A23 Attachment ARPA-H-MAI-24-01-03 AMD 02.pdf

MODULE ANNOUNCEMENT

FOR

ADVANCED RESEARCH PROJECTS AGENCY FOR HEALTH

ADVANCED ANALYSIS FOR PRECISION CANCER THERAPY

(ADAPT)

ARPA-H-MAI-24-01-03

AMENDMENT 02

APRIL 3, 2024

AMENDMENT 02 is being issued to clarify additional technical and phase requirements within numerous areas of the Opportunity Description. Submission requirements have changed; see Section 5 for changes. All changes are illustrated in red font; word deletions are in red font with a strikethrough (i.e., example).

TABLE OF CONTENTS

1. MODULE ANNOUNCEMENT OVERVIEW INFORMATION

2. OPPORTUNITY DESCRIPTION

3. AWARD INFORMATION

4. ELIGIBILITY

5. MODULE ANNOUNCEMENT RESPONSES

6. PROPOSAL EVALUATION AND SELECTION

7. ADMINISTRATIVE AND NATIONAL POLICY REQUIREMENTS

8. POINT OF CONTACT INFORMATION

9. QUESTIONS & ANSWERS (Q&A)

10. APPENDIX A

11. REFERENCES

ATTACHMENT 1: OTHER TRANSACTION BUNDLE (VOLUME 1)

1. MODULE ANNOUNCEMENT OVERVIEW INFORMATION

FEDERAL AGENCY NAME: Advanced Research Projects Agency for Health (ARPA-H)

OPPORTUNITY TITLE: ADvanced Analysis for Precision cancer Therapy (ADAPT)

ANNOUNCEMENT TYPE: Module Announcement under MAI ARPA-H-MAI-24-01

OPPORTUNITY NUMBER: ARPA-H-MAI-24-01-03

DATES: (All times listed herein are Eastern Time) o Module Announcement release date: 03/07/2024 o Virtual Proposers’ Day: 03/15/2024 o Questions & Answers (Q&A) due date: 03/22/2024 o Questions & Answers (Q&A) release date: 04/01/2024 o Proposal due date: 05/06/2024

2. OPPORTUNITY DESCRIPTION

The mission of the Advanced Research Projects Agency for-Health (ARPA-H) is to accelerate better health outcomes for everyone by advancing innovative research that addresses society's most challenging health problems. Awardees will develop groundbreaking new ways to tackle health-related challenges through high potential, high-impact biomedical and health research. ARPA-H seeks proposals to revolutionize new adaptive strategies for treating the evolution of cancer. The ADvanced Analysis for Precision cancer Therapy (ADAPT) Program will develop an adaptive cancer treatment platform that detects when tumors change, recommends updates to the treatment plan, and evaluates revised plans through a novel clinical trial design. The ADAPT program will examine a tumor’s behavior as it changes and will then match each patient’s evolving cancer with the best available therapy. ADAPT will revolutionize cancer care by bringing new science and medicine together to map and target tumor changes that improve survival for patients with metastatic cancer. This continuous and interconnected learning process fosters rapid innovation whereby tumor biology research provides evidence that rapidly informs and shapes clinical practice.

A. PROGRAM INTRODUCTION

Each year, 600,000 Americans die due to metastatic cancer, representing approximately 17% of all deaths.1 Reducing this number requires appropriately matching treatments (e.g., chemotherapy, immunotherapy, targeted therapy, etc.) with a patient’s particular tumor biology at multiple points in time as cancer progresses. Biomarkers measure indicators (e.g., genes, transcripts, proteins, etc.) of a tumor’s biology, and can help clinicians assess the likelihood that a particular therapy will be effective for a given patient. Few new biomarkers have recently been discovered because cancer research has been constrained by using single data types (DNA, protein, imaging) rather than combining a broad range of tumor biology measurements and clinical data. Additionally, most biomarkers are based on single gene measurements, which limits predictive accuracy. Lastly, both the research and clinical domains have remained siloed, preventing sufficient patient data from enabling the discovery and validation of new predictive biomarkers.

This chasm is particularly problematic when it comes to metastatic cancer, which requires that clinical decisions keep pace with rapidly evolving tumors. ADAPT utilizes advances in tumor biology measurement data, new data analysis methods and algorithms, and an innovative evolutionary clinical trial design to treat the right patient with the right drug at the right time.

