Approaches to designing and performing compressed biological experiments and systems for implementing the same

WO2024151979A3PCT designated stage expired Publication Date: 2024-11-14EXVIVO LABS INC
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
PCT/US2024/011445
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2023-01-12
Filing Date
2024-01-12
Publication Date
2024-11-14

AI Technical Summary

Technical Problem

Current drug discovery and precision medicine face limitations due to the limited amount of biological material from patients and the high cost of high-content phenotyping technologies, which restricts the number of drugs that can be screened on diseased tissue samples, hindering the identification of potential drug candidates and treatment options.

Method used

The implementation of a compression methodology that pools members of libraries into groups for screening, ensuring each pool includes at least three members and no two members are in the same subset, allowing for informed pooling operations that account for additive, antagonistic, and synergistic effects, and utilizing a precision screening platform to guide the process autonomously or semi-autonomously.

Benefits of technology

This approach enables the screening of an entire library with fewer pools, reduces data complexity, and allows for more accurate attribution of signals to individual members, thereby increasing the number of drugs that can be screened on a patient's sample, improving drug discovery and precision medicine efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure US2024011445_14112024_PF_FP_ABST
    Figure US2024011445_14112024_PF_FP_ABST
Patent Text Reader

Abstract

Introduced here are approaches to compressing members of libraries into pools for screening. These members could be biological targets to be screened against a drug, or these members could be drugs to be screened against a biological target. Regardless of their nature, the members can be compressed into pools in accordance with an informed pooling operation. As part of the informed pooling operation, the members can be divided among the pools such that (i) each member is in a subset of the pools, (ii) each pool includes at least three members, and (iii) no two members are in an identical subset of the pools.
Need to check novelty before this filing date? Find Prior Art

Description

APPROACHES TO DESIGNING AND PERFORMING COMPRESSED BIOLOGICAL EXPERIMENTS AND SYSTEMS FOR IMPLEMENTING THE SAMEPRIORITY CLAIM

[0001] This application claims priority to US Provisional Application No. 63 / 438,550, filed on January 12, 2023, the disclosure of which is hereby incorporated by reference herein in its entirety.TECHNICAL FIELD

[0002] Various embodiments concern computer programs and associated computer-implemented techniques for designing compressed biological experiments offering high throughput.BACKGROUND

[0003] High-throughput screening (“HTS”) is an approach that is commonly used in drug discovery. At a high level, HTS allows a researcher to quickly connect thousands, hundreds of thousands, or millions of experiments, for example, to determine chemical, genetic, or pharmacological compatibility of different chemical compounds (or simply “compounds”). Through this process, the researcher can quickly recognize the compounds that modulate biomolecular pathways of interest. The results of these experiments provide starting points for drug development.

[0004] Automation is an essential element in the usefulness of HTS. Normally, an integrated robotic system that includes one or more robots is responsible for transporting plates with wells defined therein from station to station for sample and reagent addition, mixing, incubation, and finally detection. With an integrated robotic system, many assays can be prepared and analyzed in a short period of time, allowing compounds that are representative of potential drug candidates to be discovered much more quickly.

[0005] To increase the number of assays, unique distributions of compounds can be employed across the plates. The simplifying assumption made with this approach is that any N compounds in the same well will generally not interact with each other - or the target - in a manner that fundamentally changes the ability of the assay to detect true hits. For example, imagine a plate where Compound A is in Wells 1 , 2, and 3, Compound B is in Wells 2, 3, and 4, and Compound C is in wells 3, 4, and 5. In an assay of this plate against a given target, a hit in Wells 2, 3, and 4 would indicate that Compound B is the most likely agent, while also providing three separate measurements of Compound B’s efficacy against the given target. Large-scale applications of this approach tend to involve combinations in which no two compounds ever share more than one well, in order to reduce the possibility of interference between pairs of compounds that are being screened.

[0006] In order to screen more compounds and hasten drug discovery, researchers have traditionally sought to increase the number of compounds under consideration - requiring more plates, reagents, and other necessary equipment and personnel. There are two notable downsides to this approach, however. First, increasing the number of compounds under consideration can significantly increase the cost of drug discovery. Second, increasing the number of compounds under consideration can significantly increase the amount of data to be analyzed, not only increasing costs but also potentially making it more difficult to evaluate data quality and understand data variability.BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 illustrates a network environment that includes a precision screening platform that is executed by a computing device.

[0008] Figure 2 illustrates an example of a computing device that is able to implement a precision screening platform designed to facilitate an efficient screen.

[0009] Figure 3 includes a high-level illustration of a workflow that can be implemented by a precision screening platform.

[0010] Figure 4 includes a high-level illustration of a process for compressing a library of drugs for screening against a biological target.

[0011] Figure 5 includes a chart that illustrates how the number of unique addresses A can change based on the number of pools p that are associated with each address.

[0012] Figure 6 includes a chart that illustrates how this approach to compression scales non-linearly with the number of samples screened, and therefore, the number of pools that can be created.

[0013] Figure 7 A includes a high-level illustration of a process for compressing a first library of perturbations against a second library of biological targets.

[0014] Figure 7B illustrates how the members of a library can be divided among different cohorts.

[0015] Figure 8A includes a high-level illustration of a process for pooling perturbations in a library based on labels associated with the perturbations.

[0016] Figure 8B illustrates how a pool can be modified based on a label associated with a target included in the pool.

[0017] Figure 9 includes a block diagram of a processing system in which at least some operations described herein can be implemented.

[0018] Various embodiments are shown in the drawings for the purpose of illustration. However, those skilled in the art will recognize that alternative embodiments may be employed without departing from the principles of the present disclosure. Accordingly, while certain embodiments are shown in the drawings, the technologies described herein are amenable to various modifications.DETAILED DESCRIPTION

[0019] A key issue in drug discovery and precision medicine is the limited amount of biological material that can be obtained from a patient’s diseased tissue. Another issue is the cost associated with generating data characterizing the biology of the disease tissue using high-content phenotyping technologies. For example, when conducting phenotypic drug screens on diseased tissue samples, the cost of the experiment is significant and the number of drugs that can be screened is limited by the number of cells isolated from the diseased tissue. For drug discovery, this limits the number of potential hits that can be identified for further characterization and development. For precision medicine, this limits the number of potential treatment options that can be identified and administered to the patient. For example, a core biopsy from a solid tumor can result in a sample comprising 25,000 - 50,000 viable cancer cells. If 30,000 viable cancer cells are isolated for a standard functional precision medicine screen, where each drug is screened across five dose concentrations and each concentration is screened in triplicate wells containing 1 ,000 cells each, then only two drugs can be screened for that patient. There are now hundreds of cancer drugs approved by the Food and Drug Administration (“FDA”), and even a modest increase in the number of these drugs that can be screened on a diseased tissue sample could significantly improve the fields of precision medicine and drug discovery.

[0020] Introduced here are approaches to compressing members of libraries into pools for screening. These approaches are generally called the “compression methodology.” As further discussed below, these members could be biological targets to be screened against a drug, or these members could be drugs to be screened against a biological target. Regardless of their nature, the members can be compressed into pools in accordance with an informed pooling operation. As part of the informed pooling operation, the members can be divided among the pools such that (i) each member is in a subset of the pools, (ii) each pool includes at least three members, and (iii) no two members are in an identicalsubset of the pools. The members in each pool may be referred to as “candidates.” Accordingly, if the library includes a plurality of biological targets, then each pool may include at least three biological targets - representative of different candidates - that can be combined with a drug to generate a sample for screening. Similarly, if the library includes a plurality of drugs, then each pool may include at least three drugs - representative of different candidates - that can be combined with a biological target to generate a sample for screening. Accordingly, the terms “members” and “candidates” can generally be used interchangeably.

[0021] As part of the informed pooling operation, additive effects, antagonistic effects, or synergistic effects of the members can be considered such that pairs of members with known additive effects, antagonistic effects, or synergistic effects are assigned to different pools. Such an approach not only allows an entire library to be screened with fewer pools, but also lessens the likelihood that the resulting data is difficult, if not impossible, to comprehend. By accounting for additive, antagonistic, and synergistic effects, each signal measured for a pool can be more readily attributed to one of the members in the pool as further discussed below.

[0022] This approach can be implemented by a computer program that executes a series of modules in operation. In combination, these modules may allow a user (also called an “operator”) to be fully- or semi-autonomously guided through the process by which the members of a library are examined as part of a screen through separate pooling, measuring, and deconvolving operations. Accordingly, the computer program may be referred to as a “precision screening platform” or simply “screening platform.” Users of the screening platform can include healthcare professionals interested in better understanding whether any drugs in a library are likely to invoke a response from a biological target, as well as researchers interested in better understanding responsiveness of a biological target to different drugs that are representative of candidates for treatment.

[0023] For convenience, embodiments may be described in the context of either compressing libraries of drugs for screening against a biological target or compressing libraries of biological targets for screening against a drug. These embodiments are described solely for the purpose of illustration. If an embodiment is described in the context of compressing a library of drugs for screening against a biological target, its features may be similarly applicable to an embodiment relating to compressing a library of biological targets for screening against a drug unless otherwise noted.

[0024] Moreover, unless otherwise noted, the embodiments set forth below may employ aspects of conventional techniques of molecular biology, cellular biology, synthetic biology, biochemistry, organic chemistry, medicinal chemistry, HTS, virtual screening, bioinformatics, and machine learning. Such conventional techniques include, but are not limited to, automated liquid handling, robotic plate handling, emulsion droplet generation, cell transfection and transduction, fermentation and incubation, cell sorting and analyzing, nucleic acid sequencing, biophysical measuring, and the like. Specific examples of appropriate conventional techniques are provided for reference below. However, other equivalent conventional techniques could also be used.Terminology

[0025] References in the present disclosure to “an embodiment” or “some embodiments” means that the feature, function, structure, or characteristic being described is included in at least one embodiment. Occurrences of such phrases do not necessarily refer to the same embodiment, nor do they necessarily refer to alternative embodiments that are mutually exclusive of one another.

[0026] The term “based on” is to be construed in an inclusive sense rather than an exclusive sense. That is, in the sense of “including but not limited to.” Thus, the term “based on” is intended to mean “based at least in part on” unless otherwise noted.

[0027] The terms “connected,” “coupled,” and variants thereof are intended to include any connection or coupling between two or more elements, either direct or indirect. The connection or coupling can be physical, logical, or a combination thereof. For example, elements may be electrically or communicatively connected to one another despite not sharing a physical connection.

[0028] The term “module” may refer broadly to software, firmware, hardware, or combinations thereof. Modules are typically functional components that generate one or more outputs based on one or more inputs. A computer program may include or utilize one or more modules. For example, a computer program may utilize multiple modules that are each responsible for completing a different task, or a computer program may utilize a single module that is responsible for completing multiple tasks.

[0029] The term “about” means within ± 10% of the recited value.

[0030] The term “biological target” or simply “target” may refer to any biological structure with which another entity - like a drug - can interact, resulting in a detectable response. Examples of targets include molecules (e.g., nucleic acids, lipids, and proteins, such as enzymes and receptors), genes, cells, or tissue samples.

[0031] The term “drug” may refer to any substance that is able to bind to a biological target of interest, and therefore may prompt, invoke, or otherwise produce a biological effect. Examples of drugs include those based on biological molecules and those based on compounds comprising chemical components. For convenience, embodiments may be described in the context of compounds, though the features of those embodiments may be similarly applicable to biological molecules unless otherwise noted.Overview of Screening Platform

[0032] Figure 1 illustrates a network environment 100 that includes a screening platform 102 that is executed by a computing device 104. An individual(also referred to as a “user” or “operator”) can interact with the screening platform 102 via interfaces 106. The user could be a healthcare professional, researcher, or developer (e.g., of drugs), for example. Depending on the nature and interests of the user accessing the interfaces 106, the interfaces 106 may allow for the identification of libraries to be screened, review of data generated during screens, specification of appropriate next steps (e.g., performing a secondary screen, implement a treatment regimen, initiate manufacture of a drug), and management of preferences. Some interfaces may serve as informative dashboards through which individual users can observe, manage, or guide the process by which the screening platform 102 screens libraries, while other interfaces may facilitate interactions between multiple users (e.g., who are members of the same research team or development team).

