CATALYST ISO Final.pdf

Health Science Futures

Computational ADME Tox AnaLYsis for Safer Therapeutics (CATALYST) Advanced Research Projects Agency for Health

ARPA-H-SOL-24-114

November 1, 2024

Table of Contents

1. Innovative Solutions Opening Solicitation Overview Information

2. Description of the Solicitation

3. Award Strategy

4. Eligibility Information

5. Submission Information

6. Solution Summary and Proposal Review Information

7. Award Administration Information

8. Other Information Appendix A: Solution Summary Template

1. Innovative Solutions Opening Solicitation Overview Information

Federal Agency Name: Advanced Research Projects Agency for Health (ARPA-H) Solicitation Title: Computational ADME Tox AnaLYsis for Safer Therapeutics

(CATALYST)

Announcement Type Innovative Solutions Opening Solicitation Number: ARPA-H-SOL-24-114 ISO Contact: Email: CATALYST@arpa-h.gov

ATTN: ARPA-H-SOL-24-114

Solution Summary Submission Site: https://solutions.arpa-h.gov/Submit-Solution/ Proposal Submission Site: https://solutions.arpa-h.gov/Submit-Proposal/ Proposal Submission Questions: https://solutions.arpa-h.gov/Ask-A-Question/ Dates: All times listed herein are Eastern Time (ET) Draft Release Date: October 16, 2024 Release Date: November 1, 2024 Questions & Answers (Q&A) Due Date:

Questions may be submitted to CATALYST@arpa-h.gov after the Release Date

Closing Date: Solution Summaries: November 25, 2024 at 5:00PM Proposals: January 31, 2025 at 5:00PM

Anticipated Award: Multiple awards are anticipated.

Types of Instruments That May Be Awarded:

Other Transaction (OT)

Participants/Proposers: Universities, Non-Profit Organizations, Small Businesses and Other than Small Businesses

Any Cost Sharing: Cost sharing may be encouraged or requested mailto:CATALYST@ARPA-H.gov https://solutions.arpa-h.gov/Submit-Solution/ https://solutions.arpa-h.gov/Submit-Proposal/ https://solutions.arpa-h.gov/Ask-A-Question/

2. Description of the Solicitation

2.1. Introduction

The goal of the CATALYST program is to develop AI/ML-enabled in silico human physiology modeling platforms for ADME-Tox simulation to replace the poorly predictive investigational new drug (IND) IND-enabling preclinical animal studies currently used. Drug research and development (R&D) rely on in vivo animal model testing as a preliminary step towards FDA approval. However, the current IND-enabling preclinical animal study paradigm is often an imprecise model for human physiology, leading to an inefficient and expensive drug development process that does not accurately predict human drug safety outcomes. By leveraging innovation within the areas of large-scale data mining, AI/ML predictive analytics, novel high-throughput living systems instrumentation, and simulation-based whole human physiology modeling, CATALYST will create new tools to replace the current approaches that fail to fully recreate human physiology.

CATALYST will develop validated and qualified in silico platforms for novel drug development. By demonstrating their acceptability for assessing drug safety, these platforms will create a new “Digital contract research organization (CRO)” sector that will transform the traditional CRO landscape. The program will focus on Good digital eXperimentation Practice (GXP) pharmacokinetics (including absorption, distribution, metabolism, and excretion) and pharmacodynamics (safety and toxicity) that correlate with various human physiological states. To promote adoption of these platforms upon validation, CATALYST will concurrently fund product sponsors to integrate them into novel drug First in Human (FIH) approval packages. The teaming between the methodology developers and product sponsors is required for CATALYST to be impactful in the drug development industry. Eliminating the need for animal testing in preclinical studies will substantially reduce drug development costs, time, and resources that often prevent drugs with smaller markets from progressing. By providing a more accurate model of human physiological diversity, the in silico modeling platforms have the potential to provide a more reliable estimate of drug efficacy. Improved precision of predictions through CATALYST’s in silico platforms will result in increased success rates in human clinical trials, ultimately leading to faster, less expensive, and more effective drug development benefiting both patients and the broader healthcare ecosystem.

CATALYST envisions creating in silico platforms by incorporating human-relevant preclinical and clinical data while addressing current data gaps through: 1) innovations in data discovery and deep learning approaches to predict drug outcomes while unifying the diverse landscape of public and proprietary data,

2) developing novel living system approaches designed to emulate human physiology, and using these data innovations to 3) create human-based in silico physiology models for comprehensive preclinical assessments. Given CATALYST’s focus on developing tools intended to be used in regulatory filings, methodologies developed within the program must meet good laboratory practice (GLP)-equivalent requirements. Further, these tools must be configured to replace studies that currently rely on higher-order mammals. Critical to the success of the program, CATALYST will require teaming between tool developers and product sponsors at program start to ensure the tools developed by CATALYST performers align to specific Contexts of Use (COU) in regulatory applications. The teaming between tool developers and product sponsors is required for the successful adoption and execution of these studies in Phase II of the program, in which ARPA-H will support the entirety of a candidate product’s IND-enabling study package, including the replacement of existing methods with CATALYST developed tools. The successful completion of CATALYST will radically shift the current industry practice of preclinical and clinical studies to improve patient safety and access to novel therapeutics in clinical trials. CATALYST aims to transform first-in-human trial approval.