Tumors are not static; instead, they mutate and change to keep growing, developing resistance to treatments that were initially effective. Today’s static treatment regimens lead to deaths that could be avoided with a more dynamic treatment approach. To successfully combat cancer, treatment strategies need to adapt as tumors evolve. Advances in tumor measurement and analysis technologies make it possible to detect changes in tumor traits during growth with ever-increasing fidelity, scaling from single to multiple data types. Novel measurements of tumor biology could be used to detect when cancer cells begin to resist treatment so that treatment plans can be adjusted. Armed with this information, doctors would then select more effective treatments that specifically target these new resistant traits. Selecting treatments with increased effectiveness would improve survival time for cancer patients so that they can live with metastatic disease instead of dying from it. Creating a standardized mechanism for selecting treatments that respond to changes in tumor biology is essential for prolonging the life of patients with metastatic cancers.

At its core, ADAPT will revolutionize predictive biomarkers for cancer and create the next generation of predictive biomarkers that capture changes in tumor biology over time, respond to resistant traits with better treatments, and extract insights from a diversity of tumor and patient data. Biomarkers will sequentially be discovered, tested, and implemented in the new evolutionary clinical trial to guide therapy selection as the tumor evolves pre- and post-treatments. The ADAPT program will examine a tumor’s behavior as it changes and will then match each patient's evolving cancer with the best available therapy, with the goal of improving progression free survival (PFS) by 50% in at least one patient subgroup. Each patient’s history of tumor adaptations to treatment will contribute to a collective knowledge base from hundreds of other patients with similar tumor types and treatment results. This collective evolutionary history and individual patient data will inform treatment through a new sequential evolutionary clinical trial approach that will enrich the precision and efficacy of patient care. ADAPT will revolutionize cancer care by bringing new science and medicine together to map and target tumor changes to improve survival for patients with metastatic cancer.

NATIONAL HEALTH IMPACT

Two million Americans receive a new cancer diagnosis annually with over 600,000 individuals currently living with metastatic cancer2. More than 90% of metastatic cancer patients develop resistance to therapy3.

Advanced cancer is responsible for one in six deaths in the U.S., and one in three individuals is expected to encounter cancer in their lifetime1. The ADAPT program will apply these novel approaches to three leading causes of cancer related death in the United States: breast, lung, and colon cancers. Lung cancer is the third most common cancer, yet it is the leading cause of cancer related deaths in Americans2. Metastatic non-small cell lung cancer represents a significant area for treatment improvement, as response rates to therapy are less than 50% and survival time is less than one year from diagnosis of metastasis4,5. Colon cancer is the second leading cause of cancer related deaths in Americans, and response to second line therapy is less than 25% with a progression free survival (PFS) time of less than 6 months6-7. Breast cancer is the most diagnosed cancer, and the second leading cause of cancer related death in American women1. Among patients with metastatic estrogen receptor positive breast cancer, response rates to second line therapies are <25% with a PFS of less than one year8-10.

The number of patients with metastatic and terminal cancer is on the rise; and the economic burden of metastatic cancer treatment is substantial, exceeding $200 billion in medical care costs in 2020 across all cancer types11. For example, metastatic breast cancer alone incurred over $35 billion in 2020, which is projected to surpass $86 billion annually by 203012. Similarly, the combined costs for advanced and metastatic lung and colon cancers in 2020 was over 48 billion11. Alongside rising medical costs, significant productivity and other societal costs are also associated with cancer13. Importantly, based on an economic analysis of current precision oncology, using tumor biology data in treatment decisions significantly lowered average per week healthcare costs including inpatient and outpatient expenses, resource utilization, and end-of-life costs14. Based on the assessments of cost savings from precision care in this study, a conservative estimate of savings from the ADAPT program for breast, lung, and colon cancer is 6.8 billion dollars per year.

Substantial costs are also associated with drug development and approval. The cost of developing a new prescription drug that gains market approval is expected to exceed $2.6 billion15. Furthermore, the estimated median cost of clinical trials supporting approval of oncology drugs was $45 million each based on oncology trials conducted in 2015-201716. By advancing precision care through new technologies, expenses for drug development can be reduced while ensuring patients receive the most suitable therapy.

B. TECHNICAL AREAS (TA)

Due to significant advancements in tumor measurement and analysis techniques, tremendous opportunities exist to develop biomarkers that improve drug selection for patients. These innovations will enable the identification of optimal drug targets and treatment strategies for individual patients.

The ADAPT Program is comprised of three interconnected technology areas (TAs), outlined below and in Figure 1:

• TA1 – Therapy Recommendation Techniques: Develop data-driven methods that determine resistant cancer traits, discover new predictive biomarkers of drug response, and guide optimized treatments based on a tumor’s genetic and phenotypic traits. TA1 comprises three sub-TAs including:

o TA1.1: Multi-Modal Data Fusion: Combine information from multi-modal data types to create a unified representation of tumor data that offers a more comprehensive and accurate understanding than any single data source alone.

o TA1.2: Resistant Trait Modeling: Develop tumor evolution models that increase number and precision at which resistant traits are discovered and that enable identification of new treatment strategies of targeting resistance traits.

o TA1.3: Biomarkers that Predict Drug Response: Develop and test multi-modal biomarkers that will be implemented within an evolutionary clinical trial (TA2).