[0033] As shown in Figure 1 , the screening platform 102 can reside in a network environment 100. Thus, the computing device 104 on which the screening platform 102 resides can be connected to one or more networks 108A- B. Depending on its nature, the computing device 104 could be connected to a personal area network (“PAN”), local area network (“LAN”), wide area network (“WAN”), metropolitan area network (“MAN”), or cellular network. For example, if the computing device 104 is a computer server, then the computing device 104 may be accessible to users via respective computing devices (e.g., mobile phones or laptop computers) that are connected to the Internet via LANs. As another example, if the computing device 104 is included in, or connected to, an analytical instrument that measures signals generated by samples as part of a screen, then the computing device 104 may include a display via which information can be presented to a user. Accordingly, the data to be examined by the screening platform 102 may be obtained from the computing device 104 or (e.g., one or more databases that are accessible to the computing device 104, an analytical instrument that is accessible to the computing device 104, etc.).

[0034] Additionally or alternatively, the computing device 104 may be connected to one or more other computing devices over a short-range wirelessconnectivity technology, such as Bluetooth®, Near Field Communication (“NFC”), Wi-Fi® Direct (also referred to as “Wi-Fi P2P”), and the like. As an example, the screening platform 102 could be embodied as a desktop application that is executed by a laptop computer. In such embodiments, the laptop computer may be communicatively connected - via a wireless communication channel - to a source from which to acquire data. The source could be a database that is accessible to the laptop computer via a network (e.g., the Internet), or the source could be an analytical instrument that is accessible to the laptop computer - or at least whose data is accessible to the laptop computer - via a network (e.g., the Internet). Data could additionally or alternatively be obtained from another computer program executing on the laptop computer and optionally connected to an analytical instrument that is able to generate data as part of a screen.

[0035] The interfaces 106 may be accessible via a web browser, desktop application, mobile application, or another form of computer program. For example, a user may be able to access interfaces through which to initiate and monitor a screen via a desktop application executing on a laptop computer as mentioned above. As another example, a user may be able to access interfaces through which information gleaned from the screen - such as data generated by an analytical instrument or analyses thereof - can be reviewed via a web browser. Accordingly, the interfaces 106 generated by the screening platform 102 may be accessible on various computing devices, including mobile phones, tablet computers, desktop computers, analytical instruments, and the like.

[0036] Generally, the screening platform 102 is executed - at least partially - by a cloud computing service operated by, for example, Amazon Web Services®, Google Cloud Platform™, or Microsoft Azure®. Thus, the computing device 104 may be representative of a computer server that is part of a server system 1 10. Often, the server system 110 comprises multiple computer servers. These computer servers can include different types of data (e.g., information regarding libraries, information regarding individual members of libraries, labels for members, characteristics of members, such as toxicity, preferred dose, knownadditive effects, known antagonistic effects, and known synergistic effects for drugs), algorithms for processing incoming data, and other assets. Those skilled in the art will recognize that these data could also be distributed among the server system 110 and one or more computing devices. As an example, data that is obtained (e.g., acquired or generated) by a user may be stored on, and processed by, her own computing device for security or privacy purposes. This may be useful if, for example, the user is a healthcare professional who is interested in reviewing whether any drugs included in a library interact with a biological target that is representative of, or derived from, a tissue sample taken from a patient, yet information regarding the patient should remain confidential.

[0037] Components of the screening platform 102 could also be hosted locally. That is, part of the screening platform 102 may reside on the computing device used to access one of the interfaces 106. For example, the screening platform 102 may be embodied as a desktop application executing on a laptop computer as mentioned above. Note, however, that the desktop application may be communicatively connected to the server system 110 on which other components of the screening platform 102 are hosted. Additionally or alternatively, the desktop application could be communicatively connected to an analytical instrument on which other components of the screening platform 102 are hosted.

[0038] Figure 2 illustrates an example of a computing device 200 that is able to implement a screening platform 210 designed to facilitate an efficient screen. As shown in Figure 2, the computing device 200 can include a processor 202, memory 204, display mechanism 206, and communication module 208. Each of these components is discussed in greater detail below.

[0039] Those skilled in the art will recognize that different combinations of these components may be present depending on the nature of the computing device 200. For example, if the computing device 200 is a computer server that is part of a server system (e.g., server system 1 10 of Figure 1 ), then the computingdevice 200 may not include the display mechanism 206. Conversely, if the computing device 200 is a laptop computer, then the computing device 200 may include the display mechanism 206.

[0040] The processor 202 can have generic characteristics similar to general- purpose processors, or the processor 202 may be an application-specific integrated circuit (“ASIC”) that provides control functions to the computing device 200. As shown in Figure 2, the processor 202 can be coupled to all components of the computing device 200, either directly or indirectly, for communication purposes.

[0041] The memory 204 can be comprised of any suitable type of storage medium, such as static random-access memory (“SRAM”), dynamic randomaccess memory (“DRAM”), electrically erasable programmable read-only memory (“EEPROM”), quantum memory, flash memory, or registers. In addition to storing instructions that can be executed by the processor 202, the memory 204 can also store data generated by the processor 202 (e.g., when executing the modules of the screening platform 210). Note that the memory 204 is merely an abstract representation of a storage environment. The memory 204 could comprise actual integrated circuits (also called “chips”).

[0042] The display mechanism 206 can be any mechanism that is operable to visually convey information to a user. For example, the display mechanism 206 can be a traditional panel that includes light-emitting diodes (“LEDs”), organic LEDs, liquid crystal elements, or electrophoretic elements. As further discussed below, outputs produced by the screening platform 210 (e.g., through execution of its modules) can be posted to the display mechanism 206 for review by a user of the computing device 200.

[0043] The communication module 208 may be responsible for managing communications external to the computing device 200. Here, for example, the communication module 208 is able to establish separate communication channels with sources 224 from which to obtain data that can be processed,analyzed, or otherwise used by the screening platform 210. For example, the sources 224 could include databases from which to obtain information regarding libraries or members of libraries, analytical instruments from which to obtain information regarding signals detected during screens, etc. Note that for each source, a separate script - stored in the memory 204 as part of the screening platform 210 - may be executed by the processor 202 to allow the communication module 208 to retrieve data therefrom.

[0044] The communication module 208 can be wireless communication circuitry that is able to establish wireless communication channels with other computing devices. Examples of wireless communication circuitry include 2.4 gigahertz (“GHz”) and 5 GHz chipsets compatible with Institute of Electrical and Electronics Engineers (“IEEE”) 802.1 1 - also referred to as “Wi-Fi chipsets.” Alternatively, the communication module 208 may be representative of a chipset configured for Bluetooth, NFC, and the like. Some computing devices - like mobile phones, tablet computers, and the like - are able to wirelessly communicate via separate channels, while other computing devices - like computer servers - tend to wirelessly communicate via a single channel. Accordingly, the communication module 208 may be one of multiple communication modules implemented in the computing device 200, or the communication module 208 may be the only communication module implemented in the computing device 200.

[0045] The nature, number, and type of communication channels established by the computing device 200 - and more specifically, the communication module 208 - can depend on (i) the sources 224 from which data is acquired by the screening platform 210 and (ii) the destinations 226 to which data is transmitted by the screening platform 210. Assume, for example, that the screening platform 210 resides on a computer server. In such embodiments, the communication module 208 can communicate with one or more sources external to the computing device 200 from which to obtain data. These sources could include an analytical instrument and one or more network-accessible databases, forexample. Moreover, the communication module 208 may communicate with at least one destination to which analyses of the data - or the data itself - are transmitted. As an example, this destination could be a database in which information regarding a screen, such as the top hits among the members of a library, is stored. As another example, this destination could be instrumentation that is able to manufacture a drug identified as a top hit by the screening platform 210. As another example, this destination could be another computing device that is associated with a user interested in the analyses performed by the screening platform 210. Those skilled in the art will recognize that a given computing device - say, an analytical instrument, a laptop computer, or a computer server - could be a source and destination.

[0046] For convenience, the screening platform 210 is referred to as a computer program that resides within the memory 204. However, the screening platform 210 could comprise hardware or firmware instead of, or in addition to, software. In accordance with embodiments described herein, the screening platform 210 can include a processing module 212, interactivity module 214, randomization module 216, analysis module 218, attribution module 220, and visualization module 222. These modules could be integral parts of the screening platform 210, or these modules could be logically separate from the screening platform 210 but operate “alongside” it. Together, these modules enable the screening platform 210 to design and complete screens in a compressed manner.

[0047] The processing module 212 can process data that is obtained by the screening platform 210 into a format that is suitable for the other modules. For instance, the processing module 212 can apply operations to data obtained from different sources in preparation for analysis by the other modules of the screening platform 210. As an example, the processing module 212 could filter data that is related to a library to be screened but is not necessary for the screen. As another example, the processing module 212 could concatenate data from different sources that relate to members of the same library, such that the datacan be more readily analyzed despite being obtained from more than one source. Accordingly, the processing module 212 may be responsible for ensuring that the appropriate data is accessible to the other modules of the screening platform 210.

[0048] As further discussed below with reference to Figure 3, the interactivity module 214 and randomization module 216 may be used to perform an informed pooling operation. An important aspect of the compression methodology is how members of a library are compressed into pools for more efficient screening. Consider, for example, a library of drugs to be divided into pools to be screened against a biological target. As part of the informed pooling operation, the screening platform 210 can ensure that (i) each drug is in a subset of the pools, (ii) each pool includes at least three drugs, and (iii) no two drugs are in an identical subset of the pools. To divide the drugs into the pools, the screening platform can implement either the interactivity module 214 or randomization module 216.

[0049] The interactivity module 214 can consider information regarding the drugs in order to determine what rules, if any, should be implemented in dividing the drugs into the pools. For example, additive effects, antagonistic effects, or synergistic effects of the drugs can be considered such that pairs of drugs with known additive effects, antagonistic effects, or synergistic effects are assigned to different pools. Specifically, the interactivity module 214 may maintain - and be able to implement - an algorithm that is programmed or trained to consider additive effects, antagonistic effects, or synergistic effects that are known for combinations of the drugs. In the event that new information regarding additive effects, antagonistic effects, or synergistic effects becomes available, the interactivity module 214 may update the algorithm through reprogramming or retraining. As another example, pools may be designed using drug class as a feature, such that drugs from the same class are separated into different pools. More specific rules could be implemented as well. Additional examples of information that could be considered in establishing the pools are set forth below.

[0050] The randomization module 216, meanwhile, may simply divide the drugs into the pools in a randomized manner. Generally, the randomization module 216 employs an algorithm that randomly selects a predetermined number of drugs from amongst the total number of drugs in the library, in such a manner that the three conditions set forth above are satisfied. For example, for a library that includes five drugs, the randomization module 216 may implement an algorithm that randomly selects three drugs for each pool - a “5 pick 3” operation. Similarly, for a library that includes six drugs, the randomization module 216 may implement an algorithm that randomly selects three drugs for each pool - a “6 pick 3” operation. And for a library that includes seven drugs, the randomization module 216 may implement an algorithm that randomly selects three drugs for each pool - a “7 pick 3” operation. In these examples, each pool includes three drugs. However, pools could include four drugs, five drugs, eight drugs, ten drugs, etc. As further discussed below with reference to Figures 5-6, there is generally a preferred pool size given the total size of the library and number of unique addresses desired.

[0051] In order to perform a screen, each pool defined for the library can be separately combined with a biological target to generate a sample. Assume, for example, that six pools are defined for a library of drugs, such that each pool includes three drugs. Each of the six pools can be combined with a biological target, so as to generate six samples. These samples can be examined by an analytical instrument that monitors for a measurable signal. Generally, the analytical instrument produces, as output, a data structure that includes at least one value for each sample. Each value may be indicative of a characteristic measured for the corresponding sample. The analysis module 218 may be responsible for reviewing the data structure, extracting the values and processing the values in preparation for analysis by the attribution module 220. For example, the analysis module 218 may filter out values that are deemed to be outliers, or the analysis module 218 may filter out values that are determined to relate toanother experiment (e.g., another screen unrelated to the library under consideration).

