Systems and methods for multi-dose drug classification

Abstract

A machine learning system can predict a therapeutic effect of a compound when administered to humans. The machine learning system can obtain multiple observational datasets acquired during trials in which a non-human animal was administered varying dosages of the compound. The machine learning system can apply the multiple observational datasets to a machine learning model trained to predict, based on such observational datasets, therapeutic effect(s) of the compound when administered to humans. The machine learning model can include multiple encoders, each encoder corresponding to a dosage level. The machine learning model can further include a classifier configured to generate indication(s) of therapeutic class(es) using the outputs of the encoders.

Description

Attorney Docket No.16048.0009-00304 SYSTEMS AND METHODS FOR MULTI-DOSE DRUG CLASSIFICATION CROSS-REFERENCE TO RELATED APPLICATION

[0001] The present application claims priority to U.S. Provisional App. No.63 / 594,275, filed on October 30, 2023, which is hereby incorporated by reference in its entirety. TECHNICAL FIELD

[0002] The present disclosure relates to devices, systems, and methods for capturing, analyzing, and augmenting behavioral and physiological data of subjects. The subjects can be non-human animals and the behavioral and physiological data can be captured during a trial in which a compound is administered. BACKGROUND

[0003] The evaluation of compounds (e.g., as part of a drug development program) can involve non-human animal trials. Such animal trials can provide the safety and efficacy data necessary to support subsequent human trials.

[0004] Observational assessments can be an important component of such trials. The effects of experimental substances on animal physiology or behavior can provide information about the potential clinical effects of the experimental substances on human subjects. For example, the discovery that chlorpromazine produces differential effects on avoidance and escape behavior in animals encouraged the evaluation of the behavioral effects of other experimental antipsychotic drugs.

[0005] An observational assessment can include recording the physiology or behavior of an animal treated with a compound. Observational platforms can automatically collect suchAttorney Docket No.16048.0009-00304 observational data. As compared to manual collection of observational data, automatic data collection can be more complete, repeatable, reliable, and accurate. In particular, observational platforms can collect EEG signals from behaving animals administered varying dosages of a compound. SUMMARY

[0006] Certain embodiments of the present disclosure relate to systems and methods for predicting the therapeutic effect of a compound in humans using observational datasets acquired from non-human animals. The observational datasets can correspond to multiple dosage levels of the compound. A machine learning model can use the information from the multiple dosage levels to improve the prediction of therapeutic effect(s) of the compound when administered to humans at dosages corresponding to the multiple dosage levels.

[0007] The disclosed embodiments include a non-transitory, computer-readable medium containing instructions. When executed by at least one processor of a system, the instructions can cause the system to perform operations. The operations can include obtaining a sample corresponding to a compound. The sample can include electroencephalographic spectrograms corresponding to different positions in a dosage sequence. The operations can include generating an indication of a predicted therapeutic effect of the compound when administered to a human for each position in the dosage sequence by applying the sample to a machine learning model. The machine learning model can be configured to generate separate encodings for the electroencephalographic spectrograms and generate the indication using the separate encodings. The operations can include providing the indication.

[0008] The disclosed embodiments include a system for classifying compounds. The system can include at least one processor and at least one non-transitory, computer-readable medium. The medium can include instructions. When executed by said at least one processor, theAttorney Docket No.16048.0009-00304 instructions can cause the system to perform operations. The operations can include obtaining a sample corresponding to a compound. The sample can include observational datasets corresponding to positions in a dosage sequence. The observational datasets can include a first observational dataset corresponding to a first position in the dosage sequence and acquired from a first non-human animal administered a first dosage of the compound. The operations can include generating an indication of at least one predicted therapeutic effect of the compound when administered to a human by applying the sample to a machine learning model. The machine learning model can be configured to generate separate encodings corresponding to different positions in the dosage sequence and generate the indication using the separate encodings. The operations can further include providing the indication.

[0009] The disclosed embodiments include a method of training a machine learning model to classify compounds. The method can include obtaining a first sample corresponding to a first compound. The first sample can include a first observational dataset corresponding to a first position in a first dosage sequence and acquired from a first non-human animal administered a first dosage of the first compound during a first trial. The method can include labeling the first sample with an indication of at least one first predicted therapeutic effect of the first compound when administered to a human. The method can include training the machine- learning model, using a training dataset including the first sample, to generate an indication of at least one predicted therapeutic effect of a compound when administered to a human using a sample including observational datasets corresponding to positions in a dosage sequence. The machine-learning model can be configured to generate separate encodings corresponding to different positions in the dosage sequence and generate the indication using the separate encodings.Attorney Docket No.16048.0009-00304

[0010] The foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the claims. Other systems, methods, and computer-readable media are also discussed within. BRIEF DESCRIPTION OF DRAWINGS

[0011] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate several embodiments and, together with the description, serve to explain the disclosed principles. In the drawings:

[0012] FIGs.1A to 1C depict a system for acquiring (an optionally analyzing) observational data concerning a non-human animal, consistent with disclosed embodiments.

[0013] FIG.2 depicts observational datasets generated using data acquired during a set of trials, consistent with disclosed embodiments.

[0014] FIG.3 depicts an exemplary method for training a machine learning model to predict therapeutic effect(s) of a compound when administered to a human, consistent with disclosed embodiments.

[0015] FIG.4 depicts an exemplary method for predicting therapeutic effect(s) of a compound when administered to a human, consistent with disclosed embodiments.

[0016] FIG.5 depicts an exemplary architecture of a machine learning model for predicting a therapeutic effect of a compound in humans using observational data obtained from non- human animals, consistent with disclosed embodiments.

[0017] FIG.6 depicts an exemplary system for generating machine learning models, consistent with disclosed embodiments.Attorney Docket No.16048.0009-00304 DETAILED DESCRIPTION

[0018] Reference will now be made in detail to exemplary embodiments, discussed with regard to the accompanying drawings. In some instances, the same reference numbers will be used throughout the drawings and the following description to refer to the same or like parts. Unless otherwise defined, technical or scientific terms have the meaning commonly understood by one of ordinary skill in the art. The disclosed embodiments are described in sufficient detail to enable those skilled in the art to practice the disclosed embodiments. It is to be understood that some embodiments may be utilized and that changes may be made without departing from the scope of the disclosed embodiments. Thus, the materials, methods, and examples are illustrative only and are not intended to be necessarily limiting.

[0019] Machine learning models consistent with disclosed embodiments can predict the effect(s) of a compound in humans based on the effect(s) of the compound in non-human animals. Such machine learning models can generate a prediction using observational datasets acquired during multiple trials. The trials can involve administration of different dosages of the compound to one or more non-human animals. In some instances, the trial may be intended to investigate the effect of the compound on the behavior of the animal. For example, the trial may be conducted as part of a drug discovery study of the compound.

[0020] Compounds can exhibit differing relationships between dosage and observed effects. Such relationships can be predictive of the effect of the compound in humans. For example, as the dosage of an administered compound is increased, the effect of the compound on the behavior of a non-human animal can follow a characteristic trajectory in some feature space. A machine learning model can be trained to associate this characteristic trajectory with predicted therapeutic effect(s) in humans.Attorney Docket No.16048.0009-00304

[0021] Predictions generated using observational datasets corresponding to a sequence of dosage levels may therefore be more accurate than predictions generated using observational datasets corresponding to a single dosage level. As described herein, dosage levels in the sequence of dosage levels can correspond to different dosages for different compounds, consecutive dosage levels can correspond to the same dosage, and the observational datasets corresponding to different dosage levels may be acquired from different animals.

[0022] Consistent with disclosed embodiments, a machine learning model can be configured to use either increasing sequences of dosage levels (e.g., given a sequence of dosages ^ ^^^, … , ^^^^ and corresponding dosage levels ^1, … , ^^^, then ^^^^ ^^^for ∀ ^^|1 ^ ^^ ^ ^^ ^ ^^) or decreasing sequences of dosage levels (e.g., ^^^^ ^^^for ∀ ^^|1 ^ ^^ ^ ^^ ^ ^^). For convenience of discussion, the disclosed embodiments are discussed with respect to increasing sequences of dosage levels. However, the disclosed embodiments are not so limited.

[0023] The robustness and flexibility of the disclosed embodiments provide additional benefits beyond improved accuracy. In particular, the disclosed embodiments permit reuse of historical observational data acquired under a variety of trial protocols (e.g., protocols having differing dosage levels, differing numbers of dosage levels, different animals administered different dosage levels, or the like). Historical observational data can be reformulated into observational datasets corresponding to a sequence of dosage levels and used for training or inference. In this manner, the disclosed embodiments permit reuse of existing libraries of trial data for drug discovery purposes. For at least these reasons, the disclosed embodiments amount to a technical improvement in, at least, the field of drug discovery.

[0024] Consistent with disclosed embodiments, observational data can include physiological data (e.g., electrophysiological data, such as electroencephalography signals, electrocardiography signals, electromyography signals, neural recordings, or the like; heartAttorney Docket No.16048.0009-00304 rate; respiration; oxygenation; thermal data, such as core body temperature, ocular temperature, paw temperature, or the like; or other physiological data) activity data (e.g., position, motion, force, head location, body center, paw positions, or the like); and / or other suitable measures. In some embodiments, observational data can be acquired and stored for subsequent processing.

[0025] Consistent with disclosed embodiments, the observational data can be or include multiple channels of repeated measurements over time (e.g., time series date). For example, the observational data can be or include multiple channels of EEG signals. As may be appreciated, such signals can be acquired at up to 20kHz. Accordingly, the number of measurements included in a channel can be extremely large. For example, a channel of EEG data collected at a sampling rate of 1 kHz for 30 minutes can include nearly 2 million measurements. As such, it will be recognized that the observational data can use large amounts of memory, such as memory on a graphics processing unit.