• TA2 – Evolutionary Clinical Trial: Design a trial that adjusts drug treatments based on how the tumor evolves during multiple sequences of therapy. TA2 comprises three sub-TAs including:

o TA2.1: New Tumor Measurement Technologies: Collect a broad diversity of temporal tumor measurements from the evolutionary clinical trial (TA2.2) for resistance trait identification and predictive biomarker development in TA1.

o TA2.2: Evolutionary Trial Protocol: Develop a centralized, modular, open-source protocol design capable of responding to tumor changes in near-real-time.

o TA2.3: Evaluation of TA1 Biomarkers: Test and integrate predictive biomarkers and adaptation of subsequent lines of therapy into the evolutionary trial based on data collected from prior treatments.

• TA3 – Treatment & Analysis Platform: Create open-source collaboration space for developing, analyzing, and sharing data, models, and trial protocols between researchers and clinicians.

Performers may submit a single proposal addressing any one TA (TA1, TA2 or TA3) or sub-TA of TA1 (TA1, TA1.1, TA1.2, TA1.3) or performers may submit a single proposal addressing a combination of multiple TAs; (see Section 2C). All performers are expected to collaborate with each other; therefore, performers will have Associate Contractor Agreement (ACA) language included in their respective award; see Section G.

INTENTIONALLY LEFT BLANK

B.1. TECHNICAL AREAS (TAS)

TA1 – THERAPY RECOMMENDATION TECHNIQUES

Tumors change their DNA, RNA, and proteins to keep growing, even during therapy. These changes can lead to drug resistance making it harder to treat cancer effectively. Data analysis methods have the capability to detect these changes and identify tumor resistance traits as well as develop biomarkers that predict the most effective drug treatments. TA1 will include three parallel and interconnected subsections (sub-TAs): 1) multi-modal data fusion; 2) resistance trait modeling during tumor evolution; and 3) biomarkers that predict drug response. Performers will include systems biologists, biomedical data scientists, bioinformaticians, statisticians, and mathematicians working collaboratively to leverage assessments of tumor biology, including genetic and protein-level data (DNA- and RNA-sequencing, proteomics, and circulating tumor DNA/RNA/methylome data) along with clinical data (e.g., electronic health records (EHR), imaging, pathology, and blood-based tests). Collectively, these comprehensive tumor biology measurements will be termed multi-modal data.

• TA1.1 – Multi-Modal Data Fusion: Combine information from multi-modal data types to create a unified representation of tumor data that offers a more comprehensive and accurate understanding than any single data source alone.

• TA1.2 – Resistant Trait Modeling: Develop tumor evolution models that increase number and precision at which resistant traits are discovered and that enable identification of new treatment strategies of targeting resistance traits.

• TA1.3 – Biomarkers that Predict Drug Response: Develop and test multi-modal biomarkers that will be implemented within an evolutionary clinical trial (TA2).

As explained in Section B, proposers may submit a single proposal to one sub-TA or a single proposal for multiple sub-TAs for TA1. All TA1 proposals should describe how they plan to meet the metrics outlined in Table 1.

TA1 performers will receive data from the clinical trial in TA2, and thus, need not provide data independent from the ADAPT program. As mentioned below, TA2 performers will establish the evolutionary clinical trial protocol and infrastructure during the first 6 months of the program in Phase I, Stage I. During this

Figure 1. The ADAPT Program’s TA1, TA2, and TA3 will jointly develop, test, share, and iteratively refine predictive biomarkers and new treatment regimens. TA1 will analyze tumor traits and build predictive biomarkers from multi-modal data derived from tumor measurements and clinical data collected by TA2. TA2 will develop a new type of clinical trial called an evolutionary clinical trial that will enable clinicians to dynamically adjust therapy and optimize drug selection based on near-real-time changes in tumor biology, allowing them to overcome tumor resistance and improve patient survival times. TA2 will take longitudinal tumor measurements to aid subsequent biomarker development, testing and implementation. Data and analysis algorithms will be shared in the treatment and analysis platform (TA3) to connect researchers and clinician teams as they develop new biomarker-driven drug strategies. This cycle, from tumor measurement and analysis to biomarker development, testing, and implementation within the evolutionary trial, enables a new approach in the treatment of cancer patients where data rapidly, consistently, and accurately informs patient care.

lead time, TA1 performers will be expected to establish a framework for their multi-modal data fusion (TA1.1), resistance trait modeling (TA1.2), and drug response biomarker development (TA1.3) capabilities using either their own, previously obtained data, publicly available data, or synthetic data. This framework will then be further refined when clinical data becomes available from TA2 performers.

