AI News, Artificial Neural Network Analysis of Spontaneous Preterm Labor ... artificial intelligence

Example 8—Spatio-Temporal Analysis of Uterine Contractions Related Vaginally Delivered Patients and Those Requiring Cesarean Delivery for Active-Phase Labor Arrest

The definition of labor is ‘the presence of uterine contractions of sufficient intensity, frequency, and duration to bring about demonstrable effacement and dilation of the cervix.’ See ACOG Practice Bulletin Number 49, December 2003: Dystocia and Augmentation of Labor.

Unfortunately, the cesarean delivery rate continues to rise (6% in 2003-4), due both to an increase in the primary cesarean rate and a decrease in the rate of vaginal birth after cesarean (VBAC).

Dystocia is a labor abnormality resulting in abnormal progression and may be due to problems with power (uterine contractions and/or maternal expulsive effort), passenger (position or size of the fetus), or passage (shape or size of the birth canal).

Women who labor and then fail to deliver vaginally have more complications (primarily infection and hemorrhage), incur more expense, and consume more resources (labor suite, nurse, etc.) than women who have an elective abdominal delivery.

For VBAC patients, who are at increased risk for labor dystocia, the risk of uterine rupture increases the potential for a catastrophic outcome, and elective repeat cesareans are becoming the standard.

Even when the uterus is relaxed its contents are under minimal pressure: the ‘diastolic uterine pressure.’ During contractions the intrauterine pressure increase depends on the elastic properties of the tissue and the active force exerted by the contractions of the myometrium.

The partogram—a graph of cervical dilation versus time—has been promoted by the World Health Organization for management of labor and recognition of abnormal progress (Rouse D J et al., Active-phase labor arrest: oxytocin augmentation for at least 4 hours.

The traditionally quoted 5th percentile of normal progress in spontaneous term labor is 1.2 cm/h for nulliparas and 1.5 cm/h for multiparas (Dujardin B et al., Value of the Alert and Action Lines on the Partogram.

Unfortunately, the inaccuracy of cervical dilation assessment severely limits the ability to detect a real change of one, or even two centimeters (Zhang J et al., Reassessing the labor curve in nulliparous women.

This may lead to incorrect conclusions regarding progress of labor in up to 33% of those progressing at 1 cm/hr, if checked at 2 hour intervals (Phelps J Y et al., Accuracy and intraobserver variability of simulated cervical dilatation measurements.

This electrical activity can be observed non-invasively from the surface of the maternal abdomen and has been described in some detail (Buhimschi C et al., Electrical activity of the human uterus during pregnancy as recorded from the abdominal surface.

95: 149-53) studied 47 women who presented with complaint of preterm contractions who were admitted either for tocodynamometer (toco) confirmation of contractions or for closer monitoring due to additional risk factors.

30: 402-8) studied 100 patients at 24-32 weeks gestation at high risk for preterm labor and compared ultrasound cervical length measurement, uterine artery velocimetry and EHG in the prediction of labor outcome.

They integrated the active intrauterine pressure (above the baseline), and calculated the energy of the electrical bursts by ‘multiplying the sum of the Y-values of the power density spectrum between 0.34 and 1.0 Hz by the duration of the electrical burst in seconds.’ They found a strong correlation (r=0.764, p=0.002) between these and concluded that EHG accurately reflects uterine contractile activity.

They placed both an internal sensor and seven external uterine activity monitors in 18 women in ‘normal, prolonged, and false labors.’ Though statistics are lacking, his group identified several factors that contributed to prolonged labors, the most important of which were absolute intensity of contractions and absence of fundal dominance.

They report that a right fundal onset of the contraction correlated with subsequent vaginal delivery: 2/30 women (7%) with predominant upper right origin of contractions were delivered abdominally, compared with 7/24 (29%) of women with predominance at other sites, p<0.05.

The opposite was true for the abdominally delivered women (n=6): in every patient, the lower segment mean active pressure exceeded that of the fundus both before and after oxytocin augmentation.

The authors note this fundal dominance ‘might be used to gauge the likelihood of success of oxytocin augmentation.’ As noted above, uterine activity is currently monitored by tocodynamometer or IUPC, but neither correlates directly with progress of labor.

No current method is effective in predicting labor success, and while serial cervical exams are the gold standard, repeated examinations are limited due to the potential for infection.

At present cervical examination is used, and sometimes transvaginal ultrasound or an expensive test of vaginal fluid (fetal fibronectin), but none has an adequate predictive value.