[0026] Consistent with disclosed embodiments, the observational data can be acquired during the trial using one or more testing systems. A testing system can include a controlled environment configured with sensors. The testing system can use the sensors to acquire observational data from a non-human animal disposed within the controlled environment during the trial. The non-human animal may be administered the compound by any suitable method (e.g., by injecting the compound or feeding the compound to the animal). When observational data is obtained from multiple trials using multiple testing systems, the trials can be performed using the same non-human animal (or different animals of the same type of non-human animal).

[0027] In some embodiments, a trial can include a pre-dosing and post-dosing period. In some embodiments, the pre-dosing period can be between five minutes and two hours, orAttorney Docket No.16048.0009-00304 more. In some embodiments, the post-dosing period can be between 10 minutes and four hours, or more. In some embodiments, observational data acquired in the pre-dosing period can be used to normalize observational data acquired in the post-dosing period. In some embodiments, normalized observational data from the post-dosing period can be applied to a machine-learning model to predict the effect of the compound in humans. In some embodiments, both pre-dosing and (unnormalized) post-dosing observational data can be applied to a machine-learning model to predict the effect of the compound in humans.

[0028] Consistent with disclosed embodiments, the sensors of a testing system can include electrophysiological recording devices (e.g., electroencephalography (EEG) recording devices, or the like), cameras, piezoelectric sensors, infrared detectors, radiofrequency detectors, and the like. In some embodiments, a sensor can include both an emitter such as an infrared beam, a radiofrequency, a source of heat, a source of optical signals, or other such source, as well as a receiver, such as a component configured to receive data or electronic communications. Multiple cameras can be used to acquire depth or 3D information. The particular configuration of sensors can be selected or adapted based on the subject and / or the compound administered.

[0029] Consistent with disclosed embodiments, the sensors may be in contact with the research animal (e.g., electrophysiological sensor electrodes, actigraphy accelerometers, etc.), in the experimental chamber (cameras, piezoelectric sensors, accelerometers connected to the chamber, etc), or at a distance from the experimental chamber (cameras, infrared sensors, etc), or a combination thereof.

[0030] FIG.1A depicts an exemplary schematic of a system 110 for acquiring (and optionally analyzing) observational data concerning a non-human animal 115, consistent with disclosed embodiments. System 110 may include a control computer 50 communicating 135Attorney Docket No.16048.0009-00304 with a computer-controlled enclosure 120. System 110 can include sensors configurable to acquire data during an experiment on a non-human animal 115. System 110 can include a computer-controlled enclosure 120 configured to contain the non-human animal 115 during the experiment. In some embodiments, the computer-controlled enclosure 120 can include actuators configurable to interact with non-human animal 115. For example, sensors and / or actuators may include an aversive stimulus probe 118 that may be deployed and retracted, a motor challenge 126 that may be deployed or retracted so as to force non-human animal 115 to walk on a plurality of physical obstacles 124 arranged in an array with a predefined pitch to provide a motor challenge, lighting 104 that may be configured to change illumination intensity levels and / or light wavelengths as applied to the non-human animal 115, a tactile stimulator 116 to administer tactile stimuli to the non-human animal 115, a top camera 106, a top thermal camera 102, a first side camera 128, one or more second side camera(s) 114, a floor force sensor 122, waterers and feeders 129, and additional actuators for applying any additional suitable stimuli to the non-human animal 115 and / or any additional sensors.

[0031] In some embodiments, computer-controlled enclosure 120 can include social pod 112 (e.g., shown as an opening in enclosure 120 in FIG.1A that enables social pod 112 to be affixed to enclosure 120 via the opening). Affixing social pod 112 to computer-controlled enclosure 120 can enable a second non-human animal in social pod 112 to interact with non- human animal 115.

[0032] In some embodiments, computer-controlled enclosure 120 can include a 3D camera. A 3D camera can be implemented using multiple 2D cameras configured and arranged around enclosure 120 to obtain 3D image data. In some embodiments, the 3D camera can be implemented using at least one of top camera 106, the first side camera 128, the second side cameras 114, or additional cameras.Attorney Docket No.16048.0009-00304

[0033] In some embodiments, the first side camera 128 and the second side cameras 114 may be oriented to capture images in the computer-controlled enclosure 120 from different perspectives. An angle between the orientation of the first side camera 128 and each of the one or more second side cameras 114 may be any suitable angle to capture movement throughout the computer-controlled enclosure 120. For example, the angle may be 90 degrees, though other angles may be used, such as, e.g., any angle from about 1 degree to about 179 degrees, such that imagery from both the first side camera 128 and each of the one or more second side camera 114 may be processed to determine movement within the computer-controlled enclosure 120.

[0034] FIG.1B depicts a view of computer-controlled enclosure 120, considered with disclosed embodiments. Depicted in this view are the non-human animal 115, the top camera 106, the thermal camera 102, the adverse stimulus probe 118, the tactile stimulator 116, waterers and feeders 129, motor challenge 126, and sensor interface circuitry 168. Also depicted is the location at which social pod 112 can be attached to computer-controlled enclosure 120. FIG.1C depicts a view of social pod 112, which can be attached to computer- controlled enclosure 120 at the location depicted in FIG.1B.

[0035] In some embodiments, the plurality of sensors may include sensors associated with some actuators to capture specific responses to the actuators. In some embodiments, the plurality of sensors may be combined with multiple actuators to challenge the test subject to react to various events, which are recorded and analyzed by the system. The resulting ethophysiogram, or collection of physiological and behavioral responses, may create a dataset (e.g., a content-rich dataset) suitable for use with the disclosed systems and methods.

[0036] Consistent with disclosed embodiments, the control computer 150 may include at least one processor 160 for executing suitable computer applications as described herein forAttorney Docket No.16048.0009-00304 performing the analysis of the behavior of the non-human animal 115, at least one memory and / or suitable storage device denoted by memory 162 for storing the computer code and any databases used in the analyses of the acquired data over the predetermined time period, control circuitry 164 for controlling the plurality of actuators in accordance with the experimental plan, sensor interface circuitry 168 for outputting data from the plurality of sensors, image device interface circuitry 170 for receiving the output data from any or all of the cameras and / or thermal cameras, input and output (I / O) devices 172, and / or communication circuitry 192 to enable the control computer 150 to communicate over any suitable communication network.

[0037] In some embodiments, the I / O devices 172 may include, for example, a display 186 and / or a keyboard 184. Keyboard 184 may allow a user or operator of system 110 to input data to the control computer 150. At least one processor 160 may control a graphic user interface (GUI) 188 shown on the display 186. The GUI 188 may display any suitable parameters and / or data visualizations related to the experimental session and / or results of analyses of the data acquired by the plurality of sensors coupled to enclosure 120 in accordance with the experimental plan.

[0038] In some embodiments, a head mount 190 may be placed on the subject’s head (e.g., skull). The head mount may include at least one electrode to measure brain electrical activity such as electroencephalographic (EEG) signals, for example. In some embodiments, the head mount 190 may also include at least one accelerometer. The signals from the at least one electrode and / or the at least one accelerometer in the head mount 190 may be coupled via wires to circuitry 108 that can relay the signals for processing to the at least one processor 160.Attorney Docket No.16048.0009-00304

[0039] In some embodiments, system 110 can be configured to acquire electrophysiological data, such as pharmaco-EEG (pEEG) data or the like. System 110 can be configured to record actigraphy and quantitative pEEG from one or more brain regions of unanesthetized non- human animals before and after administration of a compound. Such electrophysiological data can be used, consistent with disclosed embodiments, to identify novel compounds that have a desired effect on electrophysiological activity (e.g., pEEG activity).

[0040] In some embodiments, system 110 can be configured to phenotype disease models including autism spectrum disorder, rare genetic epilepsies, Huntington Disease, Alzheimer’s Disease, or the like. As electrophysiological data can yield pharmaco-dynamic signatures specific to pharmacological action, such data can be used to evaluate translational biomarkers in CNS disorders and rapidly screen compounds for potential activity at specific pharmacological targets to provide valuable information for guiding the early stages of drug development.

[0041] In some embodiments, the observational dataset can include or depend upon an EEG signal acquired from the non-human animal. In some embodiments, the observational dataset can be or include the time-dependent frequency components of the EEG signal. For example, the observational data can be or include EEG frequency band data (e.g., signal amplitude or power in the delta, theta, alpha, beta, or gamma bands) or spectrogram data. For convenience of explanation, the disclosed embodiments will be described with respect to spectrogram data generated from a single EEG channel. An observational dataset may include multiple such spectrograms corresponding to multiple EEG signals. However, the disclosed embodiments are not so limited. For example, the observational dataset can include EEG signals (e.g., times series of voltage or current amplitudes, or the like), or can be a multivariate dataset additionally including other physiological data or activity data of the non-human animal.Attorney Docket No.16048.0009-00304

[0042] The disclosed embodiments are not limited to any particular parameters for generating the spectrogram data from the EEG signal (e.g., frequency range, window duration or shape, sampling frequency, or window overlap). Suitable spectrogram generation parameters can be determined based on the animal and the observed frequency components of the EEG signal. In some embodiments, the frequency range can be between a lower bound less than 1, 5, or 10 Hz and an upper bound greater than 5, 10, 20, 50, or 100 Hz. In some embodiments, the spectrogram can include, for example, 32 to 512 frequency values, or more. These frequency values can cover a frequency range from a lower value (e.g., a value between 1 and 100 Hz) to a higher value (e.g., a value between 50 and 500 Hz). The frequency values can be distributed within the frequency range uniformly, logarithmically, or according to some other distribution. In some embodiments, the spectrogram can include 30 to 480 samples, or more. These samples can cover a duration of 30 minutes to 4 hours (e.g., corresponding to a sample period ranging from 3 seconds to 10 minutes). In some embodiments, the samples can be distributed uniformly, logarithmically, or according to some other distribution within the duration of the trial.