TA1.1 - MULTI-MODAL DATA FUSION

Multi-modal measurements enable the study of a tumor from different angles using high-dimensional data.

However, the large amount of information derived across technology platforms poses multiple challenges for data processing. Patient data can span different tumor measurements (e.g., DNA, RNA, single cell sequencing, proteomics) and clinical data (e.g., EHR and imaging data); however, biomarkers are historically established using a singular data modality. Further, obstacles such as a lack of usable data, sparsity of diverse data, and limited multi-modal interpretability and standardization hinder the discovery and analysis of new biomarkers. TA1.1 will devise data fusion methodologies that integrate multi-modal data types thereby enhancing the predictive efficacy of biomarkers developed by TA1.3.

Strong TA1.1 proposals should describe novel data fusion approaches that go beyond the state of the art.

Proposers should describe how their fusion approach will integrate genomic, proteomic, and protein features from tumors and/or clinical data such as EHR, imaging, and pathological data. In addition, proposers should describe feature selection strategies that reduce the dimensionality of EHR, imaging, and multi-omic datasets from millions of dimensions down to the tens or hundreds that are most relevant to therapy recommendation.

To this end, TA1.1 proposals must describe the design, development, and evaluation of fusion techniques capable of generating a body of high quality, complex, and integrated features from multi-modal data that will enhance the quality of downstream biomarker development and resistant trait analysis. Strong proposals will describe data integration methodologies for both structured and unstructured data. TA1.1 performers will also standardize multi-modal data for mathematical and machine learning (ML) algorithms by rectifying distribution, variance, and structure differences across data sets. TA1.1 proposals must describe normalization and integration approaches, which may include but are not limited to early/middle/late fusion strategies, multimodal ML, transformer neural networks, perceiver architecture, and transfer learning methods. Fused data will be provided to TA1.2 and TA1.3 performers for resistant trait modeling and use in the development of predictive biomarkers.

The following multi-modal data fusion characteristics are required:

1. Approaches must address either early-data fusion, middle-data fusion, late-data fusion, or combinations of these, both including fusion of research-omic data and clinical data from TA2 performers as well as integration of biomarkers from individual data modalities into composite biomarkers.

2. Approaches that include multiple data types for fusion.

3. Methods must be reproducible and transparent.

The following multi-modal data fusion characteristics are out of scope:

1. Approaches that do not address early-fusion, middle-fusion, or late-fusion of research and clinical data.

2. Approaches that do not include multiple data types for fusion.

3. Methods that are not reproducible or transparent.

TA1.2 - RESISTANT TRAIT MODELING

To date, researchers have developed low-fidelity models of tumor progression; however, such efforts have been limited by the number of time samples collected, and work has predominantly focused on one or two data types at a time. TA1.2 performers will increase the fidelity of tumor models by expanding the breadth of data types captured in mechanistic models, improving the temporal resolution of models, and increasing the number of resistant traits identified. Such advances will convert discrete time series measurements into a continuous representation of tumor size and progression, enabling near-real-time treatment recommendations to optimize clinical decision making. TA1.2 proposals must describe different mathematical approaches to model tumor evolution and acquisition of drug resistant traits. These methods may include but are not limited to probabilistic generative long short-term memory, feed forward neural networks, causal artificial intelligence (AI), hierarchical generalized additive models, matrix analysis, Gaussian processes, clustering methods (point-wise distance-based, feature-based, model-base), ARIMA (autoregressive integrated moving average) and Hold-Winters (Triple Exponential Smoothing).

The following resistant trait modeling characteristics are required:

1. Approaches must analyze multi-modal data, including research and clinical data from TA2 performers, across multiple time points.

2. Approaches must identify pathway and gene/protein-level traits and drug targets that emerge during treatment.

3. Approaches must include dynamical/temporal modeling, tracking tumor evolution.

4. Methods must be reproducible and transparent.

The following resistant trait modeling characteristics are preferred:

1. Include both modeling (dry lab) components to identify resistant traits using multi-modal data from cancer patients as well as a pharmacology (wet lab) component to validate drug resistance mechanisms and drug efficacy. Wet lab components may use human cell lines, human iPSC-derived materials, or primary samples from human patients that have been collected independently of the ADAPT program. Tissue collected as part of the ADAPT program will not be available for TA1.2 experiments.