A preterm labor detection method would (1) enable preparation for preterm delivery (betamethasone administration to mature lungs, transfer to a tertiary care center) as well as (2) allow those who are not in labor to be sent home sooner.

In summary, the majority of modern research into labor monitoring focuses on detecting intrapartum fetal asphyxia, which affects about 2% of all deliveries, and is the indication for about 10% of cesareans.

There is a need for a non-invasive means of diagnosing preterm labor, determining labor progress, identifying labor arrest, monitoring response to oxytocin augmentation, accurately distinguishing arrest from very slow labor, and early identification of future labor dystocia.

A reliable monitor with these features could improve outcomes of preterm deliveries, reduce the cesarean delivery rate (by identifying those labors that are merely slow), and shorten labor (by aiding in oxytocin titration and early administration, as well as identifying true arrest so cesarean delivery can proceed).

The result would be improved patient satisfaction, reduced use of health-care resources, and reduced infectious complications (by reducing IUPC usage, frequency of cervical exams and labor duration).

In one embodiment, the subject invention utilizes spatio-temporal information provided from multiple electrodes (also referred to herein as EHG channels) to derive parameters that quantify efficient contractions.

Contraction efficiency enables clinical assessment of various aspects of labor, including: Progress of labor monitoring: the ability to track efficient versus inefficient contractions allows the clinician to know when a patient's labor is unlikely to progress to delivery.

This early detection of failure to progress can save the hospital money, save the staff time and effort, and avoid problems that might be caused by prolonged attempts to deliver vaginally.Effectiveness of labor augmentation: the ability to identify the need for oxytocin, and to rapidly assess its effect and speed its accurate titration, with a minimum of cervical exams, thus reducing the risk of chorioamnionitisTransition during labor induction: the ability to recognize when a misoprostil or other induction is ready for oxytocin, limiting cervical exams and speeding inductions by avoiding unnecessary delays.Preterm labor monitoring: many patients arrive at the hospital with preterm labor symptoms.

This invention may be used to rapidly triage patients and send home those patients who are not in preterm labor.Home monitoring: either similar to the above discussion for continuous preterm labor monitoring at home, or for parents who would like a better indication of when they should leave for the hospital (e.g.

The subject invention can predict and detect problems with overall uterine activity.Patient feedback on effective methods of ‘pushing’ With the subject invention, muscle electrical activity is measured externally Uterine contractions are the result of the coordinated actions of individual myometrial cells.

The subject invention provides a contraction efficiency indicator that estimates the dilation at the next cervical exam (short term evaluation) as well as a dystocia prediction indicator that estimates the likelihood of eventual vaginal delivery or a preterm delivery.

According to the subject invention, regions of data from arrested patients before and after oxytocin augmentation can be utilized to establish parameters that are significantly different between periods of pre-arrest, positive oxytocin response, and lack of oxytocin response.

Models, composed of the parameters shown to correlate with the short-term ability of the contractions to produce dilation, are created and analyzed to determine the ability of the models to predict if the patient is currently dilating adequately.

6 is a snapshot of the contractile map of the uterus (COMU), which is a colorized image of the spatial contraction intensity over time, at a time t recorded during contraction where red corresponds to high power and blue to low power.

17 graphically illustrates receiver operating characteristics curves for likelihood of labor arrest and cesarean delivery using matching criteria (gestational age, body mass index (BMI), spontaneous versus induced labor, and dilation time of study) with or without the addition of EHG data.

In one embodiment, the systems and methods of the invention enable quantification of uterus contraction for use in monitoring the progress of labor as well as predicting labor success and diagnosing real preterm labor.

(3) a computing means for receiving and analyzing the spatio-temporal information in the sensor input to extract EHG results and derive parameters, including without limitation, time of contraction, location of contraction, extent of contraction, direction of propagation (gradient), speed of propagation over the abdomen, average contraction over time, spatial location of the peak of contraction and its variance, spread of the contraction, abdominal intensity distribution, abdominal frequency distribution, patterns of intensity over time, distance of peak power propagation, power changes over time, frequency changes over time;

and using the parameters to quantify uterine contraction (i.e., to establish a contractile map of the uterus (COMU)) and for clinical use in monitoring progress of labor, monitoring preterm labor, remote or home monitoring of preterm labor, external IUPC prediction, and pharmaceutical efficacy (i.e., titration of oxytocin).