[0052] As part of a screen, a drug could be screened at different concentrations ranging from a lowest concentration to a highest concentration. In such a scenario, multiple data structures may be obtained by the screening platform 210. These data structures could be generated by a single analytical instrument or multiple analytical instruments. With these multiple data structures, the analysis module 218 can generate a dose-response curve that describes the magnitude of the response of the biological target, as a function of exposure to different concentrations of the drug. Further, the analysis module 218 can calculate a metric that is indicative of the response of the biological target. In the event that the biological target is one of multiple biological targets that are introduced to the pools of drugs, the metric can be used by the screening platform 210 to better understand the relative magnitude of responses. One example of a metric is drug-response score (“DRS”). DRS can be calculated as a continuous function of multiple parameters of the non-linear dose-response curve, including the slope at the half-maximal effective concentration (also called “EC50”) and the top and bottom asymptotes of the dose-response curve. DRS can be calculated as follows:where y is a continuous function of the variable dose x, a is the maximal response (i.e., the top asymptote of the curve), b is the slope of the curve, c is the IC50 (i.e., half-maximal inhibitory concentration), and d is the minimal response (i.e., the bottom asymptote of the curve). Other examples of metrics include those output by sigmoidal and Hill slide functions that, like DRS, incorporate multiple aspects of the dose-response curve into a continuous function.

[0053] The attribution module 220 may be responsible for determining how to attribute measurable signals. Assume, for example, that the data received froman analytical instrument and processed by the analysis module 218 indicates that a measurable signal was detected by the analytical instrument for a sample. As mentioned above, the sample may include at least three drugs, and therefore it is important to know which drug is responsible - or at least primarily responsible - for interacting with the biological target. To accomplish this, the attribution module 220 can employ either a first deconvolution methodology (also called “SUM-based deconvolution”) or a second deconvolution methodology (also called “MAX-based deconvolution”). SUM- and MAX-based deconvolution are further discussed below. After one of the drugs contained in the same is identified as responsible for interacting with the biological target, the attribution module 220 may indicate as much in a data structure. Accordingly, the attribution module 220 may be responsible for identifying and then cataloging the drugs that interact with the biological target, and therefore are considered “hits.”

[0054] Users can benefit from the ability to visualize data that is obtained and generated by the modules of the screening platform 210 to empower informed decision making throughout the screen. The visualization module 222 may be responsible for generating interfaces to which these data - and analyses of these data - can be posted for review. Through these interfaces, users may not only be able to select libraries or individual members to be screened but also review analyses of the screen. For example, the visualization module 222 may generate an interface that identifies the top N hits for a biological target after a screen involving a library of drugs is completed. As another example, the visualization module 222 may generate an interface that includes a dose-response curve, so as to visually illustrate how responsiveness of a biological target relates to concentration of a drug.

[0055] Figure 3 includes a high-level illustration of a workflow that can be implemented by a screening platform 210. As shown in Figure 3, the workflow can involve the screening platform 210 executing a series of operations, namely, a pooling operation, a measuring operation, and a deconvolving operation, in order, to screen a library 302 of drugs against a biological target. To screen alibrary of biological targets against a drug, the screening platform 210 may execute the series of operations in a similar manner.

[0056] Initially, information related to a library 302 of drugs can be obtained by the screening platform 210. For example, a user may specify the library itself or its various drugs through an interface generated by the screening platform 210, and the screening platform 210 may obtain the information in response thereto. As another example, the drugs may be specified in an output produced by another computer system or computer program (e.g., one that employs a machine learnt model to discover potential candidates for development or treatment), and the screening platform 210 may obtain the information in response thereto. The information could be stored locally, for example, in the memory 204 of the computing device 200 that executes the screening platform 210. Alternatively, the information could be stored remotely, in which case the information may be received at the communication module 208.

[0057] As part of the pooling operation, the screening platform 210 can divide the drugs into multiple pools using either an interactivity module 214 or randomization module 216. The different approaches employed by the interactivity module 214 and randomization module 216 to define the multiple pools is discussed in greater detail below. Each pool includes at least a predetermined number of drugs. For example, each pool could include three drugs, four drugs, five drugs, or more than five drugs. Moreover, each drug is assigned to a different subset of the pools. Each subset of the pools uniquely identifies a corresponding drug, and therefore may be called an “address” for the corresponding drug.

[0058] Thereafter, each pool can be separately combined with a biological target to generate samples, and the samples can be loaded into the wells of one or more plates. This could be done manually, or this could be done automatically (e.g., by a robotic system). The samples can then be screened - by an analytical instrument - for a measurable signal. The analytical instrument could rely on anyof a broad range of analytical techniques, including thermal shift assay, surface plasmon resonance, microscale thermophoresis, mass spectrometry, nuclear magnetic resonance spectroscopy, Raman spectroscopy, crystallography, fluorescence polarization, fluorescence, and absorbance. Generally, the analytical instrument produces, as output, a data structure 304 that includes at least one value for each sample. Each value may be indicative of a characteristic measured for the corresponding sample. In some embodiments, each sample is associated with a single value. In other embodiments, each sample is associated with a series of values in temporal order, indicative of the characteristic as measured at different points in time across an interval of time.

[0059] As part of the measuring operation, the screening platform 210 can receive the data structure 304 output by the analytical instrument and identify each value that is indicative of a physical interaction between the biological target and at least one drug included in the corresponding sample. Using the analysis module 218, the screening platform 210 can identify those values - and therefore, those samples - that are worth further analysis.

[0060] As mentioned above, each sample may be representative of a different pool of drugs combined with a biological target, generally accompanied by an aqueous solution. As part of the deconvolving operation, the screening platform 210 - using the attribution module 220 - can attribute each value that is identified as part of the measuring operation to one of the drugs included in the corresponding sample. Consider, for example, a sample that is generated by combining the biological target with three drugs, namely, Drug A, Drug B, and Drug C. With the attribution module 220, the screening platform 210 can attribute the signal to Drug A, Drug B, or Drug C. Generally, this is accomplished by indicating, in a data structure, the drug to which the signal is attributable. The attribution module 220 may employ an algorithm (e.g., a deconvolution algorithm) that determines which drug is responsible for the signal based on an analysis of how the signals compare to the addresses defined for the drugs, as further discussed below.

[0061] Thereafter, the screening platform 210 may produce an output that conveys, summarizes, or otherwise indicates its findings. For example, the output could be a visualization that identifies the drugs, if any, determined to elicit a response from the biological target. The visualization could be presented for review on the display mechanism 206, or the visualization could be transmitted, via the communication module 208, for review via an external display mechanism. As another example, the output could be an instruction to perform further analysis of one or more drugs determined to be candidates for development, or the output could be an instruction to initiate manufacture of one or more drugs determined to be candidates for development. In some embodiments, these instructions are in human-readable form, and therefore are actionable by an individual. In other embodiments, these instructions are in machine-readable form, and therefore are actionable by a computing system (e.g., a robotic system) that automatically performs further analysis, initiates a manufacturing process, etc.

[0062] Accordingly, the screening platform 210 may be able to perform, partially or fully autonomously, a method for screening a library that comprises multiple biological targets against a drug. Initially, the screening platform 210 can divide the multiple biological targets into multiple pools in accordance with an informed pooling procedure, such that (i) each of the multiple biological targets is in a subset of the multiple pools, (ii) each of the multiple pools includes at least three biological targets, and (iii) no two biological targets are in an identical subset of the multiple pools. Each of the multiple pools may be combinable with the drug to generate multiple samples.

[0063] As part of the informed pooling procedure, the screening platform 210 may consider the additive effects, antagonistic effects, or synergistic effects of the multiple biological targets, such that pairs of biological targets with known additive effects, known antagonistic effects, or known synergistic effects are assigned to separate pools. Alternatively, the screening platform 210 may divide the multiple biological targets among the multiple pools in a randomized manner,yet still ensuring that the aforementioned criteria are satisfied. For each subset of the multiple pools, the screening platform 210 may assign a unique address thereto. Said another way, the screening platform 210 may assign, to each subset of the multiple pools, a unique address that serves as an identifier for the corresponding biological target.

[0064] Thereafter, the screening platform 210 can receive data that is representative of signals measured by an analytical instrument for the multiple samples. Based on the data, the screening platform 210 can identify at least one of the multiple samples for which the corresponding signal is indicative of a physical interaction between the drug and a biological target contained therein. The screening platform 210 can indicate, in a data structure, that a given biological target is responsive to the drug in response to a determination that at least a predetermined number or a predetermined percentage of pools (and therefore, samples) that include the given biological target are among the identified samples.

[0065] The screening platform 210 may also be able to perform, partially or fully autonomously, a method for screening a library that comprises multiple drugs against a biological target. Initially, the screening platform 210 can divide the multiple drugs into multiple pools in accordance with an informed pooling procedure, such that (i) each of the multiple drugs is in a subset of the multiple pools, (ii) each of the multiple pools includes at least three drugs, and (iii) no two drugs are in an identical subset of the multiple pools. Each of the multiple pools may be combinable with the biological target to generate multiple samples.

[0066] As part of the informed pooling procedure, the screening platform 210 may consider the additive effects, antagonistic effects, or synergistic effects of the multiple drugs, such that pairs of drugs with known additive effects, known antagonistic effects, or known synergistic effects are assigned to separate pools. The screening platform 210 could also consider the classes of the multiple drugs, such that drugs from the same class are separated into different pools.Alternatively, the screening platform 210 may divide the multiple drugs among the multiple pools in a randomized manner, yet still ensuring that the aforementioned criteria are satisfied. For each subset of the multiple pools, the screening platform 210 may assign a unique address thereto. Said another way, the screening platform 210 may assign, to each subset of the multiple pools, a unique address that serves as an identifier for the corresponding drug.

[0067] Thereafter, the screening platform 210 can receive data that is representative of signals measured by an analytical instrument for the multiple samples. Based on the data, the screening platform 210 can identify at least one of the multiple samples for which the corresponding signal is indicative of a physical interaction between the biological target and a drug contained therein. The screening platform 210 can indicate, in a data structure, that a given drug is responsive to the biological target in response to a determination that at least a predetermined number or a predetermined percentage of pools (and therefore, samples) that include the given drug are among the identified samples.Introduction of Approach to Compressed Screening

[0068] Figure 4 includes a high-level illustration of a process for compressing a library 400 of biological targets for screening against a drug. As mentioned above, the process may be similar if a library of drugs is to be screened against a biological target.

[0069] As shown in Figure 4, each biological target within the library 400 can be assigned an address 402. Each address serves as a representation of a subset of pools 404 to which a corresponding biological target has been assigned. Specifically, Address 1 is associated with Pool 1 , Pool 2, and Pool 5, Address 2 is associated with Pool 3, Pool 4, and Pool 5, Address 3 is associated with Pool 1 , Pool 3, and Pool 6, and Address 4 is associated with Pool 2, Pool 4, and Pool 6. Generally, each address is a string of alphabetic characters and / or numeric characters that uniquely identifies a subset of pools that include a corresponding biological target. Addresses 402 can be designed such that (i) notwo addresses correspond to the same subset of pools 404 and (ii) no two pools 404 contain the same subset of biological targets from the library 400.

[0070] Each pool 404 can then be combined with a drug to generate samples 406 for screening 408. The samples 406 can be screened for a measurable signal using various analytical instruments and analytical techniques, examples of which are provided above. Regardless of how the signals are detected, each signal may be indicative of a physical interaction - that results in a biological effect - between the drug and a biological target included in the corresponding sample. As an example, an analytical instrument may generate at least one measurement of a characteristic for each of the samples 406. The nature of the characteristic will depend on the nature of the analytical instrument. In some embodiments, the analytical instrument generates a single measurement for each of the samples 406. In other embodiments, the analytical instrument generates multiple measurements for each of the samples 406, in which case the multiple measurements may be populated into a data structure in temporal order so as to indicate how the characteristic changes over time.

[0071] Accordingly, as part of the screening 408, a signal can be generated for each of the samples 406. In some embodiments, the signal can be directly analyzed by the screening platform 210. In other embodiments, additional action may need to be taken before insights can be gleaned from the screening 408. For example, the screening 408 may involve a step where each of the samples 406 is separated into two fractions - namely, a positive fraction 410 and a negative fraction 412 - based on an analysis of the corresponding signal. For example, a cell viability screen may use a viability marker such as a DNA dye as a signal. Cells that stain positive for the DNA dye are dead, and cells that stain negative for the DNA dye are alive. The cells can be sorted using fluorescence- activated cell sorting (“FACS”) to separate the viable cells from the dead cells. In some embodiments, the cells can be further analyzed post-sorting using a variety of instruments, such as DNA sequencing or microscopy.