[0043] FIG.2 depicts observational datasets generated using data acquired during a set of trials, consistent with disclosed embodiments. In each trial, a non-human animal was administered at different dosage of Psilocybin. A sample, as described herein, can include the observational datasets arranged in a sequence of dosage levels, based on the administered dosages of Psilocybin. Here the observational datasets are arranged in a sequence of increasing dosage levels. As may be appreciated, the order in which the trials were conducted need not match the sequence of dosage levels. In this example, the observational dataset comprises EEG spectrograms generated from EEG signals acquired during the trials. Each EEG spectrogram corresponds to approximate 60 mins of trial time. The EEG spectrograms begin upon the administration of the Psilocybin and depict the normalized power of the EEGAttorney Docket No.16048.0009-00304 signal in frequency bands from zero to 100 Hz. In this example, the power was normalized against a baseline recording taken prior to the administration of Psilocybin. The normalized power is depicted as the percent change in power as compared to the baseline recording.

[0044] As can be observed from FIG.2, the response to Psilocybin in this set of trials was characterized by an increase in power at lower frequencies that developed over time following administration of the compound.

[0045] FIG.3 depicts an exemplary method 300 for training a machine learning model to predict therapeutic effect(s) of a compound when administered to a human, consistent with disclosed embodiments. The machine learning model can be trained to predict the effect(s) of the compound when administered to a human based on the effect(s) of administering the compound at varying dosages to one or more non-human animals. In some embodiments, the machine learning model can be trained to predict the effect(s) of the compound when administered to a human at a particular dosage. In some embodiments, the input to the machine learning model can be a sequence of observational datasets corresponding to trials at a sequence of dosage levels of the compound. For convenience of description, such observational datasets are described as being EEG spectrogram(s). However, the observational datasets can alternatively or additionally include other physiological or activity data, as described herein. Consistent with disclosed embodiments, method 300 can include generating a suitable training dataset and training the machine learning model.

[0046] For convenience of discussion, method 300 is described as being performed by a training system. However, this description is not intended to be limiting. In some embodiments, multiple systems can collectively perform method 300. In some embodiments, multiple systems can collectively perform one or more steps of method 300 (e.g., multiple systems can coordinate obtaining a dataset and processing it into a training dataset). In someAttorney Docket No.16048.0009-00304 embodiments, different systems can perform different steps of method 300 (e.g., a first system can generate a training dataset and a second system can train the machine learning model using the training dataset).

[0047] In step 310 of method 300, the training system can obtain an original dataset 310, consistent with disclosed embodiments. In some embodiments, the training system can receive or retrieve at least a portion of the original dataset. For example, the training system can access memory, database, or another system to obtain suitable observational datasets. Similarly, a user of the system can provide the training system with at least a portion of the original dataset. The observational datasets can include measurements obtained from non- human animals during trials in which compounds were administered to the non-human animals. In some embodiments, the training system can create at least a portion of the original dataset. For example, the training system can be or include one or more systems for acquiring observational data (e.g., system 110, or the like). A user of the system for acquiring observational data can conduct trials, which can provide observational data to the training system.

[0048] In step 320 of method 300, the training system can create samples for a training dataset, consistent with disclosed embodiments. The samples can be created using the original dataset obtained in step 310. In some embodiments, a sample can include a sequence of a predetermined number of observational datasets. The machine learning model can include a corresponding sequence of encoders.

[0049] In some instances, creating the samples can include converting or formatting the observational data into a format expected by the machine learning model. For example, when the machine learning model expects a sequence of EEG spectrograms, the training system can generate suitable EEG spectrograms from EEG signals. Such generation can includeAttorney Docket No.16048.0009-00304 preprocessing the EEG signals, prior to generating suitable EEG spectrograms (e.g., cropping or zero-padding the EEG signal to accord for differences in trial length, filtering the EEG signal, removing, or correcting artifacts or noise, changing or normalizing EEG signal amplitudes, resampling the EEG signal, or other suitable operations). The training system can then generate EEG spectrograms from the processed EEG signals according to any suitable method.

[0050] In some instances, creating the samples can include performing data augmentation operations. Such operations can include generating observational datasets using other observational datasets, generating multiple samples using the same observational datasets, generating samples using observational datasets acquired from multiple different animals, or other suitable data augmentation operations. As may be appreciated, such data augmentation operations can improve the performance of the machine learning model. For example, such data augmentation can reduce overfitting, address imbalances in amounts of training data for different therapeutic classes in the original dataset (e.g., arising from unequal numbers of trials having been performed with different compounds), improve the robustness of the machine learning model to inter-subject variability, improve the robustness of the machine learning model to missing observational data (e.g., missing dosage levels), and the like.

[0051] Consistent with disclosed embodiments, the training system can be configured to generate observational datasets from other observational datasets. For example, the training system can be configured to generate an observational dataset by combining other observational datasets.

[0052] In some instances, the generated observational dataset and the other observational datasets can correspond to the same combination of compound and dosage level. For example, the training system can be configured to generate an EEG spectrogramAttorney Docket No.16048.0009-00304 corresponding to a 0.1 mg / kg dose of Psilocybin by averaging other EEG spectrograms corresponding to a 0.1 mg / kg dose of Psilocybin. The other EEG spectrograms can correspond to trials conducted in different non-human animals (e.g., a first trial may have been performed using a first rat and a second trial may have been performed using a second rat). The average can be a weighted average. In this manner, the training system can generate additional observational datasets (e.g., to increase the number of samples for a class).

[0053] In some instances, the generated observational dataset and the other observational datasets can correspond to differing dosage levels of the same compound. For example, the training system can be configured to generate an EEG spectrogram corresponding to a 0.32 mg / kg dose of Psilocybin by averaging an EEG spectrogram corresponding to a 0.1 mg / kg dose of Psilocybin (e.g., acquired from a non-human animal, such as a first rat) and an EEG spectrogram corresponding to a 1.0 mg / kg dose of Psilocybin (e.g., acquired from the same or a different non-human animal, such as a second rat). The average can be a weighted average. In this manner, the training system can generate observational datasets corresponding to intermediate dosage levels (e.g., when such observational datasets are not available, or insufficient in number, in the original dataset). For example, the original dataset may include observational datasets corresponding to three repeats for each of four of six dosage levels, but only a single repeat for each of the second and fourth dosage levels. The training system can generate two additional repeats for the second dosage level using observational datasets for the first and third dosage levels (e.g., an EEG spectrogram for the second level can be generated as a weighted average of EEG spectrograms obtained for the first and third dosage levels). The training system can also generate two additional repeats for the fourth dosage level using observational datasets for the third and fifth dosage levels (e.g., an EEG spectrogram for the fourth level can be generated as a weighted average of EEG spectrogramsAttorney Docket No.16048.0009-00304 obtained for the third and fifth dosage levels). In this manner, the training system can ensure that there are three repeats for each of the six dosage levels.

[0054] Consistent with disclosed embodiments, the training system can be configured to generate multiple samples using the same observational datasets. For example, the original data set may include observational datasets corresponding to three repeats each for six dosage levels. The repeats may or may not have been performed in the same animal. The training system can use these observational datasets to generate, for example, 729 samples (e.g., a sample for each unique combination of repeat and dosage level). As may be appreciated, additional “repeats” can be generated by averaging pairs of samples within each dosage level, permitting the generation of additional samples.

[0055] As a nonlimiting example, a compound has been administered at three dosages to two mice (e.g., ^^^^for ^^ ൌ 1,2,3 and ^^ ൌ 1,2) in six trials. In some embodiments, the training system can be configured to generate additional samples by blending samples at different dosages and / or in different animals.

[0056] As a non-limiting example, additional sample can be generated as a linear combination of existing samples acquired in trials using different dosages and / or in different animals: ^^^ൌ ^ ^^^^^^^where ^^ ∈ animals, ^^ ∈ dosages, and ^^^^∈ ^0,1^ and∑^^^^ൌ 1. In some embodiments, such a linear combination can include samples drawn from fewer than two animals subjects. In some embodiments, when generating a sample for a dosage level, such a linear combination can include only samples at the dosage level, or at adjacent dosage levels. For example, if there are six dosage levels and the sample is being generated for the fifthAttorney Docket No.16048.0009-00304 dosage level, the linear combination can only include samples obtained at the fourth, fifth, or sixth dosage levels.

[0057] As a nonlimiting example, a compound has been administered at three dosages to two mice (e.g., ^^^^for ^^ ൌ 1,2,3 and ^^ ൌ 1,2) in six trials. The training system can be configured to generate additional samples ^^^^for ^^ ^ 2 using ^^ ൌ ^ ^^^^^ for ^^ ^ 2. In some embodiments, ^^^^வଶൌ ^^ ^^^ ^^்^^^^, where ^^ is thetrials and ^^^is an n by m blending matrix for generating dosage ^^ by blending samples at across dosages and animals (where n is the number of dosages, e.g., three in this example, and m is the number of animals, e.g., two in this example). For example: ^^^^^^^ଶ^^^^ ^^^ଶ^^ ൌ ^ ^^ଶ^^^ଶଶ൩ , ^^^ ൌ ^ ^^ଶ^ ^^ଶଶ൩,^In the above example, ^^^^∈ ^0,1^ and ∑ ^^^^ൌ 1. In some embodiments, the blending matrix can be further parameterized. For^^ ^^ ^^^1 െ ^^^^^^^^^^ ^^ ^^^^∙∙Thus, the first dosage of the third generated sample can be a combination of the observational data for the first and second dosages of the two measured samples, according to the blending coefficients 0 ^ ^^, ^^ ^ 1. In this example, ^^ controls the weight given to samples at the first dosage level versus the second dosage level, while ^^ controls the weight given to the samplesAttorney Docket No.16048.0009-00304 at the first animal versus the second animal. In some embodiments, the value of ^^ can be chosen randomly each time a new sample is generated.