In one embodiment, the system of the invention further includes an intelligence system that can use the parameter data generated by the processor in offering support/advice for making clinical decisions (i.e., to interpret labor progress, likelihood of delivery within a period of time;

An intelligence system of the subject invention can include, but is not limited to, artificial neural networks, fuzzy logic, evolutionary computation, knowledge-based systems, optimal linear or nonlinear filtering, and artificial intelligence.

In one embodiment, a neural network system is provided in the monitoring system of the invention to enable real-time assistance in providing additional clinical data (i.e., classification of labor progress and prediction of preterm labor).

In a related embodiment, a portable system enables continuous non-invasive uterus monitoring for beneficial assessment of labor (i.e., presence of effective contraction to determine labor).

One embodiment of the invention utilizes a set of electrodes (at least 4 electrodes but preferably 6 up to 200 electrodes) provided on a mesh (or vest), wherein the mesh can function as an electrode-stabilization component.

In a preferred embodiment, 15 signals are created from the 8 signals generated from the electrodes, where the 15 signals are provided by pairwise subtraction between electrode neighbors to subtract the commonality of the reference and providing local information about the uterus contractility pattern.

In a related embodiment, filters are also applied to signals during and after processing by the computing means (i.e., filters applied prior to and during parameter extrapolation operations).

According to the subject invention, filtering operations that can be used, either alone or in combination, in extracting vital signal components include, but are not limited to, finite impulse response (FIR) filters and infinite impulse response (IIR) filters, such as band-pass filters, high pass filters, low pass filters, band-stop (or “notch”) filters, and also to non-linear filters such as the median filter.

The computing means can also detect and act upon user input via user interface means known to the skilled artisan (i.e., keyboard, interactive graphical monitors).

In another embodiment, the subject invention is practiced in a remote setting, for example, personal residences, mobile clinics, vessels at sea, rural villages and towns without direct access to healthcare, and ambulances, wherein the patient is located some distance from the physician.

In a related embodiment, the computing means is a custom, portable design and can be carried or attached to the patient in a manner similar to other portable electronic devices such as a portable radio, or interwoven in the patient clothing as a wearable computer.

12, the computing means used in accordance with the subject invention can contain at least one user-interface device including, but not limited to, a keyboard 10, stylus 12, microphone 14, mouse 16, speaker 18, monitor 20, and printer 22.

The algorithm operations 36, including the filtering operations 34 and 38, can be embodied in the form of computer processor usable media, such as floppy diskettes, CD-ROMS, zip drives, non-volatile memory, or any other computer-readable storage medium, wherein the computer program code is loaded into and executed by the computing means.

The memory capacity of the invention can support loading a computer program code via a computer-readable storage media, wherein the program contains the source code to perform the operational algorithms 36 of the subject invention.

In addition, as understood by the skilled artisan, the memory capacity of the computing means can be expanded with additional hardware and with saving data directly onto external mediums including, for example, without limitation, floppy diskettes, zip drives, non-volatile memory and CD-ROMs.

The computing means can further include the necessary hardware and software to convert processed signals into an output form 40 readily accessible by the trained physician, nurse practitioner, midwife, or technician.

For example, without limitation, an audio device 42 in conjunction with audio speakers 18 can convert and play a processed uterine information into an audio signal, and/or a graphical interface 44 can display ECG signals in a graphical form on a monitor 20 and/or printer 22.

In one embodiment, the patient is hospitalized, and clinical data generated by a computing means is transmitted to a central location, for example, a monitoring station located in a maternity ward, or to a specialized physician located in a different locale.

For example, patients can be located at personal residences, mobile clinics, vessels at sea, rural villages and towns without direct access to healthcare, and ambulances, and by using the contraction/labor progress monitoring system of the invention, still provide clinical data to the health care provider.

Advantageously, mobile stations, such as ambulances, and mobile clinics, can monitor maternal-fetal health by using a portable computing means of the subject invention when transporting and/or treating a patient.

Communication devices such as wireless interfaces, cable modems, satellite links, microwave relays, and traditional telephonic modems can transfer clinical data from a computing means to a healthcare provider via a network.

Amplifier output is subsequently communicated to a variety of operational algorithms 36 for processing EHG signals into parameters for establishing clinical data (i.e., progress of labor, preterm labor, IUPC prediction, etc.) and subsequent presentation of clinical data to the user 65.

Operational algorithms 36 can include, without limitation, average contraction over time operations 67, spatial location of contraction peak and variance operations 69, and contractile map modeling operations 71.