The following resistant trait modeling characteristics are out of scope:

1. Approaches that only apply to a single data type.

2. Approaches that do not discover new drug resistant tumor traits.

3. Approaches that do not model tumor measurements over time.

4. Methods that are not reproducible and transparent.

5. Methods incorporating patient derived xenograft (PDX) or other animal models.

6. In vitro models using primary patient samples from TA2, including spheroid/organoid cultures.

Performers should not include data storage and analysis costs, as these will be covered by the ADAPT program and provided as a Government Furnished Resource.

TA1.3 - BIOMARKERS THAT PREDICT DRUG RESPONSE

TA1.3 seeks to build first-in-class biomarkers for standard therapies across at least three cancer types while increasing the accuracy of biomarkers to predict the right therapy for each patient using new advanced tumor characterization methods. Selecting an effective treatment plan hinges on accurately determining the current state of a tumor and predicting the best therapy; yet current clinical trials do not use biomarkers to adjust treatment regimens. To integrate biomarker discovery and evaluation with clinical trial design, TA1.3 performers will develop predictive biomarkers by analyzing comprehensive patient data such as genetic profiles, circulating tumor DNA in blood samples, medical imaging, and electronic health records from TA2 performers using advanced tumor characterization methods, which will then be tested and/or implemented in the TA2 clinical trial. TA1.3 proposals must provide detailed explanations of mathematical, statistical, and computational methods for evolutionary cancer analysis and biomarker development, which may include but are not limited to ML approaches (e.g., support vector machine (SVM), random forest, gradient and boosting, LASSO, ridge, elastic nets, deep neural networks), regression analyses (e.g., binary regression, mixed effect models, logistic regression), and mechanistic models (e.g., ordinary differential equations (ODEs), partial differential equations (PDEs), Lotka-Volterra equations, Ricker models, adaptive dynamic models, and reaction-diffusion equations). TA1.3 biomarkers will be evaluated based on the ability to accurately predict drug response in a clinical setting. Performers are welcome to propose approaches that leverage any of the aforementioned algorithms or others not on the list, so long as there is a sound plan to generate effective predictive biomarkers in clinically relevant timeframes.

The following biomarker characteristics are required:

1. Biomarkers must be developed using advanced mathematical and/or computational approaches (e.g., AI, ML, mechanistic models, etc.) for prediction of drug response. Algorithms for biomarker development should aim to be applicable in a diversity of cancer settings, facilitating generalizability of methods across cancer types and settings.

2. Approaches must use multi-modal data to test drug prediction capabilities through TA2’s evolutionary clinical trial.

3. The method used to develop biomarkers must generalize to different cancer types.

4. Approaches must dynamically adapt to tumor response during treatment.

5. Biomarkers must predict drug response to a currently used therapy or drug class in metastatic cancer treatment.

6. Biomarkers must use a complexity of human data for predictions with a minimum of >5 mutations, transcripts, or proteins.

7. All biomarkers used in ADAPT, whether pre-existing or novel, will be subject to rigorous independent validation to ensure efficacy and patient safety.

8. Approaches must be integrated into the TA3 platform and be testable by external reviewers.

9. TA1 performers must agree to collaborate to develop common Application Programming Interfaces

(APIs) and data models to transfer algorithmic outputs and supporting data to TA2 and TA3, when applicable. By the end of the program, these open APIs and data models should form a foundation for future open standards. They should relevant open standards and data models, including Health Level 7 Fast Healthcare Interoperability Resources (HL7 FHIR) and United States Core Data for Interoperability Plus (USCDI+) whenever possible.

The following biomarker characteristics are preferred:

1. Existing biomarkers previously applied to patient data and have demonstrated significant predictive capability are welcomed. A general pipeline should be proposed for biomarker development; those already tested in patient data for at least one drug are preferred but not necessary.

2. Models that are informed by tumor biology/signaling.

3. Proposers must that show evidence that they have developed at least one biomarker that has been independently validated in multiple patient cohorts as evidence of past performance.

4. Biomarkers developed during the ADAPT program should be interpretable, and they should reflect a tumor phenotype and biological cancer characteristics rather than empiric characteristics.

The following biomarker characteristics are out of scope:

1. Biomarkers that are not derived using state-of-the-art computational techniques including AI, ML, mechanistic models, and other advanced mathematical approaches.

2. Biomarkers that only include a limited number (<5) gene/mutations, transcripts, or proteins.

3. Biomarker development approaches that cannot be generalized to multiple therapies and/or tumor types.

4. Biomarkers that are not relevant for metastatic cancer.

5. “Black box” biomarkers that cannot be validated. All biomarker algorithms must be open for review.

