Interactive Psychometrics for Autism with the Human Dynamic Clamp: Interpersonal Synchrony from Sensory-motor to Socio-cognitive Domains

Background: The Human Dynamic Clamp (HDC) is a human-machine interface for studying realistic social interaction under controlled and reproducible conditions. Here, we propose to probe the validity of the HDC as psychometric instrument for quantifying social abilities in children with Autism Spectrum Disorder (ASDs) and neurotypical development. Methods: To study behavioral synchrony, we derived from interaction with the HDC avatar, five standardized scores following a gradient from sensory-motor and motor to higher socio-cognitive skills, in a sample of 155 individuals (113 with ASDs, 42 typically developing participants; aged 5 to 25 years; IQ>70). Results: We conducted regression analyses with normative modeling on global scores according to four sub-conditions (avatar behavior "cooperative/competitive", human task "in-phase/anti-phase", diagnosis and age at inclusion). Children with ASDs tend to have significant lower scores than controls for motor skills. Independently of the phenotype, socio-cognitive skills increase with developmental age, while being affected by the ongoing task and the behavior of the avatar. Discussion: The weaker performance in participants with ASDs for motor skills suggests convergent validity for this score of the HDC during social interaction. Results provide additional evidence of a relationship between sensory-motor and socio-cognitive skills. As we found a significant main effect of age at inclusion, HDC may be used as a marker of aging of socio-cognitive skills during real-time social interaction. Conclusion: Through its standardized and objective evaluation, the HDC not only represents a valid paradigm for the study of interpersonal synchrony but also a clinically relevant psychometric instrument for the evaluation and stratification of socio-motor dysfunctions. of emotions, suggesting that greater cognitive abilities correlates with a higher levels of affect recognition skills. Results attest of lower motor skills among participants with higher probability


Autism Spectrum Disorders (ASDs) are a family of complex neurodevelopmental disorders
American Psychiatric Association (APA) (2013) which are defined by the co-occurrence of significant differences in the development of social communication and interaction, and the restricted and repetitive nature of behaviors and interests. The prevalence of these disorders has increased in recent years from less than 1 in 1000 individuals to 1 in 58 (Baxter et al. 2015;Fombonne 2003). The strong heterogeneity of the conditions complicates the development of psychometric assessment tools that allow for a personalized and thorough evaluation of a child's skills (Vivanti et al. 2014). Indeed, ASDs are highly diverse at both phenotypical (Georgiades et al. 2013) and genetic levels (Huguet, Ey, et Bourgeron 2013), with more than 50% of the patients suffering from at least four other psychiatric comorbid conditions (Soke et al. 2018). The major societal challenge of improving early diagnosis thus requires the discovery of robust and scalable biomarkers related to social communication dysfunction in humans.
An important key for increasing our comprehension and early detection of ASD-specific social dysfunction may lie on the Interpersonal Synchrony (IS). IS can be defined as a rhythmic matching of actions in time and in phase with another person, based on non-verbal behaviors (Mogan et al. 2017). IS comprises multiple components, such as socio-cognitive, sensory-motor, and motor skills, as well as adaptive capacities (Nebel et  The Human Dynamic Clamp (HDC) is a new paradigm of human-machine interaction for the study of neurobehavioural processes involved in IS (Dumas et al. 2014; Kelso et al. 2009). It allows to recreate a dynamic bidirectional interaction in real time between a human and a virtual avatar, controlled by empirical grounded models. The HDC paradigm has already been validated in an adult with a neurotypical development through the study of underlying neural processes linking sensory-motor and socio-cognitive processes (Dumas et al. 2019), induction of emotional reaction (Zhang et al. 2016), and in a Virtual Teacher (VT) configuration of HDC which allows to change the behavioral repertoire by internalizing new interpersonal coordination patterns, thus opening HDC to rehabilitation (Kostrubiec et al. 2015).
IS seems to be substantively impaired in children and adolescents with ASDs (Xavier et al. 2018).
A few studies among children (6-11y) (Romero et al. 2018) and adolescents (10-16. One hypothesis currently under investigation suggests that motor and sensory-motor skills development are linked to social cognition and cognitive development (Bhat et al. 2016;Kaur et al., 2018). ASDs are frequently found to be associated with difficulties to attribute mental states to oneself and to others (Zalla et Korman 2018), where intention attribution is characterized by an appraisal based on intention underlying someone else's action (Hilton et  ASDs. This observation reflects the difficulty of assessments by reliable and scalable markers of IS over several ages and the need for a personalized analysis.
In this work, we first want to validate in children with neurotypical development how the HDC can measure different behavioural mechanisms involved in social dynamic interactions, and then to evaluate how the HDC can assess the alteration of IS in ASDs. The secondary objective has required to standardize the test and develop indicators that measure and identify socio-cognitive and sensory-motor markers. Finally, to take into account the heterogeneity of ASDs, we also integrated the developmental trajectories with normative modeling and tried to test how the HDC behavioral measures can provide reproducible and reliable clinical markers.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. was not peer-reviewed) (which The copyright holder for this preprint . http://dx.doi.org/10.1101/19013771 doi: medRxiv preprint first posted online Dec. 6, 2019 ;