In certain instances, prior to EHG signal processing, signals received from an amplifier are communicated to filter operations for each sensor channel Raw signals extracted by sensors of the invention are a mixture of several sources, namely maternal-fetal vitals signs (i.e., maternal ECG, fetal ECG, EMG signals) and noise.

In accordance with the subject invention, the filtered signals are then followed by appropriate operations for obtaining desired parameters (i.e., EHG results) for use in generating useful clinical data (such as progress of labor, preterm labor, IUPC prediction).

contractile map extraction operations, average contraction over time operations, spatial location of contraction peak and variance operations, and contractile map modeling operations.

In accordance with the subject invention, at the output of EHG extraction operations, contractile map extraction operations can be performed on a computing means (such as a computer processor), preferably in real time.

In one embodiment, from the EHG signals, location signals are generated by pair-wise subtraction between signal neighbors to remove common signal characteristics and provide local information about the uterine contractility pattern.

Following derivation of the location signals, the spatio-temporal patterns of these location signals are displayed in a grid for a specified amount of time, preferably a 5×5 grid, 10 times a second.

For example, a COMU of the invention can estimate the center of a contraction by a calculation similar to the center of mass (see section 5 below) and the power (see section 4 below) as specified later.

7A-7F, it is readily apparent that the illustrated contraction begins near the fundus, its apex moves down towards the cervix but stays midline, increases in strength, and subsides following the reverse pattern.

The second panel shows the movement of the center of the contraction as defined herein, and in all these cases, the fundus to cervix movement during the ascending part of the contraction (T1) and the retraction during the descending part of the contraction (T2) are illustrated.

In one embodiment, the fitting is done using a Levenberg Marquadt and is checked by the residual of the fitting, yielding the following equation: Gauss(t)=GaussAmp(t)*∑i=15∑i=15exp(-((Zi,j(t)-meanGaussX) ⋀2)varGaussX)-((Zi,j(t)-meanGaussY) ⋀2)vatGaussY) The derivative in space of the model represents the Laplacian of the contraction, which is associated with the direction of propagation of the force generated by the contraction, i.e.

All patents, patent applications, provisional applications, and publications referred to or cited herein are incorporated by reference in their entirety, including all figures and tables, to the extent they are not inconsistent with the explicit teachings of this specification.

These patients' records were each carefully reviewed for accuracy of surgical indication (based on the traditional definition of arrest: 2 hours of adequate uterine activity without cervical change) and then matched with two vaginally delivered controls with the following characteristics: a normal labor curve;

In 16/20 (80%) of SVD controls, the predominant pattern during the second half of the contraction was an upward movement of the CM (that is, the contraction power was greater in the fundus than near the lower uterine segment [LUS]).

Without being bound to any particular theory, it is hypothesized that this CM apex movement is a quantitative parameter of “fundal dominance” Recognizing that the definition of permanent arrest of dilation requiring cesarean delivery is a matter of debate, and that, with additional time, many of the cesarean patients would have eventually delivered vaginally, it is intriguing to speculate that CM outliers might actually be the group that were merely on a protracted labor curve.

Two distinct, but not mutually exclusive possibilities are considered herein: (1) there are features that predict short-term labor progress and (2) there are features (such as fundal dominance) that predict eventual labor outcome.

While clinically known that oxytocin will often help labor progress by increasing the frequency of contractions, and often their power, it may be that oxytocin merely speeds an already ‘functional’ labor, while doing little for a truly dysfunctional labor (i.e.

According to the subject invention, a large sample of labor segments can be used to train a multi-parameter model, the contraction efficiency indicator (CEI), to predict whether a subject is dilating normally.

Information stored in the database includes: maternal age, height, weight, race, any diagnoses, obstetric history, labor onset (spontaneous versus induced), membrane status (artificial versus spontaneous rupture, time), gestational age, fetal presentation, estimated fetal weight, all cervical examinations (including dilation, effacement and station), all medications administered with dose and time, continuous cardiotocographic and all data from the abdominally sited electrodes including individual impedances, and information regarding the type of delivery, newborn weight and Apgar scores.

Patient subsets from this database are suitable for initial design of the predictive models: Dystocia case-control cohort: as described above, 10 patients who required cesarean delivery for active-phase labor arrest were each matched with two vaginally delivered controls with a normal labor curve, who were monitored at the same dilation as the arrest patients.

Chart review will be used to confirm spontaneous onset of labor, indication for oxytocin augmentation (dilation <1 cm/hr), cervical exam at least every 3 hours after oxytocin initiation, and lack of complicating factors such as chorioamnionitis and macrosomia.