[0058] In some embodiments, the training system can have a set of blending matrices for each dosage level and select among them (e.g., deterministically, or according to a probability distribution). For example, the training system can generate a second dosage level by selecting among three blending matrices ^^ଶ|^, for ^^ ൌ 1,2,3 according to the following probabilities: ^^ 0^^ଶ|^ൌ^0 ^1 െ ^^^ ൩, ^^ ൌ 33%In these three differentcombination of a dosage at a first dosage level in the first animal combined with a dosage at another (sometimes different) dosage level in the second animal. The degree of contribution from each animal depends on the value of 0 ^ ^^ ^ 1. In this example, there is no blending across different dosage levels within an animal, but the disclosed embodiments are not so limited.

[0059] As may be appreciated, the disclosed embodiments are not limited to embodiments that expressly perform matrix multiplication using blending matrices (or expressly sample such blending matrices). Instead, the disclosed embodiments can include approaches that similarly generate new samples by combining observational data for different dosage levels and / or different animals.Attorney Docket No.16048.0009-00304

[0060] In some embodiments, the training system may include multiple observational datasets corresponding to the same dosage in a sample (e.g., duplicate observational datasets) for a compound. The number of duplicate observational datasets included in the sample can depend on the number of dosages for which observational datasets are available for the compound in the original dataset. The number of duplicate observational datasets included in the sample can further depend on the architecture of the machine learning model (e.g., the number of encoders in the machine learning model). For example: ^^ௗൌ max ^^ ^^^െ ^^^^, 0^ where ^^ௗis the number of duplicate observational datasets, ^^^is the number of encoders, and ^^^is the number of dosage levels for which observational data is available in the original dataset.

[0061] As may be appreciated, the training system can be configured to associate the same dosages with multiple differing dosage levels when creating a sample. For example, given a sequence of six dosage levels (e.g., ^ ^^^, ^^ଶ, ^^ଷ, ^^ସ, ^^ହ, ^^^^) and an original dataset including, for a particular compound, observational datasets corresponding to three dosages (e.g., ^ ^^^, ^^ଶ, ^^ଷ^), the training system can be configured to map the same dosage to multiple dosage levels. For example, training system can map the lowest dosage to multiple dosage levels (e.g., ^ ^^^, ^^^, ^^^, ^^^, ^^ଶ, ^^ଷ^). As an additional example, training system can map the highest dosage to multiple dosage levels (e.g., ^ ^^^, ^^ଶ, ^^ଷ, ^^ଷ, ^^ଷ, ^^ଷ^). As a further example, the training system can repeat each dosage once (e.g.,^^^^, ^^^, ^^ଶ, ^^ଶ, ^^ଷ, ^^ଷ^). After mapping dosages to dosage levels in the sequence, the training system can select, for each dosage level, an observational dataset corresponding to the dosage associated with that dosage level. These observational datasets can be included in the sample.Attorney Docket No.16048.0009-00304

[0062] In the prior example, the machine learning model can be configured to accept a sample including observational data for six dosage levels. The training system can be configured to obtain a sequence of observational data for the first three dosage levels (e.g., ^ ^^^^, ^^ଶଶ, ^^ଷସ^, with ^^^^being observation data from a first animal at a first dosage level, ^^ଶଶbeing observation data from a second animal at a second dosage level, and ^^ଷସbeing observation data for a third dosage level created using blending matrices as described herein, or the like). The training system can be configured to generate subsequences of these three dosage levels and then pad them out to the final sequence length of six samples. For example: ^ ^^^^, ^^ଶଶ^ → ^ ^^^^, ^^ଶଶ, ^^ଶଶ, ^^ଶଶ, ^^ଶଶ, ^^ଶଶ^ ^^^^^, ^^ଶଶ, ^^ଷସ^→^^^^^, ^^ଶଶ, ^^ଷସ, ^^ଷସ, ^^ଷସ, ^^ଷସ^

[0063] In some embodiments, the same dosages can be associated with the same positions in the sequence of dosage levels for samples corresponding to the same compound. For example, all samples corresponding to Psilocybin may include three dosage levels corresponding to dosages of 0.1 mg / kg, 0.32 mg / kg, and 1.0 mg / kg. However, different dosages can be associated with the same positions in the sequence of dosage levels for samples corresponding to the different compounds. For example, a dosage of 0.1 mg / kg, can correspond to the first dosage level for a sample corresponding to Psilocybin. As a further example, dosages of 5 mg / kg, 10 mg / kg, and 20 mg / kg can correspond to the first, second, and third dosage levels for samples corresponding to Tiagabine. As a further example, dosages of 5 mg / kg, 20 mg / kg, and 40 mg / kg can correspond to the first, second, and third dosage levels for samples corresponding to Nefazodone. As a further example, dosages of 2 mg / kg, 5 mg / kg, and 10 mg / kg can correspond to the first, second, and third dosage levels for samples corresponding to Oxazepam.Attorney Docket No.16048.0009-00304

[0064] In some instances, the training system (or a user thereof) may obtain a mapping of dosages to dosage levels for a compound. For example, historical trial data may exist for ten dosages of a compound, but the machine learning model may be configured to use only six dosage levels. The training system may therefore need to select a subset of the historical data for inclusion in the training dataset. As an additional example, a study may specify that trials will be performed at six dosage levels (e.g., because a machine learning model is configured to use only six dosage levels). As may be appreciated, the study may benefit from selecting, for the six dosages levels, dosages ranging from sub-therapeutic to supra-therapeutic dosages. Selecting such a range of dosages may enable the study to provide more suitable data for classification of therapeutic effects than a selection of dosages that are either entirely sub- therapeutic or entirely supra-therapeutic.

[0065] Accordingly, consistent with disclosed embodiments, a mapping for a compound from a dosage to a position in the sequence of dosage levels can be determined manually (or at least in part automatically). In some instances, a user of the training system can specify the mapping based on prior knowledge of the effect of the compound in the non-human animal tested. For example, the user can map dosages to dosage levels based on a known sigmoidal relationship between dosage and observational data. The user can map dosages to dosage levels to cover the sub-linear, linear, and supra-linear regions of the sigmoidal relationship. In some instances, the training system can determine a relationship between dosage levels and observational data for one or more compounds (e.g., based on observational datasets in the original data). The training system may then create or suggest a mapping from dosage to dosage level for the compound. This mapping can be used to select historical data (e.g., by selecting observational data obtained in trials in which the dosage of a compound administered matched a specified dosage level of that compound) or guide future studies.Attorney Docket No.16048.0009-00304

[0066] In step 330 of method 300, the training system can associate samples with therapeutic class(es), consistent with disclosed embodiments. The samples can include samples generated in step 320 of method 300 (or otherwise obtained).

[0067] In some embodiments, the training system can associate a sample corresponding to a compound with therapeutic class(es) based on known therapeutic effect(s) of the compound when administered to humans. For example, a sample corresponding to Psilocybin can be associated with the therapeutic class “hallucinogen” based on the hallucinogenic effects of Psilocybin in humans. Likewise, a sample corresponding to Sertraline can be associated with the therapeutic class “antidepressant” based on the antidepressant effects of Sertraline in humans.

[0068] In some embodiments, the training system can associate dosage levels in a sample with therapeutic class(es). As described herein, each dosage level can be associated with a dosage, and the dosage level can be associated with therapeutic class(es) based on known therapeutic effect(s) of the compound when administered to humans in that dosage. In some embodiments, the sample can therefore be associated with a vector of indications of therapeutic class(es). For example, when administered to humans at a first dosage, a compound (e.g., a benzodiazepine) may act as an anxiolytic. The dosage level in the sample corresponding to the first dosage can be associated with an indication of the therapeutic class “anxiolytic.” However, when administered to humans at a second, higher dosage, the compound may act as a sedative. The dosage level in the sample corresponding to the second dosage can be associated with an indication of the therapeutic class “sedative.”

[0069] In some embodiments, the therapeutic class(es) can include antidepressant (MAO inhibitors, adrenergic and serotonergic receptors antagonists, mixed or specific serotonin and norepinephrine reuptake inhibitors, triple monoamine reuptake inhibitors, tricyclicAttorney Docket No.16048.0009-00304 antidepressants, etc.), analgesic (opiates, NSAIDs, etc.), anxiolytic (serotonin 1A receptor agonists, benzodiazepines, mGLUR5 antagonists, neurosteroids, etc.), antipsychotic (typical, atypical, and other novel antipsychotics), cognitive enhancer (nicotinic cholinergic agonists, cholinesterase inhibitors, phosphodiesterase-4 inhibitors, histamine receptor ¾ antagonists, etc.), anticonvulsant (GABAb activators, channel blockers, etc.), hallucinogen (NMDA antagonist, serotonin 2a receptor agonist, other mixed mechanisms of action, etc.), mood stabilizer (tricyclic mood stabilizers, lithium, channel blockers, GABA reuptake inhibitors, etc.), psychostimulant and / or ADHD medication (monoamine releasers, reuptake inhibitors and agonists, mGLUR5 antagonists, etc.), anesthetic, or sedatives and hypnotics (histamine 1 receptor antagonists, benzodiazepines, adrenergic agonists, etc.). In some embodiments, the therapeutic class(es) can include a therapeutic antidepressant, high dose side-effect antidepressant, hallucinogen / psychedelic / treatment-resistant depression treatment, therapeutic antipsychotic, high dose antipsychotic, anxiolytic, sedative, cognitive enhancer / Alzheimer's treatment, psychostimulant / ADHD, mood stabilizer / anticonvulsant, side effect / toxic, other therapeutic area / inactive, sedative, vehicle, anxiogenic, analgesic, or the like. As may be appreciated, the disclosed embodiments are not limited to any particular enumeration of therapeutic classes.