2.2. Program Overview

Current methods used in preclinical studies have demonstrated limited utility and are often confounding when compared to human data. The most common in vitro and in vivo preclinical testing methods lack relevance to human physiology, leading to high failure rates of drug candidates entering human clinical trials. For example, in vitro systems can fundamentally alter cell behavior due to incorrect cell microenvironments and are not yet currently developed for systems integration. Despite attempts to ‘humanize’ pre-clinical models, animal models are often poor predictors of human reactions to therapeutics due to interspecies differences in physiology and pathophysiology. Finally, the current computational models rely on allometric scaling and in vitro to in vivo extrapolation (IVIVE), which rely on the data generated from the animal models and in vitro models that fail to adequately represent humans (as mentioned above). These computational tools are generally not predictors of success across the drug development continuum and are not enabled for use in regulatory applications. This results in the present success rate of candidate therapeutics below 10%, with substantial failures during clinical trials and during post-licensure surveillance due to safety issues. Additionally, while recent changes in law and regulations have sought to ensure that clinical trials include diverse populations, barriers still exist to including pediatric, women of child-bearing age, pregnant women, and geriatric populations. Additional factors such as genetic and epigenetic variance, age, gender, metabolic rate, and disease background are too diverse to be accounted for in current clinical trial designs and preclinical animal studies. Innovation is needed to shift the paradigm and enable the ability to evaluate safety and efficacy in these populations preclinically.

In silico human physiology modeling platforms have the potential to revolutionize current preclinical and clinical studies. Successful implementation of in silico models will result in a substantial decrease in study costs and unnecessary animal usage, while significantly increasing the success rate of clinical trials through accurate prediction of the safety, toxicity, and availability of therapeutic candidates. They can also optimize FIH clinical trial design while informing other ways to optimize the success of a drug candidate.

CATALYST will revolutionize current preclinical and clinical study practices with the following deliverables:

• AI/ML-enabled in silico human physiology modeling platforms for ADME-Tox simulation, that replace IND-enabling preclinical animal studies. The platforms will be adaptable to create fit-for-use in silico models of various major organ systems in the human body.

• Living system tools that generate new human data, and fill human data gaps, for the development of in silico platforms for ADME-Tox simulation.

• Independently verified in silico platforms using existing or new human clinical data and comparable validated assays.

• Demonstrated predictive capabilities of the in silico platforms through regulatory qualification of these platforms, and proof-of-concept FIH approvals based on data generated by these platforms.

Figure 1. Examples of the inputs and outputs of the end state pursued by CATAYLST through the creation of “Digital CRO” that will add GXP in silico methods for use in regulatory applications in addition to current advancements in drug discovery and development further upstream.

Success in CATALYST will yield adopted methodologies that have been applied to IND-enabling studies, enhanced predictability of human outcomes in clinical trials, and a path towards digital clinical trials, more predictive Phase 0 studies, and drug candidates that have been evaluated in backgrounds that represent the population diversity of the United States. Technologies and methods developed in CATALYST will also create new market opportunities for “Digital CROs” to support more predictive regulatory science interaction that will enable easier clinical entry of drugs with smaller markets or that target underserved diseases and populations.

The ambitious goals of the CATALYST program align to broader U.S. government strategies such as the, White House Office of Science and Technology policy’s AI Aspirations for Health (https://ai.gov/aspirations/). However, the approaches pursued here differ from other U.S. government research that seeks to improve new approach methodologies (or NAMs) in biomedical R&D broadly, such as the NIH Common Fund Complement-ARIE (Complement Animal Research In Experimentation) effort.

CATALYST will pursue in-depth development of a predictive and regulatory-aligned platform with detailed requirements to be pursued in the first phase of CATALYST, and the CATALYST program will exclude generalized NAM approaches to be more broadly applied to biomedical R&D. CATALYST teams will then demonstrate COU for drug development and adoption of technologies to advance pharmaceutical candidates into FIH clinical trial approval in Phase II of the 5-year program.