Prospective arrest cohort: A prospective study will be performed throughout the study period using the same data collection protocol described in Preliminary Studies: term (≥37 weeks gestation) patients in spontaneous labor with a single viable fetus in cephalic presentation at ≥5 cm (active phase) who are diagnosed with arrest (no cervical change in two hours) and have no bleeding or uterine scar will be eligible for inclusion.

After informed consent and initiation of EHG monitoring, an IUPC will be placed and oxytocin augmentation begun according to the standard labor unit protocol (increases at 30-minute intervals until MVUs are 150-250 or there is cervical change).

Additional data to be collected includes maternal demographic data, labor information (details of onset, membrane rupture, cervical exams, oxytocin dosages, and interventions), delivery newborn information (mode of delivery, newborn weight and Apgar scores), and the continuous output of the cardiotocograph monitor.

The motion in each portion is then quantified using the following steps: Determine the beginning and end of the contraction using a dynamic threshold on the CM amplitude.Identify the spatial location of contraction onset.Divide the contraction into its ascending part (from the beginning to the peak, T1) and descending part (from the peak to the end of the contraction, T2).

In the subject invention, those contractions with higher fundal power at the end of the contraction (‘Up’ during T2) are defined as fundal dominant, and those whose initial onset is at the fundus are defined as having fundal origin.

The initial parameterization will be as follows: Fundal Dominance: Percentage of contractions that exhibit upward movement of the CM apex during T2, duration of time there is a fundal to LUS gradient of uterine activity, and the ratio of the total uterine activity in each segment.

The parameters that are uncorrelated with other retained parameters are preferentially retained to provide a list of parameters that independently correlate with cervical dilation in response to augmentation, short-term markers of contraction efficiency.

From the entire database of term laboring women, segments from active labor that are bounded by cervical examinations and are 90-180 minutes in length can be extracted to train a computing means (i.e., intelligence means such as an artificial neural network).

For the DPI model building, both the Dystocia cohort (all monitored patients who delivered by cesarean for failure to progress, regardless of whether they were monitored at the time of arrest) and all patients who delivered vaginally can be utilized.

From these subjects, each 2-hour segment from 5 cm dilation to the onset of pushing or end of collection is extracted, together with dilation at the start of the segment (estimated by assuming linear dilation between the most recent and first subsequent cervical examinations), and eventual labor outcome.

Both linear and non-linear (neural network, in particular multi-layer perceptron) models can be evaluated for this task and compared using positive and negative predictive values and ROC curves with AUC analysis.

Cervical exams continue at two-hour intervals, but an additional cervical check can be indicated in the following circumstances: (1) for the patient with initially inadequate MVUs, 30-minutes after the target range (150-250) is achieved;

Each point includes the elapsed time since the last cervical examination, and three dichotomous values: presence of cervical dilation (Dil+ if change noted), CEI status (CEI+ if it predicts cervical change over the segment), and MVU status (MVU+ if ‘adequate’ at 150-250).

It is estimated that an average of 3-4 data points can be generated for each patient, based on the 2-hour maximum interval between cervical examinations, the ‘modern’ average active phase dilation time of 5.5 h to 7.7 h for nulliparas and 5.7 h for multiparas (Zhang J et al., Reassessing the labor curve in nulliparous women.

(b) ability to identify effective uterine contractions better than the IUPC (CEI+ when MVU− yet cervical dilation ensues) and identifies the minimum effective oxytocin dose, perhaps lowering the total dose required, together with side effects;

Patients who undergo cesarean delivery for other indications are excluded (e.g., fetal heart rate tracing abnormalities or failure of descent after complete cervical dilation), which is an estimated 25% of the patients.

Patients who underwent primary cesarean delivery for labor arrest after achieving at least 5-cm cervical dilation and had EHG monitoring with the final amplifier design (two-versions) for at least 30-minutes during the period of arrest were analyzed.

Patients were excluded if they had a uterine scar due to the increased rate of repeat cesarean delivery in this population (Landon M B et al., Maternal and perinatal outcomes associated with a trial of labor after prior cesarean delivery.

The subjects were matched for gestational age ±2 weeks, BMI ±10, parity (nulliparous or parous), induction versus spontaneous labor, and EHG monitoring during dilation within ±1 cm of the dystocia.