[0070] The disclosed embodiments are not limited to any particular implementation of a sample, or to any particular implementation of associating a therapeutic class with a sample. In some embodiments, a sample can include observational datasets. For example, a sample can be a data structure that contains the observational datasets. In some embodiments, the sample can include one or more references to one or more observational datasets. For example, the observational datasets can be stored in a database and a sample can include references to particular observational datasets stored in the database. In some embodiments, associating a sample with therapeutic class(es) can include storing an indication of theAttorney Docket No.16048.0009-00304 therapeutic class(es) in the sample (e.g., the sample can be a data structure containing an element that indicates the therapeutic class(es)). In some embodiments, the sample can include a reference that associates the sample with an indication of therapeutic class(es). For example, a first database can associate samples with key values, and a second database can associate the key values with indications of therapeutic class(es).

[0071] Consistent with disclosed embodiments, the training system can be configured to collect samples associated with therapeutic classes into a training dataset. In step 340 of method 300, the training system can train a machine learning model to predict therapeutic class(es) of a compound from a sample, consistent with disclosed embodiments. The predicted therapeutic class(es) can indicate therapeutic effect(s) of the compound when administered to a human. In some embodiments, therapeutic class(es) can be predicted for the overall sample. In some embodiments, therapeutic class(es) can be predicted for each dosage level in the sequence of dosage levels.

[0072] Consistent with disclosed embodiments, the machine learning model can include an encoder portion and a classifier portion. In some embodiments, the encoder portion can include one or more neural networks (e.g., feed-forward neural networks, recurrent neural networks, attention-based neural networks, or the like). In some embodiments, the classifier portion can include one or more neural networks, random forests or decision trees, regression classifiers, support vector machine, nearest neighbor classifiers, or other suitable classifiers. The encoder portion can map the sample to a position in a high-dimensional space. The classifier portion can map the position in the high-dimensional space to the predicted therapeutic class(es). In some embodiments, the encoder and classifier portions of the machine learning model can be trained together. In some embodiments, the encoder and classifier portions can be trained at least partially separately.Attorney Docket No.16048.0009-00304

[0073] As described herein, in some embodiments, samples can include multiple portions of observational data acquired using multiple testing systems. In such embodiments, the machine learning model can be adapted to accommodate the different portions of observational data. Such accommodation can occur in the encoder portion of the machine learning model or the classifier portion of the machine learning model.

[0074] In some embodiments including multiple portions of observational data, the encoder portion of the machine learning model can include multiple sub-encoders. Each sub-encoder can be configured to encode a portion of observational data acquired by a corresponding testing system. For example, a first testing system can be configured to acquire first observational data using a first collection of sensors (e.g., electrophysiological data, or the like). Another testing system can be configured to acquire observational data using a second, differing set of sensors (e.g., visual or infrared cameras, accelerometers, piezoelectric sensors, or the like). A first sub-encoder can correspond to and process the first observational data into a first encoding. A second sub-encoder can correspond to and process the second observational data into a second encoding. A classifier portion of the machine learning model can be configured to generate a classification from the encodings generated by the multiple sub-encoders. In some embodiments, sub-encoders can correspond to each combination of observational data portion and dosage level. For example, where there are two testing environments and six dosage levels, there may be 12 sub-encoders.

[0075] In some embodiments including multiple portions of observational data, the encoder portion of the machine learning model can include a single encoder. The single encoder can be configured to accept observational data acquired from any one of multiple testing systems. In some embodiments, a source feature can be input to the encoder. The source feature can specify which testing system generated the observational data. Based on the source feature and the observational data, the encoder can generate an encoding that is input to the classifier.Attorney Docket No.16048.0009-00304

[0076] In some embodiments including multiple portions of observational data, the classifier portion of the machine learning model can include multiple sub-classifiers. In some embodiments, each sub-classifier can receive an encoding from a corresponding sub-encoder. For example, a first portion of observational data generated using a first testing system can be input to a first sub-encoder to generate a first encoding. The first encoding can be input to a first sub-classifier. The first sub-classifier can be associated with the first testing system.

[0077] In some embodiments, the sub-encoders can share a common stem or base layer. For example, one or more encodings can be input to a common base layer of the classifier. Subsequently the classifier can branch into multiple sub-classifiers.

[0078] Consistent with disclosed embodiments, the output of multiple sub-classifiers can be combined to generate a final classification output. In some embodiments, the multiple outputs can be combined using one or more common output layers. For example, the outputs of two sub-classifiers can be combined into an input for a final classification layer or set of classification layers. In some embodiments, the multiple outputs can be combined using ensemble learning techniques (e.g., using a Bayesian classifier, bagging, boosting, model combination or averaging, or the like).

[0079] In some embodiments, the machine learning model can be configured to determine whether a compound has unspecified activity. In such embodiments, the machine learning model can include an activity detector portion, in addition to the therapeutic prediction portion discussed above. The activity detector portion of the machine learning model can be implemented using a separate encoder and classifier, a separate classifier, or a separate classifier branch that shares one or more layers with other classifier(s) in the machine learning model. The activity detector portion can be configured to classify the compound as “active” (or an equivalent label) or “inactive” (e.g., vehicle, control, default, or someAttorney Docket No.16048.0009-00304 equivalent label). In some embodiments, the output of the overall machine learning model can depend on the output of the therapeutic prediction portion (e.g., the therapeutic class(es) predicted for the different dosage levels) and the output of the activity detector portion.

[0080] In some embodiments, the therapeutic prediction portion and the activity detector portion can output probability values. When the sum of the probabilities for all the therapeutic class(es), for a dosage level and excluding any “control” or semantically similar class (e.g., “vehicle,” “no response”, “default,” or the like), exceeds the output probability for the “activity” class of the activity detector portion, then the overall output of the machine learning model can be the probabilities output by the therapeutic prediction portion. Otherwise, the overall output of the machine learning model can indicate that the class of the compound is unknown. In some embodiments, a suitable probability value can be assigned to an “unknown – active” class, or a semantic equivalent. In some embodiments, a suitable probability value can be assigned to a “control” or semantically similar class.

[0081] A machine learning model can be trained according to any suitable training arrangement, consistent with disclosed embodiments. In some embodiments, the training dataset can be divided into at least a training portion and a validation portion. The machine learning model can be trained using the training portion of the training dataset and validated using the validation portion of the training dataset. The machine learning model can be initialized and training hyperparameters selected according to known methods.

[0082] In some embodiments, when the observational data includes multiple portions of observational data from multiple testing system, sub-machine learning models can be trained separately. For example, a first sub-machine learning model can be trained using the observational data from the first testing system and a second sub-machine learning model can be trained using observational data from the second testing system. The two sub-machineAttorney Docket No.16048.0009-00304 learning models can then be combined. For example, a new architecture can be trained using a unified classifier and the two encoders of the sub-machine learning models. As an additional example, the outputs of the two existing classifiers can be combined (e.g., using ensemble techniques, or the like). In some embodiments, the combined models can be further trained. Such training can be performed using a loss function configured to prioritize contributions from sub-machine learning models known to have superior performance with particular classes of outputs (e.g., anticonvulsants versus sedatives, or the like).

[0083] In some embodiments, training can be performed in batches. In each batch, a set of samples can be applied to the machine learning model to generate a set of predicted therapeutic class(es). The predicted therapeutic classes can be compared to the “ground truth” therapeutic class(es) associated with the samples to generate an update. The machine learning model can then be updated based on the update. For example, weights or parameters of the machine learning model can be updated based on the update.

[0084] Consistent with disclosed embodiments, the training system can train the machine learning model until a termination condition is satisfied. The termination condition can depend on the performance of the model (e.g., accuracy, precision, recall, a confusion matrix, a loss function value, or another suitable measure), a training time, an amount of data trained upon (e.g., a number of epochs, or the like), or any other suitable measure.

[0085] Consistent with disclosed embodiments, the training system can store the trained machine learning model, use the trained machine learning model in a production environment (e.g., to generate predictions of therapeutic classes for compounds lacking known therapeutic effects), or provide the trained machine learning model to another system.

[0086] FIG.4 depicts an exemplary method 400 for predicting a therapeutic effect of a compound on a human using a trained machine learning model, consistent with disclosedAttorney Docket No.16048.0009-00304 embodiments. The machine learning model can be trained as described with regards to method 300. The machine learning model can include an encoder portion and a classifier portion, as described herein. The machine learning model accepts as input a sample comprising a sequence of observational datasets. For convenience of description, such observational datasets are described as being EEG spectrogram(s). However, the observational datasets can alternatively or additionally include other physiological or activity data, as described herein. The machine learning model can predict therapeutic effect(s) based on the input sample.

[0087] For convenience of discussion, method 400 is described as being performed by a production system. However, this description is not intended to be limiting. In some embodiments, multiple systems can collectively perform method 400. In some embodiments, multiple systems can collectively perform one or more steps of method 400 (e.g., multiple systems can coordinate to generate encodings). In some embodiments, different systems can perform different steps of method 400 (e.g., a first system can obtain an observational dataset and a second system can generate encodings, generate an indication of therapeutic effect(s), and provide the indication of therapeutic effect).

[0088] In step 410 of method 400, the production system can obtain a sample, consistent with disclosed embodiments. Consistent with disclosed embodiments, the sample can include a sequence of observational datasets corresponding to a sequence of dosage levels. The observational datasets can include or depend upon observational data acquired from non- human animals during trials in which the non-human animals were administered varying dosages of a compound. In some instances, the observational datasets can be or depend upon observational data acquired from the same non-human animal (e.g., the same rodent). In some instances, the observational dataset can include or depend upon observational dataAttorney Docket No.16048.0009-00304 acquired from different non-human animals (e.g., different rodents of the same species, or same strain or type, or the like).