2.3. Program Structure and Technical Approach

2.3.1. Program Structure

CATALYST is divided into two phases:

Phase I: Accelerated Technology Development and Method Qualification (36 months) – In Phase I, teams will focus on developing technologies that meet the in silico predictive goals of CATALYST. This will be through the pursuit of three technical areas (see Section 2.3.2) led by teams built around the “Digital CRO” concept. At the end of Phase I, teams will submit their technologies into novel drug development tool qualification pipelines that will facilitate their use in IND-enabling studies. Proposing teams must also include a Product Sponsor (see Section 2.3.2) will advise on COU and provide additional support to platform development in Phase I. Upon successful Phase I completion, Product Sponsors will lead the https://ai.gov/aspirations/ adoption of these platform methodologies in Phase II, as described below;

Phase II: COU Concept Demonstration and Methodology Adoption – In Phase II, the team composition will shift towards Product Sponsor leadership, in which drug developers will pursue regulatory filings that include the in silico platform developed in Phase I in place of traditional, less predictive methods that rely on higher-order mammal experimentation. CATALYST will support IND-enabling experiments to include the replacement of traditional methods and any additional data generation needed to support regulatory approval towards human clinical trials. Successful execution in Phase II of the program will demonstrate the potential of the in silico platform methodologies developed in Phase I and the reality of the “Digital CRO” concept supporting more accurate, cheaper, faster, and most importantly safer outcomes for clinical trial participants and users of licensed products.

Figure 2. CATALYST Program Overview. The in silico Method Developers and Researchers (Left) include all three TAs. The teaming (Center) requires all facilitators as team members, and the Phase I and II of CATALYST (Right) are led by method developers and product sponsors, respectively.

The Other Transaction (OT) agreement is anticipated to be utilized in a team structure, with “team” meaning a group of organizations working together to accomplish a common goal, with members sharing resources, knowledge, and expertise, as opposed to a more traditional prime/subperformer structure. Although one member may lead the team and serve as its authorized agent for administrative purposes such as executing documents or receiving payment on behalf of the team, each member must be bound to the team membership agreement, must be a party to the resultant OT award with ARPA-H, and each member must perform substantive technical work as part of the team. By employing this teaming approach to the CATALYST execution, proposing teams will pursue technology development, proof-of-concept evaluation, and method adoption across the two phases. This approach is also not exclusive to a single methodology developer and a single product sponsor. Instead, it can involve larger groups of developers and sponsors who may have ready applications for the developed tools within their current drug development pipelines. More information about this type of teaming approach can be found at Section 4.5.

2.3.2. Technical Areas (TAs) and Product Sponsors

CATALYST will create predictive modeling platforms for qualified drug safety applications by utilizing AI/ML modeling, supported by data capture in public and governmental spaces, and by developing novel methods to address data gaps and validate the platforms. The development of underlying technologies will facilitate the generation of human-relevant data, enabling the production of verifiable and comparable datasets. The program components include three technical areas (TA): data discovery methods and predictive drug safety models (TA1), living systems tools for model development (TA2), and in silico simulation-based human physiology models (TA3).

TA1, Data discovery methods for predictive drug safety models TA1 performers will capture and unify a wide range of existing preclinical, clinical, and electronic medical record (EMR) datasets that are available in the public and government domains, harmonize and curate these in a repository that can also integrate future datasets from the method development community, and produce a GXP environment and data governance platform to assure data quality and security. TA1 will procure foundation datasets for accurate deep learning methodologies to be developed within the team. The platform should also incorporate enhanced governance features for security, privacy and access control. The end result of this TA is the compilation of existing data into a framework that, either alone or with data gathered from TA2 and TA3, can be used to predict ADME-tox for a given drug.

TA2, Living systems tools for model development TA2 will utilize both existing and emerging technologies, such as human and animal MPS, organoids, instrumented tissues, and ex vivo models to develop innovative tools that will generate datasets to inform the model development to occur in TA3. TA2 should be integrated with the other TAs that require data generation using novel methods for model development. TA2 technologies will result in novel data streams that understand, at fundamental levels, human physiology and its direct interaction with candidate pharmaceuticals.

TA3, In silico human physiology models TA3 will utilize the datasets prepared and produced by TA1 and TA2 to generate and train generalized predictive platforms that are interpretable and explainable using various mathematical rules and AI algorithms and test the fidelity of the platforms using validated datasets (TA1 and TA2) and methods (TA2). Trained models should take multiple chemical, biochemical, and preclinical inputs and produce whole human predictive ADME-Tox modeling, including confidence metrics and other detailed outputs, needed for GXP validation and inclusion in IND submissions.

In addition to the identified TAs, the proposer1 teams for CATALYST should identify a pharmaceutical product sponsor as an enabling element of the program and must have a Product Sponsor on their team within twelve months of award. The Product Sponsor will implement the methodologies created in CATALYST Phase I to IND-enabling studies that will be performed in Phase II. Although the bulk of their work will be in Phase II, their inclusion in the team during Phase I will allow for the evaluation of COU and other product specific requirements for the tools in creation.

Product Sponsor, Novel drug development activity The Product Sponsor will test the predictive platforms for specific COUs and provide feedback to refine TA1/2/3 for regulatory qualifications. As a drug developer, the Product Sponsor will utilize the platforms developed in Phase I of CATALYST in IND-enabling applications and regulatory submissions in Phase II.