[0072] Accordingly, the screening platform 210 may separate, based on the corresponding signal, each of the samples 406 into a positive fraction and a negative fraction. For each of the samples 406, the screening platform 210 can classify that sample as either a positive sample or a negative sample based on a comparison of the positive and negative fractions. For example, the screening platform 210 may classify a given sample as a positive sample in response to a determination that the positive fraction exceeds the negative fraction. Similarly, the screening platform 210 may classify a given sample as a negative sample in response to a determination that the negative fraction exceeds the positive fraction. For example, in the cell viability screen described above, the FACS instrument can count the number of dead cells versus live cells in each sample during the sorting step. Samples with more dead cells (positive fraction) could be classified as a positive sample, and samples with more viable cells (negative fraction) could be classified as a negative sample. The threshold for classifying samples as positive or negative can be adjusted based on the type of screen and objectives.

[0073] The screening platform 210 can identify “hits” from the library 400 based on the frequency of positive samples relative to negative samples. For each address 402, the screening platform 210 can establish whether the corresponding biological target responds to the drug by determining whether (i) a ratio of positive samples to negative samples across the corresponding pools exceeds a threshold or (ii) a total number of positive samples across the corresponding pools exceeds a threshold. Consider, for example, Address 1 that is associated with Pool 1 , Pool 2, and Pool 5. Address 1 is associated with three pools, and therefore a threshold of two positive samples may be used to classify the corresponding biological target as a hit.

[0074] The presence of multiple hits in the same pool can be resolved through a subsequent deconvolution operation, as further discussed below. The deconvolution operation could involve computational techniques or secondary screens of those pools with multiple hits. Assume, for example, that thescreening platform 210 determines there are two hits. If one of the two hits is associated with an address having three positive samples, it is more likely to be a true hit than another of the two hits that only has two positive samples. Alternatively, if one of the two hits is associated with an address having a single positive sample, it is more likely to be a true hit than another of the two hits that is exclusively associated with addresses having multiple positive samples.

[0075] In some embodiments, the pools 404 of biological targets in the library 400 are combined with different concentrations of a drug to generate a larger set of samples. Accordingly, multiple series of samples 406 may be generated, and each series of samples 406 may be representative of the pools 404 combined with a different one of multiple concentrations of the drug. If the drug is determined to represent a hit, the concentration gradient could be used to construct a dose-response curve based on the signal quantity in each positive sample that includes the drug. The dose-response curve could be used to further deconvolute true hits from false positives in the same samples.

[0076] Each address 402 is associated with an equal number of pools 404 in some embodiments. In other embodiments, the addresses 402 may be associated with different numbers of pools 404. For example, some addresses may be associated with three pools while other addresses may be associated with four pools or five pools. The number of pools p associated with an address should always be greater than one and less than the total number of pools P. Meanwhile, the total number of pools P should be less than the total number of members N in the library to be screened. The minimum number of pools required for the library is based on the total number of unique addresses A that can be created. The total number of unique addresses can be computed as follows:where P is the total number of pools and p is the number of pools associated with each address. For example, if every address comprises a unique subset of three pools out of a total of six pools, then a total of 20 unique addresses can becreated. This means that a library of 20 biological targets can be screened in just 6 samples, with each pool represented in a different sample.

[0077] Figure 5 includes a chart that illustrates how the number of unique addresses A can change based on the number of pools p that are associated with each address. Said another way, the number of unique addresses A can change based on how many pools are assigned to each unique address. Referring again to the example in Figure 4, each address is associated with three pools (p) out of a total of six pools (P). As shown with reference numeral 500, this is the optimal number of pools per address to maximize the number of unique addresses (4), and therefore, throughput of the screen. If the number of pools in each address is greater than three or fewer than three, the number of unique addresses (4) decreases. If a total of seven pools (P) are created, then the optimal number of pools per address (p) is either three or four as shown with reference numeral 502, each resulting in 35 unique addresses.

[0078] Figure 6 includes a chart that illustrates how this approach to compression scales non-linearly with the number of samples screened, and therefore, the number of pools that can be created. At low counts of samples (e.g., fewer than ten total pools), the gains in throughput are modest. However, as the number of samples increases, the number of unique addresses (4) increases at a far greater magnitude, enabling much higher multiplexing and throughput. If a total of 14 pools (P) are created and each address is associated with 7 pools (p), then the total number of unique addresses (4) is 3,432. That is nearly a 100-fold increase in addresses with 14 pools compared to 7 pools, as shown with reference numerals 600 and 602. Thus, doubling the sample count from 7 to 14 results in 50-fold leverage compared to singleplex screens (i.e., one drug or biological target per sample). The leverage continues to scale non- linearly with increased sample count. For example, doubling the sample count from 14 to 28 results in over 5,000-fold leverage with over 40 million unique addresses.

[0079] In some embodiments, the total number of unique addresses {A) that are actually assigned to members of a library is less than 100 percent. Said another way, more unique addresses (A) may be created than there are members in the library. This may be due to practical constraints such as the size of the library that can be created, the number of members (e.g., drugs) that can function within a single sample, interaction effects between members in a pool, and noise from false positives due to a high hit rate. As the fraction of the total number of unique addresses (A) used in a screen decreases, the robustness of the screen and the deconvolution of true hits should improve. Small pilot screens can be run with various parameters (e.g., for P, p, and A) adjusted to measure the signal-to-noise (“SNR”) ratio and determine the minimum number of samples required for a full multiplexed screen.Approaches to Compressing Biological Target Libraries

[0080] In some embodiments, the screen for which the library is compressed is a biochemical screen. In such embodiments, a library of drugs is screened against a biological target - or multiple biological targets - of interest. For example, the biological target could be a disease target, and the drugs could be compounds or biological molecules that might bind to the disease target.

[0081] For compounds, large libraries can be generated by the screening platform 210 using traditional chemistry or combinatorial chemistry and screened using traditional HTS assays or DNA-encoded library (“DEL”) technologies. A compound that interacts with a disease target can be identified by washing away compounds that do not interact with the disease target and then determining the identity of the bound compound. For traditional HTS assays, the bound compound could be identified using mass spectrometry. For DEL screens, the bound compound could be identified by sequencing the deoxyribonucleic acid (“DNA”) labels of the compounds remaining in the sample. The disease target interacting with a compound can then be identified based on the frequency that atarget molecule is present in the samples where the compound was identified relative to the samples where the compound was not identified.

[0082] For biological molecules, large libraries can be generated by the screening platform 210 using display technologies, such as phage display, yeast display, etc. For example, a library of biological molecules could be pooled in accordance with the approaches discussed above and then screened against a mixture of peptides encoded by nucleic acid constructs. Peptides that interact with a biological molecule can be identified by washing away peptides that do not interact with the biological molecule and then sequencing the nucleic acid constructs encoding the remaining peptides in the sample. The biological molecule interacting with a peptide can then be identified based on the frequency that the biological molecule is present in the samples where the peptide was identified relative to the samples where the peptide was not identified.

[0083] In another example, a library of analytes are screened against biosensors that might bind to the analytes and generate a signal. The analyzes could be pooled in accordance with the approaches discussed above and then screened against a mixture of biosensors encoded by nucleic acid constructs. Biosensors that interact with an analyte can be identified by sorting cells or droplets containing a biosensor that generates a signal and then sequencing the nucleic acid construct encoding the biosensor. The analyte interacting with a biosensor can then be identified based on the frequency that the analyte is present in the samples where the biosensor generated a signal relative to the samples where the biosensor did not generate a signal. For biosensors, diverse aptamer or transcription factor libraries can be generated and screened. Because these large chemical or biological libraries are expensive to generate and screen, screening these libraries against multiple targets (e.g., disease targets or analytes) at the same time reduces the cost of screening on a per-target basis. The approaches to compression discussed above also enable a library of targets to be pooled and screened against one or more drugs. Screening large libraries of targets against multiple drugs in parallel also differentiates between selectivebinders versus non-specific binders, offering a further advantage for biochemical screens.

[0084] In other embodiments, the screen for which the library is compressed is a phenotypic screen. For expensive phenotyping technologies such as singlecell sequencing, compressing the library reduces the number of samples that are required to screen the full library. In such embodiments, one or more drugs can be screened against a library of cells. For example, a library of cell lines or cell types of interest can be compressed into pools in order to reduce the number of samples required to screen all of the cells.

[0085] In order to demonstrate the compression methodology, a library of cancer cell lines was screened against several cancer drugs using a cell viability assay. Total cell viability was measured for each cancer cell line using an Adenosine TriPhosphate (“ATP”) assay, and more specifically an ATP-based luminescence assay (i.e., CellTiter Gio). Each cancer drug was screened at multiple concentrations ranging from one nanomolar (nM) to ten micromolar (pM) in order to generate a dose-response curve and calculate a drug response score (“DRS”) for each cancer cell line. The DRS was calculated as a continuous function of multiple parameters of the non-linear dose-response curve, including the slope at the half-maximal effective concentration (also called “EC50”) and the top and bottom asymptotes of the dose-response curve. Cancer cell lines were screened individually in order to measure the actual DRS for each cancer drug, as shown in Table I. Cancer cell lines were also screened in compressed pools, as shown in Table II, in order to determine the performance of the compression approach, as shown in Table III. After total cell viability was measured and DRS was calculated, the DRS of each cancer cell line in the pool was deconvolved for each cancer drug.

[0086] To summarize, the screening platform 210 performed a pooling operation in which cancer cells were divided into a plurality of pools, each of which is separately combinable with a cancer drug to generate a plurality ofsamples, a measuring operation in which signals were measured by an analytical instrument for the plurality of samples, and a deconvolution operation in which each signal was attributed to one of the cancer cells in the corresponding sample. Thereafter, the screening platform 210 computed, for each sample, a metric that indicates how the cancer cells in that sample responded to the cancer drug. Based on these metrics, it is possible to determine whether the cancer drug is a suitable candidate for development or treatment.

[0087] A similar method could be performed by the screening platform 210 to screen a library of drugs against at least one cancer cell that is either derived from a tumor sample taken from a human body or derived from a cell line that is representative of a population of cancer cells established and maintained for a given type of cancer. Initially, the screening platform 210 can perform a pooling operation in which the drugs are divided among a plurality of pools, each of which is separately combinable with the at least one cancer cell to generate a plurality of samples. The drugs can be divided among the plurality of pools such that (i) each drug is in a subset of the plurality of pools, (ii) each pool includes at least three drugs, and (iii) no two drugs are in an identical subset of the plurality of pools. Moreover, the screening platform 210 can perform - or facilitate - a measuring operation in which signals are measured by an analytical instrument for the plurality of samples and a deconvolving operation in which each signal is attributed to a drug in the corresponding sample of the plurality of samples. Thereafter, the screening platform 210 can compute, for each sample, a metric that indicates how the drugs in that sample affected the at least one cancer cell. Based on these metrics, it is possible to determine whether any of the drugs are suitable candidates for treatment of the cancer of which the at least one cancer cell is representative. Thus, the screening platform 210 may identify a given drug from among the drugs as a potential candidate for treatment based on an analysis of these metrics computed for the plurality of samples.Table I includes several examples of cancer drugs and their DRSs across each cancer cell line in the aforementioned experiment. These cancer drugs were selected based on their hit profile in order to demonstrate performance of the compression approach across distinct scenarios.Table II includes a list of cancer cell lines and respective pools used in the aforementioned experiment.Table III includes DRS as measured for the cancer drugs across the compressed sample and target pools.

[0088] A benefit of the compression methodology is the ability to compress members of a library into pools for co-culture screens or compress individual cell cultures into pooled samples for measurement. For example, leukemia cell lines can interact in co-cultures through their immune receptors, which may result in cell activation, cytokine production, and other effects that may impact drug sensitivity and resistance. Indeed, initial deconvolution of pools of cell line cocultures proved to be difficult due to these co-culture effects. Instead of creating compressed target cells in pooled co-cultures, compression was instead used to enable individual cultures of each cell line, with pooled measurements to reduce reagents and instrument time. With individual cultures and sample compression, it was possible to accurately deconvolute DRS for each drug on each cell line, as shown in Table IV.Table IV includes DRS as calculated for the cancer drugs from the deconvolution of the compressed sample and target pools.

[0089] As shown in Table II, the library of five cancer cell lines was compressed into random pools of three cell lines either as co-cultures (“Target Compression” or simply “Target”) or pooled samples (“Sample Compression” or simply “Sample”).