[0089] Consistent with disclosed embodiments, the sequence of dosage levels can include observational datasets corresponding to a predetermined number of dosage levels. However, the dosage values corresponding to these dosage levels can differ from the dosage values used in training the machine learning model. As described herein, the machine learning model considers the sequence of dosage levels, not the dosage values.

[0090] In some embodiments, the production system can receive or retrieve the sample from a user of the production system, a memory, a database, or another system. In some embodiments, the production system can create the sample. In some embodiments, the production system can create the sample using observational data received or retrieved from a user, a memory, a database, or another system. In some embodiments, the production system can create the observational data. For example, the production system can be or include a system for acquiring observational data (e.g., system 110, or the like). A user of the system for acquiring observational data can conduct trials, which can provide observational data to the production system.

[0091] In some embodiments, creating the sample can include converting or formatting the observational data into a format expected by the machine learning model, as described herein with regards to step 320. For example, when the machine learning model expects a sequence of EEG spectrograms, the training system can generate suitable EEG spectrograms from EEG signals. In some embodiments, when the observational data corresponds to an insufficient number of dosage levels, the production system can add duplicate observational datasets to the sample or interpolate additional levels between existing levels.Attorney Docket No.16048.0009-00304

[0092] In step 420 of method 400, the production system can generate a sequence of encodings using the sample, consistent with disclosed embodiments. The production system can generate the sequence of encodings by inputting the sample to an encoding portion of the machine learning model. In some embodiments, the encoding portion of the machine learning model can include a sequence of encoders. In some embodiments, each encoder in the sequence can correspond to a dosage level in the sequence of dosage levels. As described herein, the sample can include observational datasets corresponding to the sequence of dosage levels. For each dosage level, the production system can apply the corresponding observational dataset to the corresponding encoder to generate a separate encoding in the sequence of encodings. For example, the production system can apply the first observational dataset to the first encoder to generate a first encoding corresponding to the first dosage level in the sequence of dosage levels.

[0093] In step 430 of method 400, the production system can predict therapeutic effect(s) of the compound, consistent with disclosed embodiments. The production system can generate the prediction by inputting the sequence of encodings to a classifier portion of the machine learning model.

[0094] In some embodiments, the classifier portion can generate a prediction of therapeutic effect(s) for the overall sample. For example, the classifier portion can output a vector with elements corresponding to therapeutic classes. An element corresponding to a predicted therapeutic class can be one-valued. The remaining elements can be zero-valued. As an additional example, the elements corresponding to the therapeutic classes can have likelihood values. The greater the likelihood value, the more likely the compound belongs to the therapeutic class.Attorney Docket No.16048.0009-00304

[0095] In some embodiments, the classifier portion can generate a prediction of therapeutic effects for each dosage level in the overall sample. For example, the classifier portion can output a vector with elements corresponding to therapeutic classes for each dosage value. The elements of the vector for each dosage value can be zero or one-valued, or can be likelihood values. For example, a compound may act as an anxiolytic at low dosages and a sedative at high dosages. For such a compound, the likelihood values for the element corresponding to the anxiolytic class can decrease as the dosage level increases, while the likelihood values for the element corresponding to the sedative class can increase as the dosage level increases. Alternatively, the element corresponding to the anxiolytic class can have the value one for lower dosage levels (e.g., zero otherwise) and the element corresponding to the sedative class can have the value one for higher dosage levels (e.g., zero otherwise).

[0096] In step 440 of method 400, the production system can be configured to provide an indication of the predicted therapeutic effects. Providing an indication of the predicted therapeutic effect(s) can include displaying the predicted therapeutic class(es) to a user of the production system (e.g., in a graphical user interface), storing an indication the predicted therapeutic class(es) in a memory or database, or providing an indication of the predicted therapeutic class(es) to another system.

[0097] FIG.5 depicts an exemplary architecture 500 of a machine learning model for predicting a therapeutic effect of a compound in humans using observational data obtained from non-human animals, consistent with disclosed embodiments. In some embodiments, a machine learning model trained according to method 300 or used to predict therapeutic effect(s) according to method 400 can be implemented using architecture 500. In accordance with architecture 500, such a machine learning model can be configured to generate, using an encoder portion, n separate encodings (e.g., encodings 520-1 to 520-n) from a sample 501.Attorney Docket No.16048.0009-00304 Such a machine learning model can further be configured to generate class label(s) 599 from the n separate encodings using a classifier 530.

[0098] Sample 501 can include a sequence of n observational datasets corresponding to n dosage levels, where n can be greater than one. For example, n can be between 3 and 10, or more. In some embodiments, each of the observational datasets can include multiple channels. For example, when the observational datasets corresponding to EEG spectrograms, different channels can correspond to different EEG signals obtained by different electrodes (or different electrode pairs, or the like). In some embodiments, each of the observational datasets can include as many as 64 channels, or as many as 256 channels, or more channels. In some embodiments, fewer channels may be used. For example, in some embodiments, each of the observational datasets can include between 2 and 32 channels.

[0099] Consistent with disclosed embodiments, each observational dataset can be applied to a corresponding encoder (e.g., encoder 510-1 to encoder 510-n). In some embodiments, each of the encoders can have the same or similar architectures. In some embodiments, each of the encoders can be or include a convolutional neural network.

[0100] In some embodiments, an encoder can include an input component (e.g., input component 511-1), one or more repeats of a body component (e.g., body component 512-1), and an output component (e.g., output component 516-1).

[0101] In some embodiments, the input component can include one or more convolutional neural network layers configured to convert the observational dataset into a feature set. In some embodiments, the feature set can include more channels, but lesser time and frequency resolution than the input observational dataset. In some embodiments, the input component can include a normalization layer configured to batch normalize the output of the convolutional layers. In some embodiments, the input component can include an activationAttorney Docket No.16048.0009-00304 layer configured to apply a suitable activation function to the output of the normalization layers. In some embodiments, the input component can apply a pooling or dropout layer to the output of the activation layer.

[0102] In some embodiments, each repeat of the body component can be configured to accept input features from a preceding component (e.g., the input component or another body component) and generate output features for provision to a subsequent component (e.g., the output component or another body component). In various embodiments, there can be between 1 and 10 repeats, or more, of the body component in the encoder.

[0103] In some embodiments, a body component can include two processing pathways. A first processing pathway can include one or more convolutional components (e.g., convolutional component 513-1). A convolutional component can include a convolutional layer. A convolutional component can further include additional batch normalization and activation layers. The output of the one or more convolutional components can be applied to an adaptive weighting block (e.g., weighting block 514-1) to reweight the channel in the output of the convolutional layer. The adaptive weighting block can be a squeeze and excitation block (or another suitable adaptive weighting block). A second pathway can include a second convolutional components (e.g., convolutional component 515-1). The second convolutional component can include a convolutional layer. The second convolutional component can further include additional batch normalization and activation layers. As may be appreciated, the second pathway can improve the training characteristics of the encoder by skipping the blocks in the first pathway. In some embodiments, a combination component can generate the output of the body component by combining the outputs of the first and second pathway. In some embodiments, the combination component can include a summation (or other function) layer (and optionally an activation layer).Attorney Docket No.16048.0009-00304

[0104] In some embodiments, the body components can be configured to progressively reduce the time and frequency resolution of the feature sets, while increasing the number of channels. For example, the input layer can generate a feature set having dimensions (16, 58, 62), where 16 is the number of channels and each channel has dimensions 58 by 62. A first body component can accept the feature set having dimensions (16, 58, 62) and produce a feature set having dimensions (64, 29, 31). A final body component can produce a feature set having dimensions (128, 4, 4).

[0105] In some embodiments, the output component can be configured to accept the multi- channel input and generate a single channel output. In some embodiments, the output component can reshape the multi-channel input into a single channel and input that channel to at least one set of layers (e.g., one to ten sets of layers, or another suitable number). Each set of layers can include a feedforward layer (and optionally a batch normalization, activation, and / or dropout layer). Consistent with disclosed embodiments, the output of the output component can be the encoding (e.g., encoding 520-1 to 502-n).

[0106] Consistent with disclosed embodiments, classifier 530 can be configured to accept the encodings. In some embodiments, classifier 530 can be a convolutional neural network. The convolutional neural network can include convolution layers and activation layers. The convolutional neural network can further include batch normalization and dropout layers.

[0107] In some embodiments, classifier 530 can be configured to convert the set of n encodings into an m by 1 vector, where m is a number of possible therapeutic classes. In some embodiments, the m by 1 vector can be a vector of likelihoods, indicating the likelihood that the compound associated with the sample belongs to each therapeutic class. In some embodiments, an element of the m by 1 vector can be one-valued, indicating a most likely therapeutic class, and the remaining elements can be zero-valued.Attorney Docket No.16048.0009-00304

[0108] In some embodiments, classifier 530 can be configured to convert the set of n encodings into n m by 1 vectors, where m is a number of possible therapeutic classes. Each of the n vectors can correspond to a dosage level in the sample. In some embodiments, an m by 1 vector corresponding to a dosage can be a vector of likelihoods, indicating the likelihood that the compound associated with the sample, when administered at the dosage corresponding to the dosage level, belongs to each therapeutic class. In some embodiments, an element of an m by 1 vector can be one-valued, indicating a most likely therapeutic class for the compound, when administered at the dosage corresponding to the dosage level, and the remaining elements can be zero-valued.

[0109] As may be appreciated, architecture 500 is exemplary and not intended to be limiting. In some embodiments, the encoders can be convolutional neural networks having a different internal structure than the structure disclosed in FIG.5. In some embodiments, the classifier may not be a convolutional neural network. Instead, the classifier may be another type of classifier, such as a random forest or decision tree, regression classifier, support vector machine, nearest neighbor classifier, or another suitable classifier.