2.3.3. Proposal Options

1 Proposer refers to all respondents to this Innovative Solutions Opening at all stages of the ISO.

Applicants may submit proposals for one of the following options:

Option A) A focused proposal targeting only TA1 Option B) A focused proposal targeting both TA2 and TA3 Option C) A comprehensive proposal addressing TA1, TA2, and TA3

The proposer teams for all options should include a Product Sponsor, as the proposer team’s ability to execute Phase I and Phase II will be evaluated in evaluation Criterion #2 (see Section 6.1.2). Proposals addressing combination of TAs other than the specified options will not be reviewed.

2.3.3.1. TA1, Data discovery methods for predictive drug safety models

TA1 will build the tools needed to develop deep-learning approaches to predict drug safety from large data sets while also discovering, cataloging, and creating repositories for the community to use. This platform will be equipped with a range of tools that can curate and harmonize publicly available data and leverage non-public data also. With the anticipation of large datasets produced from CATALYST teams, TA1 will facilitate and implement data standardization and automated data curation in data generation workflow as well.

An in silico human physiology simulation platform that is capable of producing data to support IND application will need to incorporate pharmacokinetic/pharmacodynamic (PK/PD) models that have been developed and refined over decades, while also seamlessly integrating sophisticated human physiology models. A central data and model repository, underpinning the platform, should encompass all aspects of human biology, ranging from the molecular level understanding of xenobiotic properties and their interaction with proteome through macroscopic human physiological effects. Modeling a physiology-level simulation requires many diverse data types. There is a vast accumulation of data from clinical/preclinical studies and DMPK (Drug Metabolism and PK) studies, which are readily available in the public domain through databases such as PubChem, DrugBank, and ChEMBL. In addition to this, other relevant data such as physicochemical properties of compounds and biologics, anatomical and physiological data, molecular-level biochemical data, and pathophysiological data are necessary. These can be combined with specific disease or patient information. An inclusive structural database, either containing predicted or experimental data on protein-protein or protein-small molecule interactions, is another essential element. The ability to predict critical points in metabolic pathways where perturbation by small molecules or biologics have in systemic effect is also crucial. Additionally, being able to predict allergenicity, immunogenicity, or non-specific interactions can be important for biologics.

While data available in the bioinformatics databases are vast in both variety and depth, including genomic sequences, experimental or predicted protein structures, metabolic pathways, xenobiotic degradation and elimination, and allergenicity or immunogenicity of protein modalities, it is critical to acknowledge that these databases often contain inaccurate or missing annotations and are not yet formatted for ML applications. Therefore, addressing these issues are vital for AI/ML-centric data curation effort and would be required for the models developed by TA1 to be accurate in their predictive potential.

A successful proposal should consider each of the following, and provide strategies to achieve each goal:

• A plan to coordinate data across TA2 and TA3 (if being pursued by the proposer team) by providing a centralized data repository platform. This platform should be capable of communicating with and be compatible with the data enclave established by the program through the development and harmonization of data dictionaries.

o Coordination and collaboration plan with the data enclave o Components of a data repository o Components of a data dictionary

• Data integration from diverse biomedical data types across biological principals. The database should include, but not limited to:

o Physicochemical property data; description of molecular weight, solubility, absorption characteristics, distribution behaviors, and stability that are critical to assess bioavailability o Genomic and proteomic data; experimental and predicted protein structures, and expression and protein-protein interactions to understand pathways and therapeutic targets o Metabolic and signaling pathway data; enzymatic activities, reaction kinetics, and specific pathways involved in the metabolism and biotransformation to understand xenobiotic metabolism o Immunological data; allergenicity and immunogenicity of various haptens and protein modalities to predict immune responses to assess profiles o Physiological and pathophysiological data; anatomical, physiological, and pathophysiological data, including tissue-specific information and variations due to diseases or patient-specific factors for accurate modeling

• To achieve data integration, the proposal and developed platform should consider:

o Automated data ingestion and preparation

Data quality enhancement AI-guided data exploration and analysis o Pattern and anomaly detection o Data standardization o Data cleaning and normalization o Ontology integration o Quality assurance o Implementation of API for user accessibility

• Discussion of potential obstacles that could require a revision in the work plan or milestones with a discussion of alternative approaches.

• Detailed schedule or timeline for each milestone and the overall goal.

• Coordination with other TAs:

o Close coordination with TA2 to integrate data from novel methods o Collaboration with TA3 to ensure data is AI/ML-ready and aligns with modeling platform requirements

By carefully curating publicly available bioinformatics databases, and potentially leveraging additional sources of data, TA1 teams will effectively integrate into a robust, high-quality repository that is ready for utilization in AI/ML applications. TA1 teams will then use these data to develop purpose built deep-learning models to be applied to CATALYST objectives. This repository and associated models can also serve as a crucial resource for developing in silico human physiology simulation platforms, ultimately contributing to advancements in the drug development process.