[0090] As shown in Table IV, for each cancer drug screened, an “anchor” cell line is identified across the target pools. DRS for these anchor cell lines are bolded and italicized. Anchors can be identified based on frequency of the cell lines in pools with high DRS versus low DRS. At a high level, an anchor is predicted to be the cell line with the lowest individual DRS. In order to deconvolute individual cell line DRS values from pool DRS values, the anchor can start with an assumed DRS value of zero. Each pool can then be expressedas a polynomial of variables (cell lines in the pool) and constants (DRS value of the pool). The DRS value of each cell line can then be solved for using the anchor value and overlapping polynomial expressions. When multiple values are calculated for the final target solved from the polynomials, then the anchor has a non-zero value that can be determined based on the difference in the calculated value of the final target.

[0091] As shown in Table IV. for both sample compression and target compression, the top two target cell lines with the highest DRS were accurately ranked for each cancer drug screened. For certain cancer drugs - including venetoclax and copanlisib - all cell lines were accurately ranked for sample compression and target compression. When cell lines can be co-cultured for target pooling, the accuracy of the compression approach appears to improve. All cell lines were accurately ranked for all cancer drugs for target compression. This suggests that inter-sample culture effect is mitigated by pooling targets in cocultures.Table V includes DRS as calculated for the compressed target pools with distinct pool designs and compression levels.Table VI includes DRS as calculated from the deconvolution of the compressed target pools.

[0092] As shown in Table V, libraries of cell lines were randomly assigned to three pools out of a total of five (“5 pick 3”), six (“6 pick 3”), and seven (“7 pick 3”) pools. Each pool represented a co-culture of three cell lines (“Target Compression”).

[0093] As shown in Table VI, for libraries of five, six, and seven cell lines compressed into five, six, and seven pools, the top two target cell lines with the highest DRSs were accurately ranked for each cancer drug screened. For certain compressed libraries - like the “7 pick 3” compressed libraries - randomized pool design resulted in overlapping polynomials that could be solved for without assuming an anchor DRS of zero. Through a series of relationships between cell lines across pools, values and ranges of DRSs can be determined without making any assumptions.

[0094] In some embodiments, only the rank of each member of a library is important. This qualitative deconvolution can identify the relative effect size across the library, resulting in a ranked order from highest effect to lowest effect. In other embodiments, the measured value for each target in a pool is important, and quantitative deconvolution is preferred. As shown in Tables V and VI, the compression methodology may be capable of determining both the relative rank between members of a library, as well as the individual values of each target in a pool.

[0095] Referring again to Table VI, the compression methodology performs well with randomized pool design and scales with larger libraries. This scaling feature is important because as libraries include more members - and therefore, more targets across different pools - more resources can be saved. Forexample, in the phenotypic screen discussed above, a typical viability assay will screen each drug with each cell line in triplicate. For the typical 5-dose curve used to measure viability and calculate DRS, that is 15 samples per drug for each cell line. A library of 5 cell lines would therefore require 75 samples to be screened. With “5 pick 3” compressed libraries, only 25 samples were screened to create the 5-dose curve for each cell line. That is a savings of 50 samples. For a library of 7 cell lines, a traditional screen would require 105 samples while a “7 pick 3” compressed library would only require 35 samples. That is a savings of 70 samples. As the library increases from 5 cell lines to 7 cell lines, 20 extra samples can be saved using the approach introduced here. The algorithm developed to perform “7 pick 3” compression also results in more accurate deconvolution based on the specific overlapping pool design, enabling deconvolution without the need for an anchor.Approaches to Compressing Perturbation Libraries

[0096] As discussed above, the compression methodology could be used to compress libraries of biological targets into different pools, for example, for screening against a drug - or multiple drugs - of interest. The compression methodology could similarly be used to compress libraries of perturbations into different pools, for example, for screening against a biological target - or multiple biological targets - of interest. Note that the term “perturbation,” as used herein, could be used to refer to drugs and variations thereof, soluble factors in media, and biological components such as cells, genes, and the like. As a specific example, gene perturbations could be created through the use of RNA interference or CRISPR-mediated regulation.

[0097] A library of perturbations could be screened against a biological target - or multiple biological targets - of interest. For example, as part of a biochemical screen, a library of compounds can be screened against a biological target of interest. As another example, as part of a phenotypic screen, a library of compounds and / or genetic perturbations can be screened against a biologicaltarget (e.g., a cell) or a biological target type (e.g., one or more cells of a given type). For expensive phenotyping technologies such as single-cell sequencing, compressing the perturbations reduces the number of samples that are required to screen the entire library.

[0098] In some embodiments, the perturbations in a library are compressed into pools in a randomized manner. If the effect of a perturbation needs to be screened in isolation, then perturbations can be cultured with targets in separate samples and the samples can be compressed at the end of the screen. This reduces the time and cost of analyzing the library.

[0099] In some embodiments, the targets are diseased cells that have been isolated from a sample of diseased tissue. In such embodiments, a library of perturbations can be compressed into pools and then cultured together in order to reduce the number of diseased cells that are needed to screen all of the perturbations. This not only reduces the cost of the screen, but it also increases the number of perturbations that can be screened given a finite number of diseased cells that can be isolated from the sample of diseased tissue.Moreover, the perturbations could be compressed into designed pools based on information that is related to, and known for, each perturbation. For example, in a precision medicine screen, information may be retrieved, received, or otherwise obtained for various drugs. For each drug, the information may specify toxicity, preferred dose, known additive effects with other drugs, known antagonistic effects with other drugs, known synergistic effects with other drugs, and the like. With this information, the screening platform 210 can minimize the confounding effects between the drugs, for example, by dividing toxic combinations of drugs into different pools.

[0100] To demonstrate compression of perturbation libraries, a library of cancer drugs was screened against a B-cell precursor ALL cell line (called “REH”) and a T-cell ALL cell line (called “MOLT-4”) using the same cell viability assay described above. Total cell viability was measured for each sample usingan ATP-based luminescence assay (i.e., CellTiter Gio). Each cancer drug was then screened at multiple concentrations ranging from one nM to ten pM in order to generate a dose-response curve and calculate a DRS for each cancer cell line.

[0101] Each cancer drug was cultured with target cells individually. At the end of the screen, samples were either analyzed individually or compressed into pools for analysis. The library was compressed alphabetically into ten cohorts, each of which contained seven cancer drugs present in three out of the seven pools, as shown in Table VII. Said another way, the screening platform 210 generated a “7 pick 3” compressed library by assigning the cancer drugs among different pools. Alphabetical pooling was used to demonstrate the performance of randomized pool design. The library was also compressed into ten cohorts with pools designed based on information known about each cancer drug, as shown in Table VIII. After cell viability measurement and DRS calculation, the DRS of each cancer drug in the samples generated based on the pools were deconvoluted and compared to the DRS for each cancer drug measured individually.Table VII includes several examples of cancer drugs and their DRSs as measured for samples generated based on randomly designed pools, where the top two drugs in each cohort are bolded and italicized.

[0102] As shown in Table VII, the compression methodology performs well with randomized (e.g., alphabetical) pooling of samples when each cancer drug is cultured with cancer cells individually. The only exception was Cohort 2, which was an outlier in terms of hit rate and effect size. The compression methodology performed well for the REH cell line but not for the MOLT-4 cell line. As shown in Table VII, the top two drugs were accurately identified and quantified on a per- cohort basis. While the remaining drugs could have been ranked by effectiveness, performance was degraded for the remaining drugs due to the high hit rate in Cohort 2. Identifying and quantifying the top hit or top two hits in a pool is generally sufficient for most, if not all, applications, including precision medicine and drug discovery. Sample pooling for perturbations can reduce the cost of expensive phenotypic drug discovery screens. For example, for a typical 5-dose curve that is used to measure viability and calculate DRS, 15 samples are needed to measure each drug as each drug is screened against a biological target in triplicate. This is 1 ,050 samples in total. With “7 Pick 3” compressed libraries, only 350 samples are needed to measure the entire library.

[0103] In parallel to the exemplary sample compression screen covered in Table VII, perturbations were compressed into pools and then screened. The pools were designed using drug class as a feature with two compression rules (or simply “rules”). The first rule required that drugs from the same class be separated into different cohorts (i.e., mutually exclusive groups of pools). The second rule separated PI3K pathway inhibitors from MAPK pathway inhibitors, as inhibiting these pathways in parallel is known to be synergistic but has not been clinically effective due to dose-limiting toxicities. A benefit of compressing perturbations is the ability to measure additive, antagonistic, and synergistic effects between drugs in the same pool. However, these effects can negatively impact deconvolution of individual drug effects. Therefore, combinations of drugs for which evidence of clinical toxicity exists can be assigned to different pools in accordance with an informed pooling operation. As new preclinical or clinical evidence of toxic combinations becomes available, the rules implemented by thescreening platform 210 - and more specifically, its algorithm for compressing members of a library - can be updated accordingly.Table VIII includes several examples of cancer drugs and their DRSs as measured for samples generated based on informed pool design, where the top two drugs in each cohort are bolded and italicized.

[0104] Pooling perturbations is different than pooling other types of targets. When pooling biological targets or samples containing biological targets, only one perturbation is cultured therewith. Therefore, each biological target in culture is affected by a single perturbation, and when biological targets are pooled during culture (i.e., target pooling) or after culture (i.e., sample pooling), the effect of the perturbation on each biological target remains isolated during measurement.Therefore, the measurement of the pool represents the sum total of the effects of a perturbation on each biological target in the pool, and the effect of each perturbation-target pair in the pool is diluted by the other perturbation-target pairs. In such a compressed screen, a first deconvolution methodology (also called “SUM-based deconvolution”) is generally appropriate.

[0105] In SUM-based deconvolution, an assumption of additivity is used to quantify the individual effects of each perturbation on each target. Additivity implies that each perturbation effect on each target is positively contributing tothe overall effect measured for that pool. This is an accurate assumption for certain experiments, such as when targets or samples are compressed. When targets are pooled - or when samples containing different targets are pooled - the effect of each perturbation on each individual target is diluted. In other words, each perturbation has an isolated effect on a target in culture, and the sum total of the isolated effects are measured when pooled. Therefore, SUM-based deconvolution breaks down the sum total measured for a pool and assigns estimated fractional effects to each compressed perturbation or target in the pool. The screening platform 210 used SUM-based deconvolution for all target and sample compression experiments described above.

[0106] For compressed perturbation experiments, multiple perturbations have an effect on the same target in culture. In such compressed experiments, there is the possibility for interactions between perturbations on the same target. Interactions could be antagonistic (i.e., the effect of a perturbation is reduced by the presence of another perturbation), neural (i.e., the effect of a perturbation masks the effect of at least one other perturbation), additive (i.e., the effect of multiple perturbations positively contribute to the measured pool effect), or synergistic (i.e., the effect of pooled perturbations is larger than the sum of the effect of each pooled perturbation on its own). Perturbation interactions can be difficult to predict, and therefore SUM-based deconvolution was compared to a second deconvolution methodology (also called “MAX-based deconvolution”).

[0107] In MAX-based deconvolution, an assumption of neutrality between the compressed perturbations is made. In other words, the perturbation with the largest effect is assumed to be masking the effects, if any, of the other perturbations in the pool. For SUM- and MAX-based deconvolution, qualitative deconvolution is first performed to rank the perturbations in each pool based on size of expected effect. Qualitative deconvolution, as described above, ranks the perturbations in a pool from largest effect size to smallest effect size. The ranking is based on the overlap of each perturbation’s unique subset of pools with the highest and lowest measured pools. The perturbation with the largest effect isexpected to be present in the pools with the highest measured signal or corresponding metric (e.g., DRS), while the perturbation with the smallest effect is expected to be present in the pools with the lowest measured signal or corresponding metric (e.g., DRS).

[0108] Following qualitative deconvolution, SUM- and MAX-based deconvolution differ in how individual effect size is estimated for each perturbation. Said another way, SUM- and MAX-based deconvolution differ in how quantitative deconvolution is performed.

[0109] As mentioned above, SUM-based deconvolution uses the overlapping pool design to identify relative relationships and ranges for the individual effects of each perturbation. When the overlapping pool design is not sufficient to solve for the polynomials representing perturbation effects, an anchor perturbation can be identified by the screening platform 210 based on rank order and assigned an effect size of zero. The anchor perturbation may be the lowest ranked perturbation in the rank order. This then enables the screening platform 210 to solve for the polynomials for the remaining perturbations.