Upon reviewing the EHG data for the controls, some had noise in the signal at the dilation of interest (determined by assuming a linear dilation rate between surrounding cervical examinations), thus the dilation at the segment used occasionally varied by more than ±1 cm of the dystocia, but never more than ±2 cm.

After channel normalization, rectification and filtering (Butterworth low pass filter with a cutoff at 0.02 Hz), these signals represented the local contraction strength at 17 locations over the abdomen, displayed as a 5×5 grid (FIG.

This grid was calculated 10 times per second and the data interpolated over the abdomen and over time generating a real-time movie of the relative intensity of the uterine muscle activity over the abdomen (FIG.

Using a 30-minute window positioned during the labor arrest, or near the same dilation in the controls, the vertical direction of CUA movement during each half of each contraction was calculated: T1=onset to peak, T2=peak back to baseline.

Finally, contraction patterns were classified as moving downward (toward the lower uterine segment (LUS)) or upward (toward the fundus) for each half of each contraction, resulting in four contraction patterns: “LUS-Fundal”, “LUS-LUS”, “Fundal-LUS”, and “Fundal-Fundal”.

For each arrest patient, the prevalence of each of the four contraction patterns was evaluated by calculating the average prevalence for the vaginal deliveries in each of the matched clusters and subtracting the prevalence of the associated cesarean delivery.

Overall, twenty-three out of twenty-four (23/24) vaginally delivered patients had a predominantly fundal CUA direction during T1 and/or T2, compared with four out of twelve (4/12) for the cesarean delivery cohort.

Logistic regression analysis using only the pairing variables—gestational age, BMI, parity, spontaneous versus induced labor, and dilation at the time of study—to predict outcome (cesarean for arrest versus vaginal delivery), resulted in an area under the ROC curve (AUC) of 0.79.

Calculating a Chi-Square value of 13.090 for the difference between two models' log likelihood score gives a p value=0.004, indicating the addition of the dilation patterns is a significant predictor of delivery type.

All patents, patent applications, provisional applications, and publications referred to or cited herein are incorporated by reference in their entirety, including all figures and tables, to the extent they are not inconsistent with the explicit teachings of this specification.

It should be understood that the examples and embodiments described herein are for illustrative purposes only and that various modifications or changes in light thereof will be suggested to persons skilled in the art and are to be included within the spirit and purview of this application.

Frontiers in Physiology

Neonates of some vertebrate species orient their first approach responses toward objects that exhibit features present in social partners and caregivers: face-like configuration, biological motion and self-propulsion.

Comparative research on human infants and newly hatched domestic chicks (Gallus gallus) found striking similarities in the static and dynamic visual cues that attract attention of these different species soon after birth (Di Giorgio et al., 2017a).

Chicks also seem to have a spontaneous preference for objects autonomously starting to move over objects set in motion after a collision (Mascalzoni et al., 2010) and for objects autonomously changing their speed over constant moving ones (Rosa-Salva et al., 2016).

Neonates at high familiar risk of ASD show significant differences compared to low-risk neonates in the preference for a face-like stimulus and for biological motion, suggesting an impairment in the development of the predisposed mechanisms for detecting animate beings (Di Giorgio et al., 2016).

To further study the effect of VPA on early predispositions, and to investigate whether the impairment for static cues is accompanied by impairment in predispositions for dynamic cues, we compared the spontaneous preference for self-propelled stimuli in VPA- and vehicle-injected chicks.

In domestic chicks, administration of 35 μmoles/egg (corresponding to approximately 100 mg/kg) has been tested between E10 and E14 with differential effects on hatching rate, showing a dramatic decrease of hatchings at E10 and a significant decrease of hatchings at E12 but no significant effect at E14 (Nishigori et al., 2013).

Controlling the visual experience during pre- and post-natal development enable to exclude any interference of visual stimuli in the expression of predispositions toward animacy cues, and to demonstrate the innate nature of these mechanisms.

The corridor was divided in three sectors: a central sector (45 cm long) delimited by two steps, that the animals had to climb to enter the two choice sectors (each 20 cm long) immediately adjacent to the two screens.

preference for the speed-change stimulus was measured by the ratio of time (in seconds) spent in the choice sector near the speed-change stimulus divided by the cumulative time spent in either of the choice sectors (preference).