The TA1 metrics are outlined in Table 2 of Section 2.4.2. Monthly technical and financial status reports will be required and discussed with the ARPA-H Program Manager Team at monthly meetings. ARPA-H may request performer data as deemed necessary throughout the program to verify the project progress.

The datasets, toolkits, and the pipeline developed by performers will be shared with Independent Verification & Validation (IV&V), which may consist of extramural and intramural USG labs, for analysis and comparison. Additionally, they may also serve as IV&V during certain phases of the program to verify findings.

2.3.3.2. TA2, Living systems tools for model development

TA2 aims to develop and deploy innovative tools based on in vitro and ex vivo models to enhance data collection and analysis to fill gaps present in the current data landscape. These gaps include the lack of data based on human systems, the need for higher throughput generation of the quality of data needed to model human physiology, and better integration of living systems to reflect human complexity. To accomplish this, TA2 will leverage biological models such as MPS, organoids, instrumented tissues, and ex vivo models. In addition, comparator animal models can be developed to assess vast preclinical animal study data. Most importantly, TA2 will strategically select and develop novel models in close collaboration with TA3 at the start of the program, focusing specifically on systems and pathophysiological states that the project team is targeting. The targeted system should pose significant medical challenges due to its complex nature and disease implications, ensuring that these novel methods are valuable tools in advancing pharmaceutical research.

A successful proposal should consider each of the following, and provide strategies to achieve each goal:

• Identification of the current state of the art (SOA) and plans for improvement to achieve the objective of closing the data gap.

o Close coordination with TA1 to scope the scientific and clinical data landscape, identify data gaps, and design and develop novel tools to bridge those gaps.

o Close collaboration with TA3 to identify limitations of the current SOA prediction platforms and design and develop novel tools to improve predictability through the generation of relevant data.

• Innovation and integration of data collection o Integration of real-time sensor readouts o Integration of high-resolution, non-invasive and non-disruptive imaging o Implementation of multi-omics technologies in the analysis pipeline o AI/ML-ready data generation following the data dictionary set by TA1

• Consideration of biomarker identification for endpoint analysis

• Consideration of comparator models with animal cell lines

• Incorporation of genetic diversities into physiology models

• Platform standardization and guaranteed reproducibility

• Discussion of potential obstacles that could require a revision in the work plan or milestones with a discussion of alternative approaches.

• Detailed schedule or timeline for each milestone and the overall goal

• Coordination with other TAs:

o Close coordination with TA1 to ensure data generated aligns with the data dictionary and is AI/ML-ready.

o Collaboration with TA3 to strategically select and develop novel models that address limitations in current prediction platforms.

The TA2 metrics are outlined in Table 3 of Section 2.4.3. Monthly technical and financial status reports will be required and discussed with the ARPA-H Program Manager Team. ARPA-H may request performer data as deemed necessary throughout the program to verify the project progress. The datasets, toolkits, and the pipeline developed by performers will be shared with IV&V, which may consist of extramural and intramural USG labs, for analysis and comparison. Additionally, they may also serve as IV&V during certain phases of the program to verify findings.

2.3.3.3. TA3, In silico human physiology models

TA3 aims to achieve revolutionary improvements in mechanistic models to address the limitations of current models and provide a more comprehensive and accurate representation of human physiology that is capable of generating animal-free IND-enabling data through collaboration with TA1 and TA2.

Recent advancements in computational science and the surge of cheminformatics and multi-omics data have significantly influenced the field of drug discovery and development. Computational modeling is at the core of drug discovery in the post-genomic era, aiding in various processes such as chemical space sampling, lead scaffold identification, implementation of synthesizability in chemical design, and risk mitigation in lead compound selection. The field of biologics development has also benefited from computational approaches, particularly in improving functional capabilities, de novo protein design, enhancing manufacturability and extending half-life.

The failure of clinical trials represents a missed opportunity to offer treatment options to patients.

Historically, high rate of clinical trial failures and post-market drug attrition rates have been attributed to toxicity and lack of efficacy. This is in part due to inadequate predictive capabilities of preclinical tools such as animal studies and in vitro models. While ongoing efforts aim to enhance human relevance in preclinical studies, the risk of clinical trial failures can be greatly reduced if accurate PK/PD modeling is possible with human relevant data, and drug attrition can be mitigated if clinical trials can predict chronic and idiosyncratic toxicities by accounting all population diversity into clinical trial subjects.

Pharmacometrics aims to mathematically model and simulate the ADME processes in both drug development and clinical application. There are two major types of models utilized: data-driven empirical models and hypothesis-driven mechanistic models. Empirical models, such as population PK, describe observed data and may include physiological parameters to account for individual variability in pharmacokinetics. On the other hand, mechanistic models provide a mathematical explanation of the drug ADME processes by leveraging physiological parameters and in vitro data.