[0110] For MAX-based deconvolution, the highest ranked perturbation in the rank order is first assigned an effect size based on the pool with the lowest measured signal or corresponding metric (e.g., DRS) that it is present in. Said another way, the screening platform 210 can identify the highest ranked perturbation in the rank order and then assign an effect size based on the measured signal or corresponding metric (e.g., DRS) of the “weakest” pool in which the highest ranked perturbation is included. The screening platform 210 can repeat this step for the next highest ranked perturbation, and so on until the measurements for each pool have been assigned to one of the perturbations. The effects for the remaining perturbations can be assigned a value of zero, since it is assumed that their effect is masked by the perturbation with the largest effect in each pool.

[0111] As shown in Table VIII, the compression methodology in combination with MAX-based deconvolution performs well with informed perturbation pooling. The only scenario for which the top two drugs were not accurately identified and quantified is Cohort 9. Gefitinib was incorrectly ranked behind venetoclax for the MOLT-4 cell line, and the effect size was inaccurately estimated to be 0.00. MAX-based deconvolution was able to correctly identify and accurately quantify the top three drugs for seven cohorts across the REH and MOLT-4 cell lines. Cohorts 4, 9, and 10 had slightly worse performance, only identifying and quantifying the top two hits - except for the gefitinib scenario mentioned above. Additive effects or synergistic effects between perturbations in these cohorts may explain the reduced performance. Therefore, performance of SUM- and MAX- based deconvolution were compared across the ten cohorts, and MAX-based deconvolution performed far better than SUM-based deconvolution except for Cohorts 4, 9, and 10.Table IX includes several examples of cancer drugs and their DRSs as measured for samples generated based on informed pool design, with results for MAX- and SUM based deconvolution and where the top two drugs in each cohort are bolded and italicized.

[0112] Unlike MAX-based deconvolution, SUM-based deconvolution resulted in the correct gefitinib ranking and accurate estimation of effect in Cohort 9. However, the two deconvolution methodologies performed similarly for the remaining drugs across the three example cohorts, as shown above. Importantly, MAX-based deconvolution has one significant benefit over SUM-based deconvolution. By assuming neutrality, MAX-based deconvolution has the potential to identify relationships (e.g., due to additive effects or synergistic effects) between pairs of perturbations in certain scenarios. With SUM-based deconvolution, the sum total effect measured for each pool is assigned toperturbations based on fractional effect size. Therefore, any additive interaction or synergistic interaction will be partially attributed to all perturbations in that pool to different degrees. This negatively impacts quantitative deconvolution and hides potential interactions between perturbations. With MAX-based deconvolution, the total effect size of the pool does not need to be fractionally assigned to perturbations. Yet the unassigned fractional effects could be assumed to be indicative of additive interactions or synergistic interactions (and accounted for by the screening platform 210).

[0113] To identify additive interactions or synergistic interactions, each pair of perturbations within each cohort shown in Table XIII was screened separately (e.g., where two perturbations were screened together instead of pools of three perturbations). An additive pair of perturbations was identified for each of the cohorts shown in Table X. In Cohort 4, bortezomib and belinostat showed an additive interaction when screened together. In Cohort 9, venetoclax and temsirolimus showed an additive interaction. In Cohort 10, Selinexor and everolimus showed an additive interaction. In order to determine if MAX-based deconvolution can identify these additive relationships, the DRSs calculated for each pool were compared to the deconvoluted effect size estimations for the additive pairs of perturbations.Table X includes several cohorts and their DRSs as measured for compressed pools of perturbations exhibiting additive interactions, where top hits for deconvoluted effect size are underlined while additive interactions are bolded and italicized.

[0114] As shown in Table X, the additive pairs of perturbations have larger effect size in pools where both are present. Since the compression methodology ensures no two perturbations are in the exact same subset of pools, MAX-based deconvolution enables the effect size to be estimated for the top hit - underlined in Table X. This allows additive and synergistic interactions to be identified in other pools containing the top hit, based on significantly larger effect size - bolded and italicized in Table X. For Cohorts 4 and 9, the additive pair is in two pools together (out of three pools). This allows the exact pair demonstrating additivity to be identified. For Cohort 4, bortezomib is the top hit for the REH and MOLT cell lines, and the effect size is assigned by the screening platform 210 based on Pool 7. Bortezomib is present with belinostat in Pools 1 and 3, and the additive interaction can be seen based on the larger effect size relative to Pool 7. Belinostat is the only other perturbation in Cohort 4 that is present in both Pools 1 and 3. For Cohort 9, temsirolimus is the top hit for the REH and MOLT cell lines, and the effect size is assigned by the screening platform 210 based on Pool 1 . Temsirolimus is present with venetoclax in Pools 4 and 6, and the additive interaction can be seen based on the larger effect size relative to Pool 1 .Venetoclax is the only other perturbation in Cohort 9 that is present in both Pools 4 and 6. For Cohort 10, Selinexor is the top hit for the REH and MOLT cell lines, and the effect size is assigned based on Pool 5. Selinexor is present with everolimus in Pool 4, and the additive interaction can be seen based on the larger effect size relative to Pool 4. However, since neratinib is also present in Pool 4, the additive effect cannot be confidently assigned to everolimus. Moreover, neratinib appears to have an additive effect with selinexor in Pool 6. Neratinib did not have a significantly additive effect with Selinexor in the paired screening mentioned above. This discrepancy may be based on the thresholdused to define an additive pair during the paired screening. The size of the additive effect also appears to be based on the effect size of each perturbation independently. This is most apparent for temsirolimus in Cohort 9 and everolimus in Cohort 10, which had much larger effects on the REH cell line relative to the MOLT cell line. The additivity with venetoclax and selinexor, respectively, were much larger for the REH cell line. In Cohort 10, neratinib had a larger effect size than everolimus for the MOLT cell line, partially explaining the additivity seen in the pooled screen.

[0115] As shown above, compressed perturbation pools can be screened and accurately deconvoluted using MAX-based deconvolution. An additional benefit of MAX-based deconvolution is the addition of additive interactions or synergistic interactions between perturbations. When an additive pair or synergistic pair is present in multiple pools together, as shown in Cohorts 4 and 9 above, the exact pair can be confidently identified by the screening platform 210. This is highly valuable when screening drugs for precision medicine. Not only does the compression methodology identify the individual effect of one or more top drugs, it also enables beneficial drug interactions to be discovered for designing novel combination regimens.

[0116] For the perturbations identified above, the compression methodology generates sparse, but almost always sufficient, information to deconvolute the individual effects of perturbations in a pool. The overlapping pool design provides quantitative information on possible ranges and relative values for each perturbation screened. This enables deconvolution using either of the aforementioned deconvolution methodologies, and the results of the different deconvolution methodologies can be compared. Further, the compression methodology performs exceptionally well in identifying and ranking the top hits in each library screened, making it useful for narrowing down possible treatment options for precision medicine or identifying bioactive molecules to pursue further for drug discovery.

[0117] The viability assay used to demonstrate appropriate target compression serves as a challenging test case for the compression methodology. The viability assay only produces one measurement (i.e., a luminescence value) related to one feature (i.e., cell viability) of the experimental system. Perturbations can only be deconvolved based on higher or lower viability, as determined based on an analysis of DRSs. The number of perturbations that can be compressed into a pool in such an experimental system is limited by perturbation hit rate, interactions, and bioactivity. For precision medicine drug screens, bioactivity is generally high and perturbation interactions can occur. Informed pooling enables highly active perturbations to be compressed into the same pools. For drug discovery screens with significantly lower hit rates (e.g., less than one percent) and minimal bioactivity, much higher compression and pool sizes can be achieved. As hit rate increases, it becomes more difficult to attribute a measurable signal or corresponding metric (e.g., DRS) to an individual perturbation. For the viability assay, the number of targets (and therefore, samples) that can be compressed is limited by hit rate and signal dilution. The more targets in a pool, the less impact of each individual target on the overall pool signal. These limitations are similar for other low-feature experimental systems, such as biochemical screens with a binary signal (e.g., indicative of binding versus non-binding).

[0118] For targets including perturbations, higher compression levels should be achievable in higher content measurement systems. For example, flow cytometry can measure multiple protein levels within a call, which can represent distinct features associated with distinct perturbation effects. Two perturbations with large effects on distinct features (e.g., proteins) of a biological target could be deconvoluted from the same pool. Microscopy can measure additional features related to the morphology of a biological target, further increasing the number of distinct perturbation effects that can be deconvoluted. Single-cell RNA sequencing of hundreds or thousands of genes has the potential to produce unique expression signatures highly specific to each perturbation, furtherincreasing the compression level that can be achieved. The higher compression level achievable in these high-content measurement systems is especially attractive due to the higher costs of these systems.

[0119] In some embodiments, the compression methodology is used to compress 3, 4, 5, 10, 15, or more than 15 targets into each pool. For example, the screening platform 210 may implement the compression methodology to compress 3, 4, 5, 10, 15, or more than 15 drugs into each pool for a drug screen. In such an embodiment, a high-content experimental system such as flow cytometry, microscopy, or single-cell RNA sequencing could be utilized. In such an embodiment, the compression methodology not only reduces the cost of the drug screen, but also increases the number of drugs that can be screened given a limited number of disease cells from a primary tissue sample (e.g., obtained via a biopsy). For example, utilizing the compression levels demonstrated in the viability assay data above, a compressed drug screen could incorporate 3-fold more drugs than a traditional drug screen given a limited number of cells. Furthermore, the compression levels demonstrated above can reduce the number of single-cell sequencing kits used three fold, which is significant since these kits typically cost about $10,000 each.

[0120] The ability to choose between library compression - whether of perturbations or biological targets - and sample compression is a general benefit of the compression methodology and increases the variety of applications that the compression methodology can be used for. For example, sample-level compression enables direct culture conditions to be screened without increasing the number of samples measured (e.g., single-cell sequenced). This can enable assay optimization with far fewer resources. As demonstrated above for the cell line co-cultures, cells can be cultured separately, then pooled and analyzed together in order to avoid any co-culture effects. If inter-sample variability is more of a concern, then targets can be cultured together as demonstrated above. If antagonism or synergism between perturbations is a concern, then perturbations can be cultured with targets separately between pooling samples for analysis.This minimizes the interaction between perturbation effects on the same biological target. In some embodiments, perturbations can be pooled together in a paired pattern in order to measure effects together and separately, and then calculate additive, antagonistic, or synergistic interactions. In such embodiments, each perturbation is in a unique subset of pools, with a subset of the pools shared with another perturbation. Such an approach to pooling enables a perturbation effect to be measured separately and together with another perturbation across multiple pools. If perturbations are also cultured with targets separately in a parallel screen and then samples are pooled, additive, antagonistic, and synergistic interactions can be confirmed while benefiting from the reduction in samples to be screened.Approaches to Compressing biological target and Perturbation Libraries

[0121] As discussed above, the screening platform 210 could employ the compression methodology to compress biological target libraries and to compress perturbation libraries. The screening platform 210 could also employ the compression methodology to compress biological targets and perturbations, for example, as part of a biological screen. At a high level, the screening platform 210 can perform a method in which members (e.g., biological targets) of a first library are to be screened against members (e.g., perturbations) of a second library. As part of the screen, measurable signals are generated. Each measurable signal relates to the physical interaction between a member (e.g., a biological target) of the first library and a member (e.g., a perturbation) of the second library. These measurable signals can be used to identify pairings (e.g., target-perturbation pairings) between the first and second libraries.

[0122] Figure 7A includes a high-level illustration of a process for compressing a first library 700 of perturbations against a second library 702 of biological targets. While the process of Figure 7 is described in the context of perturbations and biological targets, those skilled in the art will recognize that the process may be similarly applicable to other targets.

[0123] As shown in Figure 7A, a first library 700 of perturbations can be assigned addresses 704. Each address may be representative of a unique subset of the pools 708 to which the perturbations are assigned. Here, for example, Address 1 is associated with Pool 1 , Pool 2, and Pool 5, Address 2 is associated with Pool 3, Pool 4, and Pool 5, Address 3 is associated with Pool 1 , Pool 3, and Pool 6, and Address 4 is associated with Pool 2, Pool 4, and Pool 6. For convenience, the pools 708 created for the first library 700 may be called “perturbation pools.”

