders such as schizophrenia and attention-deficit hyperactivity disorder. These disorders are associated with attentional impairment, the remediation of which is important for successful therapeutic intervention. We assessed a 22q11.2DS mouse model (Df(h22q11)/+) on a touchscreen rodent continuous performance test (rCPT) of attention and executive function that is analogous to human CPT procedures. Relative to wild-type littermates, Df(h22q11)/+ male mice showed impaired attentional performance as shown by decreased correct response ratio (hit rate) and a reduced ability to discriminate target stimuli from non-target stimuli (discrimination sensitivity, or d'). The Df(h22q11)/+ model exhibited decreased prefrontal cortical-hippocampal oscillatory synchrony within multiple frequency ranges during quiet wakefulness, which may represent a biomarker of cognitive dysfunction. The stimulant amphetamine (0-1.0 mg/kg, i.p.) dose-dependently improved d' in Df(h22q11)/+ mice whereas the highest dose of modafinil (40 mg/kg, i.p.) exacerbated their d' impairment. This is the first report to directly implicate attentional impairment in a 22q11.2DS mouse model, mirroring a key endophenotype of the human disorder. The capacity of the rCPT to detect performance impairments in the 22q11.2DS mouse model, and improvement following psychostimulant-treatment, highlights the utility and translational potential of the Df(h22q11)/+ model and this automated behavioral procedure.

Selective effects of 5-HT2C receptor modulation on performance of a novel valence-probe visual discrimination task and probabilistic reversal learning in mice.

RATIONALE: Dysregulation of the serotonin (5-HT) system is a pathophysiological component in major depressive disorder (MDD), a condition closely associated with abnormal emotional responsivity to positive and negative feedback. However, the precise mechanism through which 5-HT tone biases feedback responsivity remains unclear. 5-HT2C receptors (5-HT2CRs) are closely linked with aspects of depressive symptomatology, including abnormalities in reinforcement processes and response to stress. Thus, we aimed to determine the impact of 5-HT2CR function on response to feedback in biased reinforcement learning. METHODS: We used two touchscreen assays designed to assess the impact of positive and negative feedback on probabilistic reinforcement in mice, including a novel valence-probe visual discrimination (VPVD) and a probabilistic reversal learning procedure (PRL). Systemic administration of a 5-HT2CR agonist and antagonist resulted in selective changes in the balance of feedback sensitivity bias on these tasks. RESULTS: Specifically, on VPVD, SB 242084, the 5-HT2CR antagonist, impaired acquisition of a discrimination dependent on appropriate integration of positive and negative feedback. On PRL, SB 242084 at 1 mg/kg resulted in changes in behaviour consistent with reduced sensitivity to positive feedback. In contrast, WAY 163909, the 5-HT2CR agonist, resulted in changes associated with increased sensitivity to positive feedback and decreased sensitivity to negative feedback. CONCLUSIONS: These results suggest that 5-HT2CRs tightly regulate feedback sensitivity bias in mice with consequent effects on learning and cognitive flexibility and specify a framework for the influence of 5-HT2CRs on sensitivity to reinforcement.

MAM-E17 rat model impairments on a novel continuous performance task: effects of potential cognitive enhancing drugs.