Mechanistic models, particularly physiologically-based pharmacokinetic (PBPK) models, have been increasingly successful in achieving regulatory waiver from the FDA and EMA for drug applications, effectively addressing questions such as absorption modeling and manufacturing quality support in scale-up and post-approval changes (SUPAC). However, revolutionary advances need to be made for mechanistic PK/PD predictive simulators by incorporating more human-relevant data, intensive sampling of cellular-level information, and a comprehensive understanding of human metabolism and cell signaling at the biochemical and molecular levels.

A successful proposal should consider each of the following, and provide strategies to achieve each goal:

• Identification of the current state of the art (SOA) and plans for improvement in predictability of simulation platforms.

o Leverage recent advancements in the computational science of AI/ML o Integration of proteome-wide understanding of xenobiotic metabolism o Promotion of technical push in TA2 through identification of data requirements for model development o Safety and toxicity prediction o Verification of platforms through close collaboration with TA1 and TA2

• Healthy physiology modeling with safety and toxicology focus.

o Broad impact of safety and toxicology in drug discovery and development o Prediction of biological pathways and biomarkers and endpoints for acute and chronic toxicity o Inclusion of genetic, environmental, and dietary factors for individual-specific modeling o Equity and health impact consideration

• Future-proof expandability and flexibility for application in other disease area.

o Consideration of platform design to allow integration of new data and models o Close collaboration with TA1 and TA2 to standardize the input data format o Expandable for complex simulations

• Discussion of potential obstacles that could require a revision in the work plan or milestones with a discussion of alternative approaches.

• Detailed schedule or timeline for each milestone and the overall goal.

• Coordination with other TAs:

o Close collaboration with TA1 to utilize curated data repository and ensure data is in standardized format o Coordination with TA2 to identify data requirements and integrate data from novel methods o Work with TA1 and TA2 to verify and validate developed platforms

The TA3 metrics outlined in Table 4 of Section 2.4.4 will increase in difficulty and complexity over the course of the CATALYST program. Monthly technical and financial status reports will be required and discussed with the ARPA-H Program Manager Team. ARPA-H may request performer data as deemed necessary throughout the program to verify the project progress. The datasets, toolkits, and the pipeline developed by performers will be shared with IV&V, which may consist of extramural and intramural USG labs, for analysis and comparison. Additionally, they may also serve as IV&V during certain phases of the program to verify findings.

2.3.3.4. Product Sponsor, Novel drug development activity

In silico simulation platforms bring myriad benefits, including reduced development costs, decreased drug development timelines, and enhanced predictive capacities for drug safety and efficacy profiles. The platforms developed by TA1/2/3 performers must achieve these potentials for real-world applications.

During the Phase I of CATALYST, Product Sponsor(s) will play an advisory role for performer teams in various aspects of the project. They will provide guidance on disease area selection, ensuring that the chosen areas align with the program's objectives and have the potential for significant impact. Additionally, Product Sponsor(s) will offer their expertise in navigating the drug development tool qualification process, working closely with the teams to ensure that the in silico platforms meet the required standards and are suitable for use in IND application. They will also contribute to FDA IND-application process, assisting in the preparation and submission of necessary documents and communicating with the regulatory agency. As a liaison between the FDA and the performer teams, Product Sponsor(s) will facilitate effective communication and collaboration, helping to streamline the development and validation of the in silico platforms.

In Phase II, Product Sponsor will test these innovative tools in real-world scenarios by implementing these platforms in their drug development process. They will validate and verify the platforms by comparing the results with their own preclinical and clinical study results. Ultimately, Product Sponsor will evaluate whether the simulation data produced through the platforms has the predictive power to replace the current practice of preclinical IND-enabling studies. The evaluation will also include the identification of potential improvement areas of the platforms. With this objective in mind, Product Sponsor should be in the late stage of the drug discovery process or the early development stage, having identified and validated potential drug targets and lead molecules, to participate in the program.

A successful proposal should include the following for Product Sponsor in Phase I, and provide strategies to achieve each goal:

• Past experience and success in drug development tool (DDT) development

• Disease area target selection and fit-for-use identification together with TA1/2/3

• Intellectual property (IP), market analysis and commercialization strategies

• Stakeholder identification and engagement

• Regulatory approval strategy

In addition, the proposal should include the following for Phase II, and provide strategies to achieve each goal:

• Possession of validated drug targets and lead molecules.

o Produce pseudo-clinical data by leveraging the technologies developed by TA1/2/3 o The lead molecule should be validated through medicinal chemistry studies (structure-activity relationship, or SAR), as well as in vitro and in vivo studies focusing on the mechanism of action (MOA) or mode of inhibition (MOI). Additionally, rescreening should be performed o Available preclinical data, which can be compared to and validative of the IND-enabling novel in vitro/ex vivo models and simulation platforms developed in TA1/2/3 o Identified therapeutic indications, correlated biomarkers, and efficacy endpoints for the validated drug targets and lead molecules

• Pre-IND and IND-application plans.

o Plan for in silico simulation platform integration and evaluation o FIH clinical trial design and pre-IND meeting o Sharing of clinical study data with the TA1/2/3 team

• Consideration of disease area to promote equity of the program.