In the control group (vehicle-injected), the preference for approaching the speed-change stimulus was similar to what previously observed, and the preference scores were significantly higher than chance level [t(51) = 2.365, p = 0.011;

On the contrary, VPA exposure significantly reduced the preference for the speed-change stimulus: the preference scores for approaching the speed change stimulus did not differ from chance level [t(50) = -0.406, p = 0.686;

speed-change N = 35, speed-constant N = 17), in the VPA-treated group no significant difference was found in the number of chicks that approached the two stimuli (χ2= 0.176, p = 0.78;

We investigated unlearned predispositions to orient toward animate motion cues in VPA-injected chicks compared to vehicle-injected controls, using a choice preference test between a speed-change and a constant moving stimulus.

In phylogenetically distant species of vertebrates, such as domestic chicks and humans, similar mechanisms have been described to drive early approach responses toward static and dynamic cues typically associated with animate figures.

The adaptive function of early predispositions has been hypothesized to be in directing attention toward highly important animate stimuli, enabling future learning through experience and enhancing social interactions (Johnson et al., 2015;

In chicks, predispositions are likely to orient the young animal toward the mother hen (or other brood mates), directing subsequent filial imprinting responses toward animate stimuli (Miura and Matsushima, 2016).

In human newborns, subcortical fast and automatic mechanisms have been hypothesized to underlie these social predispositions, directing attention toward animate entities to create an early social bond with the caretakers and social companions (Tomalski et al., 2009;

Several accounts suggest that abnormalities in this early social-orienting system may lead to deficits in social stimuli processing, limiting attention to salient social stimuli, decreasing their reward value and resulting in the atypical social behavior associated with ASD.

This drug has been used to model ASD core deficits in other vertebrate species (Ranger and Ellenbroek, 2016) although chicks are the first precocial species in which its effect on social behavior has been investigated (Nishigori et al., 2013;

Precocial species, like domestic chicks, are characterized by the early maturation of the motor and sensory system, that allows to perform behavioral tests soon after birth, before gaining any social experience.

Our work on VPA-mediated impairment of early predispositions, together with the deficits documented in human neonates at high risk of ASD (Di Giorgio et al., 2016), supports the hypothesis of early social orienting mechanisms shared across species whose impairment or delay might have a pivotal role in the pathogenesis of autism.


Yet, the relationship with labor is distinctive for a company that was founded on the principle of using automation to reduce manual lab testing and using machine learning to minimize reliance on human “intuition” within the hypothesis-driven scientific method.

“Zymergen has been reducing the staff devoted to each project but rapidly increasing the number of projects we take on.” Comparing Zymergen to another biotech company for new product development, Zymergen’s core research teams 23 appear to have about 50 percent fewer researchers (scientists and research associates)24 with significantly higher output (a tenfold increase in productivity compared with conventional labs).

As stated above, the core research teams at Zymergen, e.g., scientists and research associates, are about half the size of other R&D teams, as well (See Exhibit 5: Comparison of staffing models.) The lower ratio of research associates on a core team is due to the role of automation in the wet lab and machine learning at Zymergen, which has replaced many of the tasks that an RA could be responsible for in a typical R&D lab.

While core strain improvement program teams at Zymergen may be smaller by about 50 percent, the core research teams are enabled by an increased number of shared services throughout the organization.25 Compared to peers in biotech and R&D teams at clients, Zymergen likely has a higher ratio of labor dedicated to enabling functions or shared services.

From a labor perspective, Zymergen’s organization appears to have a ‘top-heavy’ research-intensive team of scientists and research associates with more diverse, specialized, and distributed teams of scientists, engineers, roboticists, and data scientists as support and shared services.

This difference in functional labor breakdown between Zymergen and its peer organizations is explained by two main factors: First, automation and machine learning tools require significant human resources to develop, support, and improve continuously.

In contrast to a traditional R&D lab which has a much larger core research team, out of about 7500 to 8600 total FTEs at Zymergen, only 34 percent are fully dedicated core research teams involved in client programs and internal product development.

The upfront costs associated with automation in a high-throughput screening lab or in machine learning models and infrastructure might not make economic sense due to: (1) fewer programs to realize marginal benefits, (2) capital intensity and limited funding, and (3) lack of available technical talent and organizational hurdles.

Within Zymergen, program staffing is more dynamic, with a larger input of resources (some fully-dedicated, some partially-dedicated) required upfront staffing in the tech transfer phase (defined below), followed by fewer fully-dedicated resources to sustain the program during the strain improvement phase.

During the first phase of a project (the tech transfer phase), Zymergen integrates the client’s specifications into the Zymergen platform and ensures that the client’s strain can be integrated into Zymergen’s “factory” workflows (systems in the high-throughput screening/automated wet lab).