RATIONALE: Impairments in attention and inhibitory control are endophenotypic markers of neuropsychiatric disorders such as schizophrenia and represent key targets for therapeutic management. Robust preclinical models and assays sensitive to clinically relevant treatments are crucial for improving cognitive enhancement strategies. OBJECTIVES: We assessed a rodent model with neural and behavioral features relevant to schizophrenia (gestational day 17 methylazoxymethanol acetate treatment (MAM-E17)) on a novel test of attention and executive function, and examined the impact of putative nootropic drugs. METHODS: MAM-E17 and sham control rats were trained on a novel touchscreen-based rodent continuous performance test (rCPT) designed to closely mimic the human CPT paradigm. Performance following acute, systemic treatment with an array of pharmacological compounds was investigated. RESULTS: Two cohorts of MAM-E17 rats were impaired on rCPT performance including deficits in sensitivity (d') and increased false alarm rates (FARs). Sulpiride (0-30 mg/kg) dose-dependently reduced elevated FAR in MAM-E17 rats whereas low-dose modafinil (8 mg/kg) only improved d' in sham controls. ABT-594 (5.9-19.4 µg/kg) and modafinil (64 mg/kg) showed expected stimulant-like effects, while LSN2463359 (5 mg/kg), RO493858 (10 mg/kg), atomoxetine (0.3-1 mg/kg), and sulpiride (30 mg/kg) showed expected suppressant effects on performance across all animals. Donepezil (0.1-1 mg/kg) showed near-significant enhancements in d', and EVP-6124 (0.3-3 mg/kg) exerted no effects in the rCPT paradigm. CONCLUSION: The MAM-E17 model exhibits robust and replicable impairments in rCPT performance that resemble attention and inhibitory control deficits seen in schizophrenia. Pharmacological profiles were highly consistent with known drug effects on cognition in preclinical and clinical studies. The rCPT is a sensitive and reliable tool with high translational potential for understanding the etiology and treatment of disorders affecting attention and executive dysfunction.

Persistent gating deficit and increased sensitivity to NMDA receptor antagonism after puberty in a new mouse model of the human 22q11.2 microdeletion syndrome: a study in male mice.

BACKGROUND: The hemizygous 22q11.2 microdeletion is a common copy number variant in humans. The deletion confers high risk for neurodevelopmental disorders, including autism and schizophrenia. Up to 41% of deletion carriers experience psychotic symptoms. METHODS: We present a new mouse model (Df(h22q11)/+) of the deletion syndrome (22q11.2DS) and report on, to our knowledge, the most comprehensive study undertaken to date in 22q11.2DS models. The study was conducted in male mice. RESULTS: We found elevated postpubertal N-methyl-D-aspartate (NMDA) receptor antagonist-induced hyperlocomotion, age-independent prepulse inhibition (PPI) deficits and increased acoustic startle response (ASR). The PPI deficit and increased ASR were resistant to antipsychotic treatment. The PPI deficit was not a consequence of impaired hearing measured by auditory brain stem responses. The Df(h22q11)/+ mice also displayed increased amplitude of loudness-dependent auditory evoked potentials. Prefrontal cortex and dorsal striatal elevations of the dopamine metabolite DOPAC and increased dorsal striatal expression of the AMPA receptor subunit GluR1 was found. The Df(h22q11)/+ mice did not deviate from wild-type mice in a wide range of other behavioural and biochemical assays. LIMITATIONS: The 22q11.2 microdeletion has incomplete penetrance in humans, and the severity of disease depends on the complete genetic makeup in concert with environmental factors. In order to obtain more marked phenotypes reflecting the severe conditions related to 22q11.2DS it is suggested to expose the Df(h22q11)/+ mice to environmental stressors that may unmask latent psychopathology. CONCLUSION: The Df(h22q11)/+ model will be a valuable tool for increasing our understanding of the etiology of schizophrenia and other psychiatric disorders associated with the 22q11DS.

A mouse model of the 15q13.3 microdeletion syndrome shows prefrontal neurophysiological dysfunctions and attentional impairment.