[0110] FIG.6 depicts an exemplary computing system 600 suitable for generating machine learning models, consistent with disclosed embodiments. In some embodiments, the machine learning system described herein can be implemented using computing system 600. Similarly, methods 300 or 400 can be performed, and / or architecture 500 can be implemented using computing system 600. As will be appreciated by one skilled in the art, the components and arrangement of components included in computing system 600 may vary. For example, as compared to the depiction in FIG.6, computing system 600 may include a larger or smaller number of processors, I / O devices, or memory units. In addition, computing system 600 may further include other components or devices not depicted that perform or assist in the performance of one or more processes consistent with the disclosed embodiments. TheAttorney Docket No.16048.0009-00304 components and arrangements shown in FIG.6 are not intended to limit the disclosed embodiments, as the components used to implement the disclosed processes and features may vary.

[0111] Processor 610 may comprise known computing processors, including a microprocessor. Processor 610 may constitute a single-core or multiple-core processor that executes parallel processes simultaneously. For example, processor 610 may be a single-core processor configured with virtual processing technologies. In some embodiments, processor 610 may use logical processors to simultaneously execute and control multiple processes. Processor 610 may implement virtual machine technologies, or other known technologies to provide the ability to execute, control, run, manipulate, store, etc., multiple software processes, applications, programs, etc. In another embodiment, processor 610 may include a multiple-core processor arrangement (e.g., dual core, quad core, etc.) configured to provide parallel processing functionalities to allow execution of multiple processes simultaneously. One of ordinary skill in the art would understand that other types of processor arrangements could be implemented that provide for the capabilities disclosed herein. The disclosed embodiments are not limited to any type of processor. Processor 610 may execute various instructions stored in memory 630 to perform various functions of the disclosed embodiments described in greater detail below. Processor 610 may be configured to execute functions written in one or more known programming languages.

[0112] Computer program code for carrying out operations, for example, embodiments may be written in any combination of one or more programming languages, including an object- oriented programming language such as PYTHON, JAVA, SMALLTALK, C++, or the like, and conventional procedural programming languages, such as the “C” programming language or similar programming languages. The program code may execute entirely on the user’s computer, partly on the user’s computer, as a stand-alone software package, partly on theAttorney Docket No.16048.0009-00304 user’s computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer may be connected to the user’s computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection may be made to an external computer (for example, through the Internet using an Internet Service Provider).

[0113] I / O device 620 may include at least one of a display, an LED, a router, a touchscreen, a keyboard, a microphone, a speaker, a haptic device, a camera, a button, a dial, a switch, a knob, a transceiver, an input device, an output device, or another I / O device to perform methods of the disclosed embodiments.

[0114] I / O device 620 may be configured to manage interactions between computing system 600 and other systems using a network. In some aspects, I / O device 620 may be configured to publish data received from other databases or systems not shown. This data may be published in a publication and subscription framework (e.g., using APACHE KAFKA), through a network socket, in response to queries from other systems, or using other known methods. In various aspects, I / O device 620 may be configured to provide data or instructions received from other systems. For example, I / O device 620 may be configured to receive instructions for generating data models (e.g., type of data model, data model parameters, training data indicators, training parameters, or the like) from another system and provide this information to machine learning framework 635. As an additional example, I / O device 620 may be configured to receive data from another system (e.g., in a file, a message in a publication and subscription framework, a network socket, or the like) and provide that data to programs or store that data in, for example, training data set 632, inference data 633, or model 634.

[0115] In some embodiments, I / O device 620 may include a user interface configured to receive user inputs and provide data to a user (e.g., a data manager). For example, I / O deviceAttorney Docket No.16048.0009-00304 620 may include a display, a microphone, a speaker, a keyboard, a mouse, a track pad, a button, a dial, a knob, a printer, a light, an LED, a haptic feedback device, a touchscreen and / or other input or output devices.

[0116] Memory 630 may be a volatile or non-volatile, magnetic, semiconductor, optical, removable, non-removable, or other type of storage device or tangible (i.e., non-transitory) computer-readable medium, consistent with disclosed embodiments. As shown, memory 630 may include inference data 633 and training data set 632, including one of at least one of encrypted data or unencrypted data. Memory 630 may also include models 634, including weights and parameters of neural network models.

[0117] Machine learning framework 635 may include one or more programs (e.g., modules, code, scripts, or functions) used to perform methods consistent with disclosed embodiments (e.g., TENSORFLOW, PYTORCH, KERAS, OPENCV, or the like). Programs may include operating systems (not shown) that perform known operating system functions when executed by one or more processors. Disclosed embodiments may operate and function with computer systems running any type of operating system. Machine learning framework 635 may be written in one or more programming or scripting languages. One or more of such software sections or modules of memory 630 may be integrated into computing system 600, non-transitory computer-readable media, or existing communications software. Machine learning framework 635 may also be implemented or replicated as firmware or circuit logic.

[0118] Machine learning framework 635 may include programs (scripts, functions, algorithms) to assist creation of, train, implement, store, receive, retrieve, and / or transmit one or more machine learning models. Machine learning framework 635 may be configured to assist creation of, train, implement, store, receive, retrieve, and / or transmit, one or more ensemble machine learning models (e.g., machine learning models comprised of a plurality ofAttorney Docket No.16048.0009-00304 machine learning models). In some embodiments, training of a model may terminate when a training criterion is satisfied. Training criteria may include number of epochs, training time, performance metric values (e.g., an estimate of accuracy in reproducing test data), or the like. Machine learning framework 635 may be configured to adjust model parameters and / or hyperparameters during training. For example, machine learning framework 635 may be configured to modify model parameters and / or hyperparameters (i.e., hyperparameter tuning) using an optimization technique during training, consistent with disclosed embodiments. Hyperparameters may include training hyperparameters, which may affect how training of a model occurs, or architectural hyperparameters, which may affect the structure of a model. Optimization techniques used may include grid searches, random searches, gaussian processes, Bayesian processes, Covariance Matrix Adaptation Evolution Strategy techniques (CMA-ES), derivative-based searches, stochastic hill-climbing, neighborhood searches, adaptive random searches, or the like.

[0119] In some embodiments, machine learning framework 635 may be configured to generate models based on instructions received from another component of computing system 600 and / or a computing component outside computing system 600. For example, machine learning framework 635 can be configured to receive a visual (e.g., graphical) depiction of a machine learning model and parse that graphical depiction into instructions for creating and training a corresponding neural network. Machine learning framework 635 can be configured to select model training parameters. This selection can be based on model performance feedback received from another component of machine learning framework 635. Machine learning framework 635 can be configured to provide trained models and descriptive information concerning the trained models to model memory 630.

[0120] Any computer program instructions may also be stored in a computer readable medium that can direct one or more hardware processors of a computer, other programmableAttorney Docket No.16048.0009-00304 data processing apparatus, or other devices to function in a particular manner, such that the instructions stored in the computer readable medium form an article of manufacture including instructions that implement the function / act specified in the flowchart or block diagram block or blocks.

[0121] The computer program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other devices to cause a series of operational steps to be performed on the computer, other programmable apparatus or other devices to produce a computer implemented process such that the instructions that execute on the computer or other programmable apparatus provide processes for implementing the functions / acts specified in the flowchart or block diagram block or blocks.

[0122] The embodiments may further be described using the following clauses:

[0123] 1. A non-transitory, computer-readable medium containing instructions that, when executed by at least one processor of a system, cause the system to perform operations comprising: obtaining a sample corresponding to a compound, the sample including electroencephalographic spectrograms corresponding to different positions in a dosage sequence; generating an indication of a predicted therapeutic effect of the compound when administered to a human for each position in the dosage sequence by applying the sample to a machine learning model configured to: generate separate encodings for the electroencephalographic spectrograms; and generate the indication using the separate encodings; and providing the indication.

[0124] 2. The non-transitory, computer-readable medium of clause 1, wherein: providing the indication comprises providing, for each position in the dosage sequence, an indication that the compound, when administered to a human at a dosage corresponding to the position in theAttorney Docket No.16048.0009-00304 dosage sequence, is an antidepressant, anxiolytic, antipsychotic, cognitive enhancer, hallucinogen, mood stabilizer, or psychostimulant.

[0125] 3. The non-transitory, computer-readable medium of any one of clauses 1 to 2, wherein: the electroencephalographic spectrograms include a first electroencephalographic spectrogram corresponding to a first position in the dosage sequence; the machine-learning model includes a sequence of encoders, the sequence of encoders including, at the first position, a first convolutional neural network encoder; and generating the separate encodings includes applying the first electroencephalographic spectrogram to the first convolutional neural network encoder to generate a first separate encoding.

[0126] 4. The non-transitory, computer-readable medium of any one of clauses 1 to 3, wherein: the machine-learning model includes a neural network classifier; and generating the indication using the separate encodings includes inputting the separate encodings to the neural network classifier.

[0127] 5. A system for classifying compounds, comprising: at least one processor; and at least one non-transitory, computer-readable medium containing instructions that, when executed by the at least one processor, cause the system to perform operations comprising: obtaining a sample corresponding to a compound, the sample including observational datasets corresponding to positions in a dosage sequence, the observational datasets including a first observational dataset corresponding to a first position in the dosage sequence and acquired from a first non-human animal administered a first dosage of the compound; generating an indication of at least one predicted therapeutic effect of the compound when administered to a human by applying the sample to a machine learning model configured to: generate separate encodings corresponding to different positions in the dosage sequence, and generate the indication using the separate encodings, and provide the indication.Attorney Docket No.16048.0009-00304

[0128] 6. The system of clause 5, wherein: the first observational dataset includes or depends upon electroencephalographic data acquired from the first non-human animal during a first trial.

[0129] 7. The system of any one of clauses 5 to 6, wherein: the indication of at least one predicted therapeutic effect includes an indication of a predicted therapeutic effect for each position in the dosage sequence.