[0124] Similarly, a second library 702 of biological targets can be assigned addresses 706. Each address may be representative of a unique subset of the pools 710 to which the biological targets are assigned. Again, Address 1 is associated with Pool 1 , Pool 2, and Pool 5, Address 2 is associated with Pool 3, Pool 4, and Pool 5, Address 3 is associated with Pool 1 , Pool 3, and Pool 6, and Address 4 is associated with Pool 2, Pool 4, and Pool 6. For convenience, the pools 710 created for the second library 702 may be called “biological target pools” or simply “target pools.”

[0125] The perturbation pools 708 can then be placed into samples 712. The number of samples created for each perturbation pool 708 may be equal to, or greater than, the number of target pools 710. The target pools 710 can also be placed into the samples 712, such that (i) each sample includes one of the perturbation pools 708 and one of the target pools 710 and (ii) each of the target pools 710 is in at least one sample with each of the perturbation pools 708.

[0126] In Figure 7A, six pools are created for the first library 700, and each of the addresses 704 is associated with three pools. Similarly, six pools are created for the second library 702, and each of the addresses 706 is associated with three pools. This results in 36 samples, each of which contains a unique pair of pools - each containing a unique pair of pools from the first and second libraries 700, 702, with each perturbation from the first library 700 combined with each biological target from the second library 702 in 9 samples. These 36 samples canbe screened, for example, in a 384-well plate 714 using 36 wells 716. This setup enables a first library of 20 different perturbations to be screened against a second library of 20 different biological targets in 36 samples.

[0127] This approach can scale non-linearly relative to the number of samples required to conduct an experiment. If the first library 700 were to be pooled in accordance with the aforementioned procedure and then separately screened against each biological target in the second library 702, it would require 20 samples for each perturbation pool 708 - equaling 120 samples in total. This is over a 3-fold gain in throughput compared to multiplexing in just one of the libraries. If each perturbation from the first library 700 and each biological target from the second library 702 were to be screened individually, it would require 400 total samples. This would require more than one 384-well plate 714.

[0128] As illustrated in Figure 7A, the size of each library can scale non- linearly relative to the number of samples required to conduct an experiment. A library of 462 perturbations can be combined into 11 pools if each perturbation is present in 5 of the 1 1 pools, and a library of 70 biological targets can be combined into 8 pools if each biological target is present in 4 out of the 8 pools. These two libraries can be screened against each other in just 88 total wells 718. If the library of perturbations was instead pooled and then individually screened against each biological target, it would require 70 samples for each perturbation pool - equaling 770 samples in total. Multiplexing both libraries represents nearly a 10-fold gain in throughput per sample relative to multiplexing a single library. If each perturbation and each biological target were to be screened individually, it would require 32,340 total samples.

[0129] In some embodiments, the first library 700 and / or the second library 702 are divided into distinct cohorts. For example, solubility can limit how many compounds can be combined into one pool. Additionally, a high hit rate can make it challenging to deconvolute true hits across many samples that generate signals indicative of a hit. For these and other reasons, a library can be divided intomutually exclusive cohorts, each of which is representative of a distinct subset of the overall library. Each cohort can be combined into a distinct set of pools, so that each perturbation or biological target is only present in pools comprising perturbations or biological targets from the same cohort.

[0130] Figure 7B illustrates how the members of a library can be divided among different cohorts. Specifically, Figure 7B shows how a library 750 of perturbations can be divided into multiple cohorts 752 and then screened against a library 756 of biological targets that are also divided into multiple cohorts 754. The cohorts established for each library can be screened against the cohorts established for the other library across multiple plates. For example, 460 perturbations can be divided into 4 cohorts, each of which includes 115 perturbations, while 60 biological targets can be divided into 4 cohorts, each of which includes 15 biological targets. Within each perturbation cohort, perturbations can be assigned addresses that comprise a unique subset of 5 pools out of 1 1 pools. Within each target cohort, biological targets can be assigned addresses that comprise a unique subset of 4 pools out of 8 pools.

[0131] As shown in Figure 7B, each perturbation cohort can then be combined into samples with each target cohort in accordance with the process described above. This results in 16 sets of samples, where each set of samples includes perturbations from one cohort and biological targets from one cohort. Sample sets can be spread across multiple plates if necessary. Consider, for example, Sample Set 1 .3 referenced with numeral 758. Within Sample Set 1 .3, perturbations from Cohort 1 and biological targets from Cohort 3 are combined and screened. Each of the 1 1 perturbation pools from Cohort 1 can be placed into one of 1 1 columns (e.g., labeled as Column A - Column K) on a plate, with each perturbation pool replicated across 8 wells within the column. Each of the 8 target pools from Cohort 3 can be placed into one of 8 rows (e.g., labeled as Row 1 - Row 8), with each target pool replicated across 11 wells within the row. The result is 88 samples, each of which is loaded into a different well and comprises a unique pair of pools - one perturbation pool from Cohort 1 and one target poolfrom Cohort 3. This process can be repeated for all cohorts across all four plates, enabling the full perturbation library to be screened against the full target library using four 384-well plates. Note that the number of plates needed will depend on the size of the perturbation and target libraries. Each plate may also include positive and negative control wells 760.Approaches to Pooling Libraries Based on Labels

[0132] In addition to the compression methodology discussed above, introduced here is an approach to pooling members of a library based on one or more labels that are associated with each member. The approach may be similarly applicable whether the members are biological targets or perturbations. The approach described below is intended to be applicable to a broad range of labels. For example, for perturbations, labels could specify identity (e.g., source, concentration, etc.), properties (e.g., molecular weight, solubility, permeability, hydrogen bond donors and acceptors, functional groups, scaffold, structure, etc.), activity (e.g., docking score, assay binding or inhibition, affinity, etc.), history (e.g., past assay parameters, pools, etc.), relationships (e.g., clusters, motifs, families, etc.), and composite scores (e.g., calculated by predictive machine learnt models or another algorithm based on features).

[0133] Figure 8A includes a high-level illustration of a process for pooling perturbations in a library 800 based on labels associated with the perturbations. Initially, the library 800 can be divided into distinct cohorts 802, 804 based on an analysis of the labels associated with the perturbations. Each cohort can comprise a distinct subset of the library 800 and a distinct set of pools 808, 812. Perturbations may only be placed into pools with other perturbations from the same cohort. Within each cohort, addresses 806, 810 can be created. Like the pools 808, 812, distinct sets of addresses 806, 810 could be used across the different cohorts. Within each cohort, each perturbation can be assigned one of the addresses created for that cohort. Then, each perturbation can be placed into the pools that are associated with the corresponding address, such that no twoperturbations are in the same subset of pools and no two pools comprise the same perturbations. Referring to Figure 8A, for example, Address 1 , Address 2, Address 3, and Address 4 are created for the first cohort 802. Address 1 is associated with Pool 1 , Pool 2, and Pool 5, Address 2 is associated with Pool 3, Pool 4, and Pool 5, Address 3 is associated with Pool 1 , Pool 3, and Pool 6, and Address 4 is associated with Pool 2, Pool 4, and Pool 6. Address 5, Address 6, Address 7, and Address 8 are created for the second cohort 804. Address 5 is associated with Pool 7, Pool 8, and Pool 11 , Address 6 is associated with Pool 9, Pool 10, and Pool 1 1 , Address 7 is associated with Pool 7, Pool 9, and Pool 12, and Address 8 is associated with Pool 8, Pool 10, and Pool 12.

[0134] In some embodiments, a library of perturbations is divided into distinct cohorts based on labels related to assay parameters. For example, perturbations with a low solubility label may be placed into a cohort comprising pools with higher dimethylsulfoxide (“DMSO”) concentration to increase solubility. As another example, perturbations with a low permeability label may be placed into a cohort comprising pools with a cell-free molecule (e.g., cell lysate or purified protein) to increase permeability. Similarly, a library of perturbations could be divided into distinct cohorts based on labels related to past assays, For example, perturbations from the same high DMSO cohort may be plated into distinct cohorts comprising pools with different DMSO concentrations in order to test an array of assay parameters across the library.

[0135] In some embodiments, a library of perturbations is divided into cohorts based on labels related to molecular scaffold. For example, perturbations with the same scaffold label may be divided into distinct cohorts to improve the resolution of hits and identify which functional groups on a certain scaffold result in a hit. Additionally or alternatively, a library of perturbations could be divided into cohorts based on labels related to predicted activity. For example, perturbations may be virtually screened using docking simulation software, and virtual hits can then be clustered based on similarity (e.g., scaffold). Clustered perturbations can then be divided into distinct cohorts in order to identify whichvirtual hits result in a true hit. As another example, a library of perturbations could be divided into cohorts based on labels related to predicted binding affinity or docking score. Perturbations within the same range of binding affinity or scoring score may be divided into distinct cohorts in order to identify hits within each predicted range.

[0136] Each perturbation can be present in a library at multiple concentrations, and in such a scenario, the perturbations can be divided into cohorts based on labels that are indicative of concentration. For example, each perturbation can be divided into distinct cohorts at different concentrations in order to identify a dose-response curve. The signal of the perturbation across the pools in a given cohort represents one point on the dose-response curve. Meanwhile, the signals of the perturbations across the pools in all cohorts can form the dose-response curve. Dividing the perturbations between distinct cohorts ensures that the same perturbation does not end up in the same pool at different concentrations.

[0137] In some embodiments, a library of perturbations is divided into cohorts based on an analysis of labels related to multiple characteristics. These labels may be referred to as “orthogonal labels.” For example, a first subset of perturbations in the library could be divided into cohorts based on labels related to molecular concentration, and a second subset of perturbations in the library could be divided into the same cohorts based on labels related to scaffold. Additionally or alternatively, perturbations could be divided into cohorts based on hierarchical labels. For example, a library may initially be split into two subsets of perturbations based on labels related to docking score. Each subset could then be divided into cohorts based on labels related to solubility. Accordingly, members of a library could be divided into cohorts by considering labels related to different characteristics, either simultaneously or sequentially.

[0138] In addition to establishing appropriate cohorts, labels could also be used for modifying pools of targets. An approach to modifying pools based on ananalysis of labels associated with the targets in those pools is also introduced here. The approach described below is intended to be applicable to a broad range of labels. For example, for perturbations, labels could specify identity (e.g., source, concentration, etc.), properties (e.g., molecular weight, solubility, permeability, hydrogen bond donors and acceptors, functional groups, scaffold, structure, etc.), activity (e.g., docking score, assay binding or inhibition, affinity, etc.), history (e.g., past assay parameters, pools, etc.), relationships (e.g., clusters, motifs, families, etc.), and composite scores (e.g., calculated by predictive machine learnt models or another algorithm based on features).

[0139] Figure 8B illustrates how a pool can be modified based on a label associated with a target included in the pool. Specifically, Figure 8B shows a scenario where a first address 854 that is initially assigned to a first target 850 is switched with a second address 856 that is initially assigned to a second target 852. Here, the first and second targets 850, 852 are biological targets, though those skilled in the art will recognize that the approach may be similar if the targets are instead perturbations. Following the switching operation, the first address 854 is associated with the second target 852, while the second address 856 is associated with the first target 850. Referring to Figure 8B, Address 1 is still associated with Pool 1 , Pool 2, and Pool 5 but is now associated with the second target 852 rather than the first target 850. Meanwhile, Address 2 is still associated with Pool 3, Pool 4, and Pool 5 but is now associated with the first target 850 rather than the second target 852. The targets that are switched as part of the switching operation can be determined based on an analysis of labels, allowing the pools to be modified in a focused manner while preserving the combinatorial pooling for the rest of the members of the library. This enables an existing library to be modified and screened more efficiently compared to reassigning all addresses and recombining all targets into new pools.

[0140] In some embodiments, addresses are switched based on labels related to past screens. For example, a first perturbation that is assigned a first address and a second perturbation that is assigned a second address may bepresent in the same pool that generates a positive signal in a previous screen. In order to deconvolute the signal and identify the true hit, the second perturbation can switch its address with a third perturbation that is assigned a third address not present in the same pools. This allows a secondary screen to be run with the first and second perturbations in different pools. As another example, a first perturbation and a second perturbation may share a first address that is associated with pools that generated positive signals in a previous screen. The second perturbation can switch its address with a third perturbation that is not present in the same pools. This results in the first and third perturbations sharing the first address, while the second perturbation is associated with the second address. This allows multiplexing groups of perturbations within an address, and subsequent deconvolution of hits through address switching and secondary screening.