6. Biomarkers developed using only PDX or other animal models. without testing in patient samples.

Note, PDX or other animal models are outside the scope of ADAPT.

TA2 – EVOLUTIONARY CLINICAL TRIAL:

Historically, cancer clinical trials have followed a uniform approach in which efficacy of a single treatment regimen is tested in patients by collecting sparse data, giving an incomplete picture of tumor biology and response. This design precludes adaptation of treatments to effectively address drug-resistant tumors. TA2 will improve cancer treatment by tightly integrating advanced tumor analysis methods from TA1.2 to identify resistant traits as they emerge so clinicians can adapt treatments to target these traits. This adaptive approach integrates the predictive biomarkers developed in TA1.3 for initial selection of optimal standard of care therapy, and subsequent identification of new treatments to overcome resistance. A single, institutional review board (IRB)-approved evolutionary clinical trial protocol will be used to test biomarkers and drug treatments in sequence, thus promoting adaptation of this model.

TA2 is comprised of three subsections (sub-TAs):

• TA2.1 – New Tumor Measurement Technologies: Collect a broad diversity of temporal tumor measurements from the evolutionary clinical trial (TA2.2) for resistance trait identification and predictive biomarker development in TA1.

• TA2.2 – Evolutionary Trial Protocol: Develop a centralized, modular, open-source protocol design capable of responding to tumor changes in near-real-time.

• TA2.3 – Evaluation of TA1 Biomarkers: Test and integrate predictive biomarkers and adaptation of subsequent lines of therapy into the evolutionary trial based on data collected from prior treatments.

TA2.1 - NEW TUMOR MEASUREMENT TECHNOLOGIES

Serial tumor measurements are required for understanding how a patient’s unique cancer changes in response to treatment. Historically, data from tumor biology studies have been limited in precision, breadth, and duration, restricting the potential of personalized cancer treatments. Recent advancements in single cell analysis methods have enabled the holistic measurement of tens of thousands of individual cells within a tumor, while other biotechnological breakthroughs have enabled serial tumor measurements from blood collection. However, interpretation of these measurements requires computational resources and analysis expertise beyond the scope of randomized controlled clinical trials. TA2.1 performers will provide multiple tumor measurement data types for incorporation into predictive biomarkers, which will in turn be used to inform patient therapy selection.

TA2.1 performers will collect a broad diversity of temporal tumor measurements from the evolutionary clinical trial (TA2.2) for resistance trait identification and predictive biomarker development in TA1. TA2 proposals must describe multi-modal data collection approaches and tissue biopsy analysis methods that enable dynamical modeling and precise characterization of a tumor’s biology. These approaches should be combined and may include but are not limited to bulk and single cell DNA- and RNA-sequencing, blood-based assays such as circulating tumor DNA/RNA/methylome sequencing, and spatial single cell assays (e.g., transcriptomics, proteomics, and mutation analysis). Proposals may also consider how the tumor environment contributes to drug response, as well as how the cancer and non-cancer cells interact to promote an environment that facilitates tumor growth. Proposals that leverage continuous patient health monitoring technologies will be considered relevant to the program. A list of required tumor measurement features is provided at the end of TA2.

TA2.2 - EVOLUTIONARY TRIAL PROTOCOL

The TA2.2 clinical trial framework will ideally combine elements of several types of trials to treat patients sequentially during multiple lines of therapy while collecting tumor measurement and clinical data for biomarker development and implementation for drug response predictions. These elements could include randomized trials (a test of response between two different treatment regimens), classical adaptive trials (which allow for modifications to the trial protocols based on interim results), n-of-1 trials (where each individual patient is given a control and experimental treatment) and umbrella trials (where a variety of treatments are tested simultaneously on patients). In the ADAPT evolutionary trial, patients remain within a single clinical trial protocol transitioning between treatments based on their tumor's unique characteristics identified from serial samples taken throughout the trial. Performers should propose a cancer setting where biomarkers are needed to optimize patient response to sequential therapies, where new resistance mechanisms need to be discovered, and where patients have a short average progression free survival on current therapies. Examples are provided in Appendix A.

ADAPT will also allow modular clinical trial approaches that can be broadly enacted at diverse sites both within the ADAPT program, but also can be repurposed and used independently by other clinical teams using the TA3 platform infrastructure. A list of required evolutionary clinical trial features for TA2 proposals is provided at the end of TA2. Performers will develop an evolutionary protocol for either metastatic breast, lung or colon cancer that follows patient treatments and response while collecting data to address critical needs for individualized and improved therapy decision capabilities. TA2 performers will work closely with TA1 performers to develop and test biomarkers. TA2 performers will select multiple sites for the clinical trial to enable diverse and rapid patient enrollment.