RATIONALE: A microdeletion at locus 15q13.3 is associated with high incidence rates of psychopathology, including schizophrenia. A mouse model of the 15q13.3 microdeletion syndrome has been generated (Df[h15q13]/+) with translational utility for modelling schizophrenia-like pathology. Among other deficits, schizophrenia is characterised by dysfunctions in prefrontal cortical (PFC) inhibitory circuitry and attention. OBJECTIVES: The objective of this study is to assess PFC-dependent functioning in the Df(h15q13)/+ mouse using electrophysiological, pharmacological, and behavioural assays. METHOD: Experiments 1-2 investigated baseline firing and auditory-evoked responses of PFC interneurons and pyramidal neurons. Experiment 3 measured pyramidal firing in response to intra-PFC GABAA receptor antagonism. Experiments 4-6 assessed PFC-dependent attentional functioning through the touchscreen 5-choice serial reaction time task (5-CSRTT). Experiments 7-12 assessed reversal learning, paired-associate learning, extinction learning, progressive ratio, trial-unique non-match to sample, and object recognition. RESULTS: In experiments 1-3, the Df(h15q13)/+ mouse showed reduced baseline firing rate of fast-spiking interneurons and in the ability of the GABAA receptor antagonist gabazine to increase the firing rate of pyramidal neurons. In assays of auditory-evoked responses, PFC interneurons in the Df(h15q13)/+ mouse had reduced detection amplitudes and increased detection latencies, while pyramidal neurons showed increased detection latencies. In experiments 4-6, the Df(h15q13)/+ mouse showed a stimulus duration-dependent decrease in percent accuracy in the 5-CSRTT. The impairment was insensitive to treatment with the partial α7nAChR agonist EVP-6124. The Df(h15q13)/+ mouse showed no cognitive impairments in experiments 7-12. CONCLUSION: The Df(h15q13)/+ mouse has multiple dysfunctions converging on disrupted PFC processing as measured by several independent assays of inhibitory transmission and attentional function.

The continuous performance test (rCPT) for mice: a novel operant touchscreen test of attentional function.

RATIONALE: Continuous performance tests (CPTs) are widely used to assess attentional processes in a variety of disorders including Alzheimer's disease and schizophrenia. Common human CPTs require discrimination of sequentially presented, visually patterned 'target' and 'non-target' stimuli at a single location. OBJECTIVES: The aims of this study were to evaluate the performance of three popular mouse strains on a novel rodent touchscreen test (rCPT) designed to be analogous to common human CPT variants and to investigate the effects of donepezil, a cholinesterase inhibitor and putative cognitive enhancer. METHODS: C57BL/6J, DBA/2J and CD1 mice (n = 15-16/strain) were trained to baseline performance using four rCPT training stages. Then, probe tests assessed the effects of parameter changes on task performance: stimulus size, duration, contrast, probability, inter-trial interval or inclusion of flanker distractors. rCPT performance was also evaluated following acute administration of donepezil (0-3 mg/kg, i.p.). RESULTS: C57BL/6J and DBA/2J mice showed similar acquisition rates and final baseline performance following rCPT training. On probe tests, rCPT performance of both strains was sensitive to alteration of visual and/or attentional demands (stimulus size, duration, contrast, rate, flanker distraction). Relative to C57BL/6J, DBA/2J mice exhibited (1) decreasing sensitivity (d') across the 45-min session, (2) reduced performance on probes where the appearance of stimuli or adjacent areas were changed (size, contrast, flanking distractors) and (3) larger dose- and stimulus duration-dependent changes in performance following donepezil administration. In contrast, CD1 mice failed to acquire rCPT (stage 3) and pairwise visual discrimination tasks. CONCLUSIONS: rCPT is a potentially useful translational tool for assessing attention in mice and for detecting the effects of nootropic drugs.

The rat's not for turning: Dissociating the psychological components of cognitive inflexibility.

Executive function is commonly assessed by assays of cognitive flexibility such as reversal learning and attentional set-shifting. Disrupted performance in these assays, apparent in many neuropsychiatric disorders, is frequently interpreted as inability to overcome prior associations with reward. However, non-rewarded or irrelevant associations may be of considerable importance in both discrimination learning and cognitive flexibility. Non-rewarded associations can have greater influence on choice behaviour than rewarded associations in discrimination learning. Pathology-related deficits in cognitive flexibility can produce selective disruptions to both the processing of irrelevant associations and associations with reward. Genetic and pharmacological animal models demonstrate that modulation of reversal learning may result from alterations in either rewarded or non-rewarded associations. Successful performance in assays of cognitive flexibility can therefore depend on a combination of rewarded, non-rewarded, and irrelevant associations derived from previous learning, accounting for some inconsistencies observed in the literature. Taking this combination into account may increase the validity of animal models and may also reveal pathology-specific differences in problem solving and executive function.