During the strain improvement phase, the program is in steady-state and is iterating through highly automated experimentation cycles to identify optimizations to the strain for the client’s desired product and program objectives (e.g., improve economics, accelerate commercialization).

It is possible that the savings resulting from a Zymergen program could increase demand for fermentation products or biomaterials, resulting in increased demand for manufacturing labor by customers, or even new downstream job opportunities associated with sale of newly developed products.26 However, Zymergen’s programs could also potentially decrease labor need associated with manufacturing: Labor at fermentation plants: If production volume can be increased due to a strain improvement program, the labor productivity of manufacturing operations will increase, i.e., in theory, less labor will be needed to manufacture the same volume of product.

If Zymergen achieves its founders’ goals, the impact on highly skilled workers will be profound: These workers could face the same drop in available jobs that others have identified in traditional manufacturing jobs, borne from factory automation.“ Our vision is to have all of the manual parts of the work done by robotics and to have all the intellectual efforts, such as design and interpretation, to be done with machine learning and big data,” a co-founder said.

number of the micro labor trends observed at Zymergen could offer clues to the future of the broader workforce:27 Zymergen’s techniques led to reported cost savings for its customers on both R&D and manufacturing, as well as higher business productivity internally.

Zymergen reports that its fast cycle times allow research teams to meet ambitious timelines due to automation in the lab and confidence in the data, leading to shorter project durations overall compared to conventional R&D labs.

The focus is typically either on improving yield, thereby saving money on raw materials (e.g., reduced sugar consumption per unit output of product), or improving the speed of production (e.g., faster fermentation with the same level of inputs).

Significant investments have included: Infrastructure investments: Zymergen made significant investment in infrastructure to enable its business, including robotics and machinery in the wet lab, sensors for automatic data capture, and IT/data infrastructure to centralize and store data.

Theoretically, as Zymergen scales the number of programs, data science and automation engineers could scale at a slower rate due to reusability of machine learning models and wet lab infrastructure across multiple programs.

Unconventional cost structure: Zymergen has a different cost base than conventional R&D teams or other biotech firms: lower estimated labor costs as a percent of total costs, with significantly higher consumables costs (reagents, DNA, raw materials for experiments) because of the increased volume of experiments performed (see next section);

Yet the company has major consumable costs associated with its data acquisition strategy: To run each experiment, the company needs to purchase an inventory of supplies, from pipettors to the microbial cultures, reagents, and DNA used in experiments, which can add up quickly with each marginal experiment.

The Road to Intelligence

A panel discussion with: Patrick Winston, Josh Tennenbaum, Laura Schulz, Bob Desimone, Shimon Ullman and Geoffrey Hinton Chaired by Tomaso Poggio ...

Living Universe | Journey To Another Stars - Documentary 4K [2018]

The Earth is Not Alone: LIVING UNIVERSE is an interstellar adventure that seeks to answer the most profound question of all: ..

Artificial Intelligence & Consciousness

Zylok discusses the development of Artificial Intelligence, the potential for AI to become conscious, the looming threats of mishandling a sentient AI, and how ...

Biblical Series III: God and the Hierarchy of Authority

Although I thought I might get to Genesis II in this third lecture, and begin talking about Adam & Eve, it didn't turn out that way. There was more to be said about ...

On the Neural Machinery of Faces

Winrich Freiwald, Professor of Neurosciences and Behavior at The Rockefeller University, discusses the connection between sociality and intelligence, the ...

Collective Intelligence

CAST Visiting Artist Agnieszka Kurant joins Stefan Helmreich, professor of Anthropology; Caroline Jones, professor of History, Theory and Criticism of ...

Cern Revelation 9 Warning - The Key To The Bottomless Pit?

This is a hard hitting video. We are not sugar coating anything here. Please do your own research on this content to come to your own conclusions. Here is some ...

Interrupt V: Jackie Wang

Interrupt 2019 Jackie Wang introduced by Harper Shalloe Jackie Wang is a poet, multimedia artist, and Ph.D. candidate in African and African American Studies ...

Mapping the mouse brain - Partha Mitra CSHL Prof.

Cold Spring Harbor Laboratory Prof. Partha Mitra, director of the Mouse Brain Architecture Project, talks about mapping the mouse brain at the Secret Science ...

Philosophy of artificial intelligence | Wikipedia audio article

This is an audio version of the Wikipedia Article: Philosophy of artificial intelligence Listening is a more natural way of learning, when compared to reading.