• Consideration of collaboration and data sharing strategy with TA1/2/3 performers.

• Discussion of potential obstacles that could require a revision in the work plan or milestones with a discussion of alternative approaches.

• Detailed schedule or timeline for each milestone and the overall goal.

The Product Sponsor’s requirements and metrics are outlined in Table 5 of Section 2.4.5. Monthly technical and financial status reports will be required and discussed with the ARPA-H Program Manager Team.

ARPA-H may request performer data as deemed necessary throughout the program to verify the project progress. The datasets, toolkits, and the pipeline developed by performers will be shared with IV&V, which may consist of extramural and intramural USG labs, for analysis and comparison. Additionally, they may also serve as IV&V during certain phases of the program to verify findings.

2.3.4. Program Execution and Options

Timeline: The overall CATALYST timeline is structured as a 54-month effort consisting of 2 phases; Phase I with a 36-month effort and Phase II with an 18-month effort. Phase I will integrate pre-clinical and clinical publicly available study data available, as well as discover methods to use non-publicly available data, to develop predictive deep-learning models (TA1) while also generating exploratory and validation data using novel living systems methods (TA2) for multi-system in silico physiology model development and validation (TA3). Phase II will be spearheaded by Product Sponsor(s), who will carry out the IND application while contracting the methodologies developed in CATALYST Phase I without experimentation.

Structure: In Phase I, TA1/2/3 performers will focus on developing predictive in silico platforms, and the submitted proposals should address all TAs, TA1 only, or TA2 and TA3 only (see Section 2.3.2). The ML-centric comparative data generation for modeling will take place within TA2, while TA1 will curate the data from TA2 as well as the existing relevant data in the public and USG space. Additionally, the curated data will be accessed and discovered through novel means. Predictive platform design will occur in TA3 using the training data curated by TA1.

Down-Selection: The baseline performance testing for validation and verification of the platform will be done at the end of Month 18, with the datasets set aside for this specific purpose by TA1. The performers will be asked to provide assessment metrics for the stage. In consultation with ARPA-H program management and commercialization/regulatory experts, the performer team will identify the COU of the predictive model by making minimal modifications to the platform algorithms to ensure its suitability for specific purposes. Additional data generation through new experimentations by TA2 and iteration will be necessary to improve models. The iteration steps should be done through collaboration among all TA performers and Product Sponsor.

The performance validation and verification of the platform by IV&V will be carried out at the end of 30-month using datasets and government partner validation capabilities that are unique to the CATALYST program. The performers will spend the final 6 months in platform refinement. The performer teams that successfully complete the verification stages will have the opportunity to bring their predictive platform to regulatory COU qualification process or another FDA-qualification program (such as the Innovative Science and Technology Approaches for New Drugs (ISTAND) Pilot Program, which will assess the platform's ability to meet the necessary standards for regulatory purposes. The platforms should also satisfy other FDA guidance (e.g., FDA-2013-D-1464, FDA-2019-D-2398, FDA-2020-D-1517, etc.) for pharmacometric modeling. This will ensure that the platforms are compliant with the current FDA guidance. All performers should also develop an FDA engagement plan prior to CATALYST inauguration to understand and familiarize themselves with the FDA's guidelines and requirements for predictive platforms.

Figure 3. The program structure and timeline of CATALYST by phase and TAs. TA: technical area, mo:

month, SOA: state of the art, COU: context of use, IND: investigational new drug (application), FIH: first in human.

Performers who have successfully developed a predictive prototype platform, which incorporates comprehensive coverage of human physiology and ADME-Tox information for drug profile prediction and have accomplished Phase I milestones and IV&V verification/validation, will be considered for CATALYST Phase II, which will be led by Product Sponsor within the proposer team. During Phase II, the performers will concentrate on concurrently developing pharmaceutical candidates through both traditional animal testing, in vitro models and pilot digital pathways. The transition phase is designed to facilitate the integration of predictive platforms and data across multiple performers, and to select the optimal sources for preclinical and clinical study candidates. Phase II involves the creation of product regulatory filings using digital-only data or hybrid of digital data and in vitro model data (No Animal Challenge) for the FDA.

Proposal Requirements: To ensure the applicability of tools developed to the broader community and for the success of CATALYST platforms, performers must have demonstrated team capabilities per the selection of TA submission pathway. Proposals that fail to address the technical area(s) required, as noted above, will be deemed non-conforming and may be rejected without further review. The proposal must also include data and platform access plans and commercialization plans including FDA meeting milestones, technology transfer milestones, platform proof-of-concept objectives, and market analysis and partnership models for commercialization. The predictive platforms must meet the specifications listed in Section 2.4.