Sample
We enrolled in the study a sample of 155 individuals composed of 113 participants with ASDs and 42 participants with typical development (Table 1) to synchronized her/his movements to those of the partner or in anti-phase meaning that s/he has to synchronize her/his movements with a half-period offset between the phase of the partner and the participant. In this experiment, all participants were instructed that the partner will be half of the time virtual (i.e., movements will be computer driven) and half of the time a real sex-, agematched human performing the same task in another room of the hospital (although in this case, the partner was also virtual). The protocol was composed of 40 trials, divided into four blocks. The instructions to the participant stayed the same within each block. The instruction for the first block was randomly assigned at the beginning of the experiment. During the trials, the VP could adopt a "cooperative" or "competitive" behavior, meaning that its shares the same goal as the participant or has the opposite goal to the one assigned to the participant (i.e., VP aims at in-phase coordination when participant aims at anti-phase; and vice versa). Behavior of the VP was randomized across all trials, disregarding block structure.At the end of each trial, we asked the participant if s/he felt like s/he was playing with a human or a VP, and if s/he could quantify the level of cooperativeness to competitiveness of the partner.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. was not peer-reviewed) (which The copyright holder for this preprint . http://dx.doi.org/10.1101/19013771 doi: medRxiv preprint first posted online Dec. 6, 2019 ;

HDC behavioral measures
In our study, five scores (between 0 and 1, 0 being the worst) of the HDC paradigm were automatically generated from sensory-motor to representational dimensions of social cognition: i/ a motor score : which measure the difference of amplitude of imitative gestures between the participant and the VP. A mean is obtained for each trial.
where %&'() and (*(+(, are respectively the amplitude of the movement of the participant and the virtual avatar, both extracted from a Hilbert transform; ii/ a coordination score which corresponds to the temporal index of imitation. This measure is based on the frequency parameter and evaluate the shift of frequency between that of the model and that of the participant. A mean is obtained for each trial.
where %&'() and (*(+(, are respectively the phase of the movement of the participant and the VP, both extracted from a Hilbert transform, | |is the norm in the complex plane, and is the total number of samples; iii/ a task score which is based on the ongoing relative phase of the VP and the participant's one, taking into account the task condition. where is the relative phase, i.e. the difference between the phase of the movement of the participant and the VP, is the goal of the task, i.e. 0 if in-phase and if anti-phase, N is the total . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. was not peer-reviewed) (which The copyright holder for this preprint . http://dx.doi.org/10.1101/19013771 doi: medRxiv preprint first posted online Dec. 6, 2019 ; number of samples, and arg() indicate the mathematical function returning the angle in radians of a complex number; iv/ an intention score which evaluates the ability for the participant to properly attribute intention towards the "cooperative" or "competitive" behaviour of the VP.
where ℎ and v/ a humanness score which reflects quantitatively the impression of the participant on the human or robotic character of the partner.
where (*(+(, is 0 when the VP is a robot and 1 when it is a human, and is the rating of humanness of the VP reported by the participant at the end of each trial.

Statistical analysis and Normative Modelling
All statistical data analyses were performed using the Python 3. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. was not peer-reviewed) (which The copyright holder for this preprint . http://dx.doi.org/10.1101/19013771 doi: medRxiv preprint first posted online Dec. 6, 2019 ;

Comparison with standardized tests
Using normative modeling we virtually corrected for the developmental bias on the HDC scores.
We then tried to observe how those corrected scores related to standard neuropsychological tests (Table 2).
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. was not peer-reviewed) (which The copyright holder for this preprint . http://dx.doi.org/10.1101/19013771 doi: medRxiv preprint first posted online Dec. 6, 2019 ; We observed a significant interaction effect between the SRS-2 and motor score (r=-0.22; p<0.05*) (Fig.3A). NEPSY-II test showed significant interaction effect between its AF score and the HDC task comprehension score (r=0.33 p<0.05*) (Fig.3B). Supplementary data contain the details of the correlation per group, and with the HDC scores before the normative modeling correction.