[0130] 8. The system of any one of clauses 5 to 7, wherein: providing the indication comprises providing an indication that the compound is an antidepressant, anxiolytic, antipsychotic, cognitive enhancer, hallucinogen, mood stabilizer, or psychostimulant.

[0131] 9. The system of any one of clauses 5 to 8, wherein: the machine-learning model includes a sequence of encoders, the sequence of encoders including, at the first position, a first encoder; and generating the separate encodings includes applying the first observational dataset to the first encoder to generate a first separate encoding.

[0132] 10. The system of clause 9, wherein: the first encoder comprises a convolutional neural network.

[0133] 11. The system of one of clauses 5 to 10, wherein: the machine-learning model includes a classifier; and generating the indication using the separate encodings includes inputting the separate encodings to the classifier.

[0134] 12. The system of clause 11, wherein: the classifier comprises a neural network.

[0135] 13. A method of training a machine learning model to classify compounds, comprising: obtaining a first sample corresponding to a first compound, the first sample including a first observational dataset corresponding to a first position in a first dosage sequence and acquired from a first non-human animal administered a first dosage of the firstAttorney Docket No.16048.0009-00304 compound during a first trial; labeling the first sample with an indication of at least one first predicted therapeutic effect of the first compound when administered to a human; and training the machine-learning model, using a training dataset including the first sample, to generate an indication of at least one predicted therapeutic effect of a compound when administered to a human using a sample including observational datasets corresponding to positions in a dosage sequence, the machine-learning model configured to: generate separate encodings corresponding to different positions in the dosage sequence, and generate the indication using the separate encodings.

[0136] 14. The method of clause 13, wherein: the first sample further includes a second observational dataset acquired from a second non-human animal.

[0137] 15. The method of clause 13, wherein: the first sample further includes: a second observational dataset generated using the first observational dataset and a third observational dataset corresponding to a third dosage of the first compound; and the second observational dataset corresponding to a second dosage of the first compound, the second dosage being greater than the first dosage and less than the third dosage.

[0138] 16. The method of any one of clauses 13 to 15, wherein: the indication of at least one predicted therapeutic effect comprises an indication of a predicted therapeutic effect for each position in the dosage sequence.

[0139] 17. The method of any one of clauses 13 to 16, wherein: the first observational dataset comprises a first electroencephalographic spectrogram.

[0140] 18. The method of any one of clauses 13 to 17, wherein: the machine learning model includes a number of encoders; and the first sample includes duplicate observational datasets, each duplicate corresponding to a different dosage of the first compound, the number of duplicates based in part on the number of encoders.Attorney Docket No.16048.0009-00304

[0141] 19. The method of any one of clauses 13 to 18, wherein: the training dataset includes samples corresponding to different compounds, the samples corresponding to different compounds including observational datasets corresponding to different dosage sequences, the different dosage sequences having equal lengths and including differing dosages.

[0142] 20. The method of any one of clauses 13 to 19, wherein: the machine-learning model includes a sequence of encoders, the sequence of encoders including, at the first position, a first encoder; and training the machine-learning model, using the training dataset including the first sample, comprises applying the first observational dataset to the first encoder to generate a first separate encoding.

[0143] As used herein, unless specifically stated otherwise, a range defined by two endpoint values X and Y includes the endpoints X and Y, except where infeasible.

[0144] As used herein, unless specifically stated otherwise, the term “or” encompasses all possible combinations, except where infeasible. For example, if it is stated that a component may include A or B, then, unless specifically stated otherwise or infeasible, the component may include A, or B, or A and B. As a second example, if it is stated that a component may include A, B, or C, then, unless specifically stated otherwise or infeasible, the component may include A, or B, or C, or A and B, or A and C, or B and C, or A and B and C.

[0145] It is understood that the described embodiments are not mutually exclusive, and elements, components, materials, or steps described in connection with one example embodiment may be combined with, or eliminated from, other embodiments in suitable ways to accomplish desired design objectives.

[0146] In the foregoing specification, embodiments have been described with reference to numerous specific details that can vary from implementation to implementation. Certain adaptations and modifications of the described embodiments can be made. OtherAttorney Docket No.16048.0009-00304 embodiments can be apparent to those skilled in the art from consideration of the specification and practice of the invention disclosed herein. It is intended that the specification and examples be considered as exemplary only. It is also intended that the sequence of steps shown in figures are only for illustrative purposes and are not intended to be limited to any particular sequence of steps. As such, those skilled in the art can appreciate that these steps can be performed in a different order while implementing the same method.

Claims

Attorney Docket No.16048.0009-00304 WHAT IS CLAIMED IS:

1. A non-transitory, computer-readable medium containing instructions that, when executed by at least one processor of a system, cause the system to perform operations comprising: obtaining a sample corresponding to a compound, the sample including electroencephalographic spectrograms corresponding to different positions in a dosage sequence; generating an indication of a predicted therapeutic effect of the compound when administered to a human for each position in the dosage sequence by applying the sample to a machine learning model configured to: generate separate encodings for the electroencephalographic spectrograms; and generate the indication using the separate encodings; and providing the indication.

2. The non-transitory, computer-readable medium of claim 1, wherein: providing the indication comprises providing, for each position in the dosage sequence, an indication that the compound, when administered to a human at a dosage corresponding to the position in the dosage sequence, is an antidepressant, analgesic, anxiolytic, antipsychotic, cognitive enhancer, anticonvulsant, hallucinogen, mood stabilizer, psychostimulant and ADHD medication, anesthetic, or sedatives and hypnotics.

3. The non-transitory, computer-readable medium of claim 1, wherein: the electroencephalographic spectrograms include a first electroencephalographic spectrogram corresponding to a first position in the dosage sequence; the machine-learning model includes a sequence of encoders, the sequence of encoders including, at the first position, a first convolutional neural network encoder; andAttorney Docket No.16048.0009-00304 generating the separate encodings includes applying the first electroencephalographic spectrogram to the first convolutional neural network encoder to generate a first separate encoding.

4. The non-transitory, computer-readable medium of claim 1, wherein: the machine-learning model includes a neural network classifier; and generating the indication using the separate encodings includes inputting the separate encodings to the neural network classifier.

5. A system for classifying compounds, comprising: at least one processor; and at least one non-transitory, computer-readable medium containing instructions that, when executed by the at least one processor, cause the system to perform operations comprising: obtaining a sample corresponding to a compound, the sample including observational datasets corresponding to positions in a dosage sequence, the observational datasets including a first observational dataset corresponding to a first position in the dosage sequence and acquired from a first non-human animal administered a first dosage of the compound; generating an indication of at least one predicted therapeutic effect of the compound when administered to a human by applying the sample to a machine learning model configured to: generate separate encodings corresponding to different positions in the dosage sequence, and generate the indication using the separate encodings; and provide the indication.Attorney Docket No.16048.0009-00304 6. The system of claim 5, wherein: the first observational dataset includes or depends upon electroencephalographic data acquired from the first non-human animal during a first trial.

7. The system of claim 5, wherein: the indication of at least one predicted therapeutic effect includes an indication of a predicted therapeutic effect for each position in the dosage sequence.

8. The system of claim 5, wherein: providing the indication comprises providing an indication that the compound is an antidepressant, anxiolytic, antipsychotic, cognitive enhancer, hallucinogen, mood stabilizer, or psychostimulant.

9. The system of claim 5, wherein: the machine-learning model includes a sequence of encoders, the sequence of encoders including, at the first position, a first encoder; and generating the separate encodings includes applying the first observational dataset to the first encoder to generate a first separate encoding.

10. The system of claim 9, wherein: the first encoder comprises a convolutional neural network.

11. The system of claim 5, wherein:Attorney Docket No.16048.0009-00304 the machine-learning model includes a classifier; and generating the indication using the separate encodings includes inputting the separate encodings to the classifier.

12. The system of claim 11, wherein: the classifier comprises a neural network.

13. A method of training a machine learning model to classify compounds, comprising: obtaining a first sample corresponding to a first compound, the first sample including a first observational dataset corresponding to a first position in a first dosage sequence and acquired from a first non-human animal administered a first dosage of the first compound during a first trial; labeling the first sample with an indication of at least one first predicted therapeutic effect of the first compound when administered to a human; and training the machine-learning model, using a training dataset including the first sample, to generate an indication of at least one predicted therapeutic effect of a compound when administered to a human using a sample including observational datasets corresponding to positions in a dosage sequence, the machine-learning model configured to: generate separate encodings corresponding to different positions in the dosage sequence, and generate the indication using the separate encodings.

14. The method of claim 13, wherein:Attorney Docket No.16048.0009-00304 the first sample further includes a second observational dataset acquired from a second non-human animal.

15. The method of claim 13, wherein: the first sample further includes: a second observational dataset generated using the first observational dataset and a third observational dataset corresponding to a third dosage of the first compound; and the second observational dataset corresponding to a second dosage of the first compound, the second dosage being greater than the first dosage and less than the third dosage.

16. The method of claim 13, wherein: the indication of at least one predicted therapeutic effect comprises an indication of a predicted therapeutic effect for each position in the dosage sequence.

17. The method of claim 13, wherein: the first observational dataset comprises a first electroencephalographic spectrogram .

18. The method of claim 13, wherein: the machine learning model includes a number of encoders; and the first sample includes duplicate observational datasets, each duplicate corresponding to a different dosage of the first compound, the number of duplicates based in part on the number of encoders.Attorney Docket No.16048.0009-00304 19. The method of claim 13, wherein: the training dataset includes samples corresponding to different compounds, the samples corresponding to different compounds including observational datasets corresponding to different dosage sequences, the different dosage sequences having equal lengths and including differing dosages.

20. The method of claim 13, wherein: the machine-learning model includes a sequence of encoders, the sequence of encoders including, at the first position, a first encoder; and training the machine-learning model, using the training dataset including the first sample, comprises applying the first observational dataset to the first encoder to generate a first separate encoding.

Images