2.3.5. ELSI

ARPA-H is committed to developing an in silico platform for drug development that will test PK/PD while carefully considering the ethical, legal, and social implications (ELSI) for broad subpopulations of Americans. The platform can address the challenges faced by drug developers when considering diverse subpopulations, such as individuals with various genetic, metabolic, age, and health characteristics, into clinical trials. By considering the broad spectrum of human diversity, the platform aims to promote more inclusive and equitable drug development processes.

ELSI objectives for the in silico platform may include:

• Ensuring equitable representation of diverse subpopulations in the platform's data and models.

• Addressing disparities in drug efficacy and safety across different subpopulations.

• Minimizing risks to vulnerable populations, such as pregnant women, fetuses, and children.

• Enhancing pediatric drug development by prioritizing safe and effective dosing regimens.

• Protecting patient privacy and data security in compliance with relevant laws and regulations.

• Fostering transparency and public trust through open communication with stakeholders.

The in silico platform should be designed to address the challenges associated with clinical trial design and account for population diversity by: 1) simulating diverse subpopulations to predict drug responses in groups that may be difficult to include in clinical studies, 2) predicting complex drug-drug interactions and optimizing dosing regimens for various subpopulations, and 3) informing and optimizing clinical trial design to ensure trials are more representative of diverse patient populations.

By prioritizing these ELSI objectives, ARPA-H can contribute to the development of safer, more effective, and more equitable therapeutics for all Americans.

2.3.6. Data Management and Sharing Plan (DMSP)

The DMSP shall include all information included in the Six Element plan format recommended by the National Institutes of Health (to view the Six Element suggested format visit https://grants.nih.gov/grants/forms/data-management-and-sharing-plan-format-page). The six elements are:

1. Data Type

2. Related Tools, Software and/or Code

3. Standards

4. Data Preservation, Access, and Associated Timelines https://grants.nih.gov/grants/forms/data-management-and-sharing-plan-format-page

5. Access, Distribution, or Reuse Considerations

6. Oversight of Data Management and Sharing

Within the DMSP, proposers for each TA must include a detailed plan of what types of data, platforms, or portions of platforms they will be sharing with the scientific community as a result of the program. All non-proprietary data should be shared with the scientific community promptly within a year of generation. The specific repository method should be discussed and chosen in agreement with the ARPA-H program manager. In addition, the proposers must provide an explicit plan for timely material, data, and platform exchange between all team members on the proposal, as the free flow of information is critical for the success of the predictive platform development and its verification. The data should be transmitted frequently, in a timely manner, and in its entirety. The development of living systems tools by TA2 and their dissemination will require performers to prepare plans for sharing and distribution of non-data resources that will be generated by the proposed project, including cell line origin, experimental tools and specifications, protocols, biomaterials, and reagents.

The proposers will need to present explicit solutions to address the significant data storage and computing challenges presented by the program, with the understanding that the plans and repository may change later in the program.

2.3.7. Assessment Checkpoints

The program metrics state the assessment decisions in Section 2.4 and the down selection description in Section 2.3.3. When specific end-of-phase metrics are not met, a No-Go decision is made. To transition to Phase II, performers must have a Product Sponsor partner as an element of the CATALYST agreement.

The Product Sponsor will utilize their predictive in silico platforms for IND applications, and the platform must be acceptable for regulatory purposes by the FDA. Any performer that is non-compliant to the data sharing requirements or the equity requirements may not be selected for Phase II.

2.4. Program Goals and Metrics

To evaluate the effectiveness of a proposed solution in achieving the stated program objectives, the following program goals and metrics will serve as the basis for determination of satisfactory progress to warrant continued funding. Although the program metrics are specified below, proposers should note that the Government has identified these goals with the intention of bounding the scope of effort while affording maximum flexibility, creativity, and innovation of proposed solutions to the goals. Proposals must cite the quantitative and qualitative success criteria that the effort will achieve at each phase’s program milestone, as well as the measurement of intermediary metrics. If the metrics are not meaningful for a particular case, proposers are expected to provide their own metrics and describe the quantitative improvement that those metrics represent over the state-of-the-art. Power analysis calculations may be needed to support the proposed metrics.

2.4.1. Overall Program Goals

The overall objectives of CATALYST are listed in Table 1 and should be referenced within the context of individual project efforts. Progress checks for Phase I include the creation of team roadmaps for constructing the in silico ADME-Tox and biodistribution/tox platform, supported by preliminary results generated by the teams. Interim assessments include the down-selection of the most promising platform approaches verified by the IV&V, and the use of the platforms by product development teams. The end of Phase II assessments for CATALYST focuses on the ability to fully replace animal models by the parallel development of animal-free in silico and in vitro/ex vivo approaches in IND-enabling data production. This results in final assessments where novel drug developers file IND applications based on data from only animal-free in silico studies.

Table 1. Overall CATALYST Objectives

Objective Specifics Data repository • Establish a data repository for current platform improvement and a foundation for future development

• Capture relevant data, such as cheminformatics, toxicological, metabolic, and multi-omics data

• Facilitate data sharing among different…