Global comparative analysis between the controls and the ASDs groups
The comparative analysis between the two groups shows a statistically significant decrease in the ASDs group for the motor score (d=-0.5; p<0.005**). We also observed a better understanding of the task among ASDs participants (d=0.23; p<0.05*). Interactions between the two groups for the other scores (Coordination: d=-0.21, p=0.12; Intention: d=-0.12, p=0.49; Humanness: d=0.12, p=0.19) were not significant (Fig.4).

HDC scores analysis by sub conditions
HDC paradigm comes with sub conditions (see Figure 5 for a summary). Multiple regressions were thus calculated to predict the different normalized HDC scores based on the diagnostic (coded as 0 = ASD and 1 = CTR), the age (in years), the avatar behavior (coded as 0 = Competitive and 1 = Cooperative), and the human task (coded as 0 = Anti-phase and 1 = In-phase).
Both the diagnostic and the human task were significant predictors of the motor score, with Control Human task and Avatar behavior were significant predictors of the Intention score. Participants tend to better detect the intention of the VP while "anti-phase" (coeff = -1.2718, p < 0.005), and while "in-phase" if the VP takes a cooperative behavior (coeff = 2.3107, p < 0.005).

Developmental aspects of socio-cognitive skills and intervention based on IS in ASDs children
Literature about IS attest of its developmental aspect and plasticity allowing therapeutic detections

Coupling between low-level sensory-motor, motor and high-level socio-cognitive skills
Better affect recognition was associated with a better comprehension (r=0.33; p=0.023*). One hypothesis that could be formulated would be that of a mediating effect of IQ on the recognition of emotions, suggesting that greater cognitive abilities correlates with a higher levels of affect recognition skills. Results attest of lower motor skills among participants with higher probability . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. was not peer-reviewed) (which The copyright holder for this preprint . http://dx.doi.org/10.1101/19013771 doi: medRxiv preprint first posted online Dec. 6, 2019 ; of dimensional diagnosis of ASDs (SRS-2: r=-0.22; p=0.01*). Motricity in ASDs will be discussed further, but a first conclusion of this results suggests a substantial link between low-level motor skills and high-level social-cognitive skills among a population of children and adolescents. Some support a strong pairing between the mirror system and the mentalizing system during communicative gestures, suggesting a cognitive-motor coupling in children (Nadel 2015). The mechanisms involved are not only mechanistic but are also part of neurobiological processes,

Motor skills as a developmental marker of children and adolescents at risk with ASDs
Motor score is the only HDC scores that allows to distinguish the two groups (coeff = 0.4413, p < 0.001**). In overall terms, this motor low-level score is found to be statistically lower among ASDs participants, confirming current data finding altered motor skills in ASDs. . Further analysis showed that the instruction given to the participant (Human task: "in-phase") is associated with a better motor score among the control group (coeff = 0.2726, p = 0.016*). Furthermore, we observed interactions by sub condition. ASDs children tend to have lower results "in-phase" while a "competitive" and cooperative" behavior of the avatar and during "anti-phase" while a "competitive" and cooperative" behavior of the avatar (Table 3). Thus, Wang and colleagues (2019) reported that during a cooperative task of synchronization (i.e. "in-phase"), children with severe diagnostic of ASDs tend to have lower action level.

Socio-cognitive skills based on Interpersonal Synchrony
Mentalizing deficits have repeatedly been described in ASDs population (Bliksted et al. 2016).
This process need preserved metacognitive skills, yet metacognitive monitoring is found diminished in children with ASDs (Grainger et al. 2016). Higher-level scores (intention attribution and task comprehension) require better use of metacognitive processes. However, for both score we did not observe an effect of the group. Thus, for the intention attribution score, it seems easier to detect the intention of the VP while "anti-phase" (coeff = -1.2718, p < 0.005), and while "inphase" if the VP takes a cooperative behavior (coeff = 2.3107, p < 0.005). Furthermore, the comprehension of the task is better if the Avatar is cooperative (coeff = 6.9602, p < 0.005). One result has caught our attention and seems difficult to interpret. ASD participants would have a better understanding of the task than controls (d=0.23; p<0.05*). One hypothesis that could be . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. was not peer-reviewed) (which The copyright holder for this preprint . http://dx.doi.org/10.1101/19013771 doi: medRxiv preprint first posted online Dec. 6, 2019 ; made is that ASDs participants tend to generate mainly the same movement "in-phase" with the avatar without taking the instruction into account. This bias could be responsible for a false positive result.