[0141] In some embodiments, addresses are switched in order to refactor a library. For example a first perturbation can be removed from the library and a second perturbation can be added to the library to replace the first perturbation in future screens. The second perturbation can be assigned a first address that was previously assigned to the first perturbation. Moreover, the second perturbation can switch the first address with a second address assigned to a third perturbation, for example, based on one or more labels in order to optimize the pooling of the refactored library. Labels can also be updated over time, either for all members of a library or for a subset of the members. Consider, for example, a library of perturbations. Labels associated with a subset of the perturbations may be updated based on the results of a virtual screen or the inference of a machine learnt model. The updated labels may require that the impacted perturbations switch addresses with other perturbations in order to optimize the pooling of the refactored library.

[0142] In some embodiments, addresses are switched across distinct cohorts. For example, a first perturbation that is assigned a first address in a first cohort may be associated with one or more labels. In the event that these labels areupdated, the screening platform 210 may determine that the first perturbation is no longer compatible with the first cohort. If the screening platform 210 determines that a second perturbation that is assigned a second address in a second cohort has a label that is compatible with the first cohort (and presumably is also compatible with the second cohort), a switch may be carried out. The second perturbation can be assigned the first address in the first cohort, while the first perturbation can be assigned the second address in the second cohort.Example of Processing System

[0143] Figure 9 includes a block diagram of a processing system 900 in which at least some operations described herein can be implemented. For example, components of the processing system 900 may be hosted on a computing device that includes a screening platform (e.g., screening platform 210 of Figure 2).

[0144] The processing system 900 can include a processor 902, main memory 906, non-volatile memory 910, network adapter 912, video display 918, input / output devices 920, control device 922 (e.g., a keyboard or pointing device such as a computer mouse or trackpad), drive unit 924 including a storage medium 926, and signal generation device 930 that are communicatively connected to a bus 916. The bus 916 is illustrated as an abstraction that represents one or more physical buses or point-to-point connections that are connected by appropriate bridges, adapters, or controllers. The bus 916, therefore, can include a system bus, a Peripheral Component Interconnect (“PCI”) bus or PCI-Express bus, a HyperTransport (“HT”) bus, an Industry Standard Architecture (“ISA”) bus, a Small Computer System Interface (“SCSI”) bus, a Universal Serial Bus (“USB”) data interface, an Inter- Integrated Circuit (“l2C”) bus, or a high-performance serial bus developed in accordance with IEEE 1394.

[0145] While the main memory 906, non-volatile memory 910, and storage medium 926 are shown to be a single medium, the terms “machine-readable medium” and “storage medium” should be taken to include a single medium ormultiple media (e.g., a centralized / distributed database and / or associated caches and servers) that store one or more sets of instructions 928. The terms “machine-readable medium” and “storage medium” shall also be taken to include any medium that is capable of storing, encoding, or carrying a set of instructions for execution by the processing system 900.

[0146] In general, the routines executed to implement the embodiments of the disclosure can be implemented as part of an operating system or a specific application, component, program, object, module, or sequence of instructions (collectively referred to as “computer programs”). The computer programs typically comprise one or more instructions (e.g., instructions 904, 908, 928) set at various times in various memory and storage devices in a computing device. When read and executed by the processor 902, the instruction(s) cause the processing system 900 to perform operations to execute elements involving the various aspects of the present disclosure.

[0147] Further examples of machine- and computer-readable media include recordable-type media, such as volatile memory devices and non-volatile memory devices 910, removable disks, hard disk drives, and optical disks (e.g., Compact Disk Read-Only Memory (“CD-ROMs”) and Digital Versatile Disks (“DVDs”)), and transmission-type media, such as digital and analog communication links.

[0148] The network adapter 912 enables the processing system 900 to mediate data in a network 914 with an entity that is external to the processing system 900 through any communication protocol supported by the processing system 900 and the external entity. The network adapter 912 can include a network adaptor card, a wireless network interface card, a router, an access point, a wireless router, a switch, a multilayer switch, a protocol converter, a gateway, a bridge, bridge router, a hub, a digital media receiver, a repeater, or any combination thereof.Remarks

[0149] The foregoing description of various embodiments of the claimed subject matter has been provided for the purposes of illustration and description. It is not intended to be exhaustive or to limit the claimed subject matter to the precise forms disclosed. Many modifications and variations will be apparent to one skilled in the art. Embodiments were chosen and described in order to best describe the principles of the invention and its practical applications, thereby enabling those skilled in the relevant art to understand the claimed subject matter, the various embodiments, and the various modifications that are suited to the particular uses contemplated.

[0150] Although the Detailed Description describes certain embodiments and the best mode contemplated, the technology can be practiced in many ways no matter how detailed the Detailed Description appears. Embodiments can vary considerably in their implementation details, while still being encompassed by the specification. Particular terminology used when describing certain features or aspects of various embodiments should not be taken to imply that the terminology is being redefined herein to be restricted to any specific characteristics, features, or aspects of the technology with which that terminology is associated. In general, the terms used in the following claims should not be construed to limit the technology to the specific embodiments disclosed in the specification, unless those terms are explicitly defined herein. Accordingly, the actual scope of the technology encompasses not only the disclosed embodiments, but also all equivalent ways of practicing or implementing the embodiments.

[0151] The language used in the specification has been principally selected for readability and instructional purposes. It may not have been selected to delineate or circumscribe the subject matter. It is therefore intended that the scope of the technology be limited not by this Detailed Description, but rather by any claims that issue on an application based hereon. Accordingly, the disclosure of various embodiments is intended to be illustrative, but not limiting, of the scope of the technology as set forth in the following claims.

Claims

CLAIMSWhat is claimed is:1 . A method for screening multiple biological targets against a drug, the method comprising: dividing the multiple biological targets into multiple pools in accordance with an informed pooling procedure, such that(i) each of the multiple biological targets is in a subset of the multiple pools,(ii) each of the multiple pools includes at least three biological targets , and(iii) no two biological targets are in an identical subset of the multiple pools, wherein each of the multiple pools is combinable with the drug to generate multiple samples; receiving data that is representative of signals measured by an analytical instrument for the multiple samples; and identifying, based on an analysis of the data, at least one of the multiple samples for which the corresponding signal is indicative of a physical interaction between the drug and a biological target contained therein.

2. The method of claim 1 , wherein as part of the informed pooling procedure, additive effects, antagonistic effects, and / or synergistic effects of the multiple biological targets are considered, such that pairs of biological targets with known additive effects, known antagonistic effects, and / or known synergistic effects are assigned to separate pools.

3. The method of claim 1 , further comprising: indicating, in a data structure, that a given biological target is responsive to the drug in response to a determination that at least apredetermined number or a predetermined percentage of pools that include the given biological target are among the at least one sample.

4. The method of claim 1 , wherein each of the multiple biological targets is representative of a different biological cell, biological molecule, or gene.

5. The method of claim 1 , further comprising: assigning, to each subset of the multiple pools, a unique address that serves as an identifier for the corresponding biological target.

6. The method of claim 5, wherein a total number of unique addresses that can be assigned ispl py, where P equals a total number of the multiple pools and p equals a number of the pools included in each subset.

7. The method of claim 1 , wherein the analytical instrument relies on thermal shift assay, surface plasmon resonance, microscale thermophoresis, mass spectrometry, nuclear magnetic resonance spectroscopy, Raman spectroscopy, crystallography, fluorescence polarization, fluorescence, or absorbance.

8. The method of claim 1 , further comprising: for each of the at least one sample, deconvolving the signal so as to establish the biological target involved in the physical interaction.

9. A method for screening multiple drugs against a biological target, the method comprising: dividing the multiple drugs into multiple pools in accordance with an informed pooling procedure, such that(i) each of the multiple drugs is in a subset of the multiple pools,(ii) each of the multiple pools includes at least three drugs, and(iii) no two drugs are in an identical subset of the multiple pools, wherein each of the multiple pools is combinable with the biological target to generate multiple samples; receiving data that is representative of signals measured by an analytical instrument for the multiple pools; and identifying, based on an analysis of the data, at least one of the multiple samples for which the signal is indicative of a physical interaction between the biological target and a drug contained therein.

10. The method of claim 9, wherein as part of the informed pooling procedure, additive effects, antagonistic effects, and / or synergistic effects of the multiple drugs are considered, such that pairs of drugs with known additive effects, known antagonistic effects, and / or known synergistic effects are assigned to separate pools.11 . The method of claim 9, wherein as part of the informed pooling procedure, a class of the multiple drugs is considered, such that drugs from the same class are separated into different pools.

12. The method of claim 9, wherein each drug is representative of a different chemical compound or biological molecule that may be able to bind to the biological target.

13. The method of claim 9, further comprising: assigning, to each subset of the multiple pools, a unique address that serves as an identifier for the corresponding drug.

14. The method of claim 9, further comprising:separating, based on the corresponding signal, each of the multiple samples into a positive fraction and a negative fraction; and classifying each of the multiple samples as either a positive sample or a negative sample based on a comparison of the positive fraction to the negative fraction.

15. The method of claim 14, further comprising: for each of the multiple drugs, establishing whether that drug responds to the biological target by determining whether (i) a ratio of positive samples to negative samples in the corresponding subset of the multiple pools exceeds a threshold or (ii) a total number of positive samples in the corresponding subset of the multiple pools exceeds a threshold.

16. The method of claim 9, wherein a total number of unique addresses that can be assigned is!(1 where P equals a total number of the multiple pools and p equals a number of the pools included in each subset.

17. A non-transitory medium with instructions stored thereon that, when executed by a processor, cause the processor to perform operations comprising: performing a pooling operation in which a plurality of drugs is divided into a plurality of pools such that(i) each drug is in a subset of the plurality of pools,(ii) each pool includes at least three drugs, and(iii) no two drugs are in an identical subset of the plurality of pools, wherein each pool is separately combinable with at least one cancer cell to generate a plurality of samples,a measuring operation in which signals are measured by an analytics instrument for the plurality of samples, and a deconvolving operation in which each signal is attributed to a drug in the corresponding sample of the plurality of samples; computing a metric for each sample of the plurality of samples based on the signals, so as to produce a plurality of metrics, wherein each metric of the plurality of metrics indicates how the at least one cancer cell responded to the at least three drugs in the corresponding sample of the plurality of samples; and identifying a given drug from among the plurality of drugs as a potential candidate for treatment based on an analysis of the plurality of metrics.

18. The non-transitory medium of claim 17, wherein as part of the pooling operation, the processor applies an algorithm that is programmed or trained to consider additive effects, antagonistic effects, and / or synergistic effects that are known for combinations of the plurality of drugs.

19. The non-transitory medium of claim 18, wherein the algorithm is updated, through reprogramming or retraining, as new information regarding the additive effects, antagonistic effects, and / or the synergistic effects becomes available.

20. The non-transitory medium of claim 17, wherein each signal is representative of a temporal sequence of discrete measurements of total cell viability.21 . The non-transitory medium of claim 20, wherein total cell viability is measured for each cancer cell using an Adenosine TriPhosphate (ATP) assay.

22. The non-transitory medium of claim 17, wherein the plurality of samples is one of multiple pluralities of samples for which the measuring operation and the deconvolving operation are performed, and wherein to generate the multiple pluralities of samples, different concentrations of the plurality of drugs are combined with the at least one cancer cell.

23. The non-transitory medium of claim 17, wherein the at least one cancer cell combined with each pool is derived from a tumor sample taken from a human body.

24. The non-transitory medium of claim 17, wherein the at least one cancer cell combined with each pool is derived from a cancer cell line that is representative of a population of cancer cells established and maintained for a given type of cancer.

Citation Information

Patent Citations

  • Machine learning systems for automated pharmaceutical molecule screening and scoring

    US11127488B1

  • Screening method for multi-target drugs and / or drug combinations

    US20180080913A1

  • Multiplexed proteomics and predictive drug candidate assessment

    US20200225242A1

  • Generating Anti-infective design spaces for selecting drug candidates

    US20220165360A1

  • Multiple target screening of molecular libraries by mass spectrometry

    WO1999005309A1