The ADAPT Program Manager team will select a Contract Research Organization (CRO) that is well suited to manage the regulatory affairs, data management, auditing, clinical trial planning, protocol management, site initiation, recruitment support, clinical monitoring, biomarker testing/integration, sample collection and distribution, among other centralized clinical trial components. TA2 proposals should not include CRO costs in their budgets, nor should they include data storage and analysis costs, as these will be covered by the ADAPT program.

TA2.3 - EVALUATION OF TA1 BIOMARKERS

The predictive accuracy of biomarkers developed by TA1.3 performers will be tested by TA2 performers, informing the iterative course of therapeutic regimens. This built-in process of biomarker identification and validation will be deployed across multiple lines of treatment and cancer types. Incorporating biomarker discovery and validation directly into the clinical trial process will improve their translational effectiveness.

Biomarkers will enter ADAPT at different levels of development and testing. The evolutionary trial will enroll patients in therapies that would benefit from biomarkers for drug selection and collect multi-modal data for biomarker development (TA1.3). Prior to biomarker testing in an evolutionary clinical trial arm, statisticians will pre-define the predictive accuracy needed for statistical significance. Once biomarkers are developed, they can be evaluated on patients independent from the biomarker development cohort within the same evolutionary trial protocol.

Proposers must provide the following information in TA2 proposals:

TA2 proposal requirements

(i) Tumor measurement characteristics:

1. Tissue or blood samples must be of sufficient mass and quality to support research data extraction (e.g., sequencing).

2. Planned tumor biology measurements should include a diversity of data types: bulk and single cell DNA- and RNA-sequencing, blood-based assays such as circulating tumor DNA/RNA/methylome sequencing, and spatial single cell assays (e.g., transcriptomics, proteomics, and mutation analysis).

Measurements may be added to the clinical trial, if feasible, based on other performer capabilities.

3. Raw and summarized data from measurements will be made available to and used by performers and researchers.

4. Analysis of measurements must be feasible within one month from collection and processing.

(ii) Clinical trial characteristics:

1. Clinical trial for metastatic cancer patients.

2. Clinical trial that includes multiple lines of therapy under a single protocol.

3. Flexible clinical trial protocol that enables changes in design (e.g., randomized, adaptive, etc.).

4. Strong capability to collect tumor and blood-based measurements.

5. Possess a treatment line with a Progression Free Survival (PFS) rate of <9 months average.

6. The protocol must include mechanisms for biomarker evaluation.

7. Modular clinical trial design enabling rapid dissemination and protocol uptake by different sites.

(iii) General characteristics:

1. Details about key clinical needs being addressed (e.g., targeting chemotherapy resistance or endocrine therapy resistance).

2. Evolutionary clinical trial design examples (see Appendix A) and plans for rapid patient enrollment (see Metrics table 2).

3. Lines of treatments to be included in the clinical trial.

4. Plans for biomarker testing and integration into the clinical trial.

5. Historical clinical trial enrollment for 2020-2023 in the specific cancer type including patient demographic and diversity metrics. Proposers should detail a plan to initiate clinical trials and start enrolling patients within 6 months of funding.

6. Information about the ability to collect high quality patient samples, specifically tumor biopsies, blood, etc. Past successes should be detailed for large scale translational research that includes a diversity of patient samples and multi-modal data types.

7. Information about the ability to collect and use clinical data such as electronic health records (EHR), imaging data, pathology reports, genomics, clinical tests, etc. Note any restrictions on data use and/or sharing.

8. Whenever possible all relevant data should be structured and transferred in accordance with the HL7 FHIR standard.

Out of Scope for TA2 proposals

(i) Tumor measurement characteristics:

1. Research data that does not measure tumor biology.

2. In vitro experiments using primary cells from patient samples, such as spheroids or organoids.

3. Mouse models such as PDX using primary cells from patient samples.

4. Research data that cannot be made available to all performers and researchers.

5. Measurements that take >1 month to process for analysis by performers.

6. Patient data that does not directly contribute towards the research on or clinical application of cancer therapies.

(ii) Clinical Trial characteristics:

1. Clinical trials that do not include metastatic cancer patients.

2. Clinical trials that do not use a single protocol with multiple lines of therapy.

3. Clinical trials that do not allow for changes in therapies.

4. Clinical trials that do not collect serial samples.

5. Clinical trials that do not allow for integration of predictive biomarkers.

6. Protocols that do not include mechanisms for biomarker evaluation.

For additional information on the design and execution of evolutionary clinical trials, see Appendix A.