Implications findings for clinical practice and public health
The aim of this study is also to give standardized values to assess the HDC. We generated percentile ranks for each HDC score from the results obtained in the control patients (see Supplementary   Table 1). In this way it is possible to estimate a child's skills for each assessment.

Conclusion
The HDC can evaluate interpersonal synchrony at both the low and high level of social cognition during live interaction, but also probe the developmental aspects of their relationship. On the other hand, the psychometric evaluation of HDC provides reliable, reproducible, objective, standardized scores, derived from a natural movement. As a new psychometric test, it could identify motor and social markers to improve early detection of neurobehavioral abnormalities during human interaction. This paradigm also provides a basis for the development of therapeutic approaches, for example the serious game "Pop'Balloons" that allows an automatic evaluation in mixed reality (demonstration video: https://vimeo.com/277085489).
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. was not peer-reviewed)

CONFLICT OF INTEREST STATEMENT
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

ACKNOWLEDGMENTS
The authors would like to thank the participants and their families who participated in this study.
They also thank Benjamin Landman, Anna Banki, and Ramon Aparicio-Garcia for early help in the project.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. was not peer-reviewed)  Mean values and respective standard deviations for continuous variables; n = sample size; ASDs = Autism Spectrum Disorders; IQ = Intellectual Quotient; SRS t-score = Social Responsiveness Scale t-score.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. was not peer-reviewed) (which The copyright holder for this preprint . http://dx.doi.org/10.1101/19013771 doi: medRxiv preprint first posted online Dec. 6, 2019 ;  . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. was not peer-reviewed) (which The copyright holder for this preprint . http://dx.doi.org/10.1101/19013771 doi: medRxiv preprint first posted online Dec. 6, 2019 ; Figure 1. Experimental design. Structure of the protocol with four blocks alternating "inphase/anti-phase". Each block is divided into ten trials (Left). Each trial starts with a phase while the participant has to synchronize with the sound of a metronome for 3 seconds. Then, the participant interacts with the avatar according to the instruction (i.e. here, "in-phase"). At the end of the trial, a report of two questions is made directly on the screen. First of all, the impression of the participant on the competitive or cooperative behavior of the avatar and his/er impression on the human nature of the avatar (Right). We represented screenshots of what participants could see (Top).
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. was not peer-reviewed) (which The copyright holder for this preprint . http://dx.doi.org/10.1101/19013771 doi: medRxiv preprint first posted online Dec. 6, 2019 ; . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. was not peer-reviewed) (which The copyright holder for this preprint . http://dx.doi.org/10.1101/19013771 doi: medRxiv preprint first posted online Dec. 6, 2019 ; is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. was not peer-reviewed) (which The copyright holder for this preprint . http://dx.doi.org/10.1101/19013771 doi: medRxiv preprint first posted online Dec. 6, 2019 ; Figure 4. Comparison between the ASDs and control groups across different behavioral scores derived from the HDC protocol and corrected with Normative Modeling. Only motor score really discriminates the two population (d = -0.5; p<0.005**), with significant lower results among ASDs. The lines represent linear regressions. Coloured areas: 95% confidence intervals (CI). Neurotypical participants : blue, ASDs : red.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. was not peer-reviewed) (which The copyright holder for this preprint . http://dx.doi.org/10.1101/19013771 doi: medRxiv preprint first posted online Dec. 6, 2019 ; Figure 5. Score analysis by sub-conditions (Human task: in-phase/anti-phase and Avatar behavior: cooperation/competition), in the two groups. Both the diagnostic (see Global) and the Human Task were significant predictors of the motor score, with Control group having greater scores (coeff = 0.4413, p < 0.001**), and "in-phase" task leading as well to better scores (coeff = 0.2726, p = 0.016*). Controls are in blue and ASDs participants in red. Scores with a statistically significant differences between ASDs and Controls are in bold typeface.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. was not peer-reviewed) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. was not peer-reviewed) (which The copyright holder for this preprint . http://dx.doi.org/10.1101/19013771 doi: medRxiv preprint first posted online Dec. 6, 2019 ;