Dissociable effects of 5-HT2C receptor antagonism and genetic inactivation on perseverance and learned non-reward in an egocentric spatial reversal task.

Cognitive flexibility can be assessed in reversal learning tests, which are sensitive to modulation of 5-HT2C receptor (5-HT2CR) function. Successful performance in these tests depends on at least two dissociable cognitive mechanisms which may separately dissipate associations of previous positive and negative valence. The first is opposed by perseverance and the second by learned non-reward. The current experiments explored the effect of reducing function of the 5-HT2CR on the cognitive mechanisms underlying egocentric reversal learning in the mouse. Experiment 1 used the 5-HT2CR antagonist SB242084 (0.5 mg/kg) in a between-groups serial design and Experiment 2 used 5-HT2CR KO mice in a repeated measures design. Animals initially learned to discriminate between two egocentric turning directions, only one of which was food rewarded (denoted CS+, CS-), in a T- or Y-maze configuration. This was followed by three conditions; (1) Full reversal, where contingencies reversed; (2) Perseverance, where the previous CS+ became CS- and the previous CS- was replaced by a novel CS+; (3) Learned non-reward, where the previous CS- became CS+ and the previous CS+ was replaced by a novel CS-. SB242084 reduced perseverance, observed as a decrease in trials and incorrect responses to criterion, but increased learned non-reward, observed as an increase in trials to criterion. In contrast, 5-HT2CR KO mice showed increased perseverance. 5-HT2CR KO mice also showed retarded egocentric discrimination learning. Neither manipulation of 5-HT2CR function affected performance in the full reversal test. These results are unlikely to be accounted for by increased novelty attraction, as SB242084 failed to affect performance in an unrewarded novelty task. In conclusion, acute 5-HT2CR antagonism and constitutive loss of the 5-HT2CR have opposing effects on perseverance in egocentric reversal learning in mice. It is likely that this difference reflects the broader impact of 5HT2CR loss on the development and maintenance of cognitive function.

The touchscreen operant platform for testing learning and memory in rats and mice.

An increasingly popular method of assessing cognitive functions in rodents is the automated touchscreen platform, on which a number of different cognitive tests can be run in a manner very similar to touchscreen methods currently used to test human subjects. This methodology is low stress (using appetitive rather than aversive reinforcement), has high translational potential and lends itself to a high degree of standardization and throughput. Applications include the study of cognition in rodent models of psychiatric and neurodegenerative diseases (e.g., Alzheimer's disease, schizophrenia, Huntington's disease, frontotemporal dementia), as well as the characterization of the role of select brain regions, neurotransmitter systems and genes in rodents. This protocol describes how to perform four touchscreen assays of learning and memory: visual discrimination, object-location paired-associates learning, visuomotor conditional learning and autoshaping. It is accompanied by two further protocols (also published in this issue) that use the touchscreen platform to assess executive function, working memory and pattern separation.

The touchscreen operant platform for assessing executive function in rats and mice.

This protocol details a subset of assays developed within the touchscreen platform to measure various aspects of executive function in rodents. Three main procedures are included: extinction, measuring the rate and extent of curtailing a response that was previously, but is no longer, associated with reward; reversal learning, measuring the rate and extent of switching a response toward a visual stimulus that was previously not, but has become, associated with reward (and away from a visual stimulus that was previously, but is no longer, rewarded); and the 5-choice serial reaction time (5-CSRT) task, gauging the ability to selectively detect and appropriately respond to briefly presented, spatially unpredictable visual stimuli. These protocols were designed to assess both complementary and overlapping constructs including selective and divided visual attention, inhibitory control, flexibility, impulsivity and compulsivity. The procedures comprise part of a wider touchscreen test battery assessing cognition in rodents with high potential for translation to human studies.