TA3 – TREATMENT AND ANALYSIS PLATFORM:

Currently many clinicians face a steep learning curve to find, access, explore, and analyze biomedical research data, especially multi-modal datasets, as the data are stored in siloed data repositories, and existing tools are designed for those with informatics skillsets. TA3 will fill this void by creating a collaborative, user-friendly platform, and testing ground that bridges the research and clinical domains to improve access to data and toolsets that facilitate adaptive therapy recommendation techniques for cancer. The platform’s data, algorithms, and clinical trial protocols will be made publicly available through the ADAPT portal thus creating a new standard for the field and alleviating roadblocks that have limited the impact of big data in basic and translational cancer research. To achieve this goal, TA3 performers will focus on five critical areas:

1. Building a collaboration ecosystem: TA3 performers will create a treatment and analysis platform that will bring together clinicians and multidisciplinary scientists and enable near-real-time availability of harmonized multi-modal and patient data, protocol distribution, and analysis of therapy regimens.

2. Creating a comprehensive data lake: performers will maintain a centralized repository (data lake) that encompasses various data types, enhancing data accessibility and management. The data lake will provide a scalable data management solution that stores the raw and processed data and will handle a wide array of data types and sources, utilizing cloud-based technologies for flexibility in storage and computing. The data lake will support various tools for data analysis, while also providing data governance and security to maintain data quality and compliance.

3. Facilitate data processing: TA3 performers will develop pipelines to rapidly process tumor biology measurement data, such as sequencing data, for use among TA1 performers. Processing of data will include quality control assessments (such as read counts, base quality scores, etc.), detection and mitigation of sampling biases across multi-source data, data curation, read alignment and mapping, variant calling and annotation, data summarization into user-friendly and standardized data formats, data de-identification (e.g., removal of personally identifiable information (PII) in text and images).

4. Promote data standardization and linkage: TA3 performers will implement data standardization and linkage processes to ensure compatibility and consistency across different tumor biology data types as well as link disparate data types across common sources using privacy-respecting common identifiers. TA3 proposals must describe prototype workflows for processing raw, multi-modal data to generate secondary and tertiary results that improve size, manipulability, and interpretability.

Performers will standardize different data formats, units, and scales for consistency, transforming data for uniformity, and cleaning it to correct errors and inconsistencies. Performers will map and link data based on common identifiers, and ensure data conveys the same meaning across sources, and manage metadata for clarity and transparency. Regular quality assurance and control will be applied to maintain the integrity and usability of the harmonized dataset.

5. Develop open APIs and toolsets: TA3 performers will develop new open APIs to facilitate data access and sharing among stakeholders to promote broad use of the treatment and analysis platform.

Proposals should describe the development of user-friendly interfaces that enable researchers and clinicians to access de-identified data and distribute evolutionary trial protocols to other users. The platform should enable the logical and efficient querying of biomedical data. Proposals should describe the design, development, and evaluation of these front-end and back-end platform interfaces. Similarly, proposers should explain how their approach will enable users to explore and make connections across multiple data types, including but not limited to multi-omics, imaging, longitudinal, and single cell datasets. Proposers are strongly encouraged to maximize the use of existing and emerging cloud-based technologies from the major cloud service providers. Proposers are encouraged to design, architect, develop, and implement intuitive AI-powered dashboards, user-friendly data exploration tools, and data visualization techniques. The key to data usability is understanding where and what data are available and obtaining enough context to understand how the data can be analyzed, explored, or used. To this end, TA3 performers will work with TA1 and TA2 performers to ensure that dashboards, auto-report generation capabilities, and any other tools meet the needs of ADAPT performers and eventually the wider scientific and clinical communities.

Data dashboards and other tools may provide near-real-time summaries of all data with a biomedical data ecosystem, including statistics by data type, disease type, number of patients, and other metrics.

The following treatment and analysis platform features are required:

1. A scalable data management solution that stores the raw and processed multi-modal data from cancer researchers and the evolutionary clinical trial.

2. A computational pipeline to rapidly process tumor biology measurements, perform quality control (QC) analysis, detect sampling bias, manage data curation, mapping, annotation, and linkage of disparate data types from the same patient.

3. Implementation plan for data harmonization to ensure compatibility and consistency across different data sources.

4. Development of open APIs, a data portal, and other tools for data access, sharing, analysis, querying, and visualization.

5. Whenever possible HL7 FHIR APIs and open data standards (e.g., USCDI+) should be leveraged.

The following treatment and analysis platform features are out of scope:

1. A limited data management plan that cannot store raw and processed patient data.

2. An inability to process patient data, perform QC analysis, detect sample biases, or curate, map and annotate data.

3. The lack of a data standardization and linkage plan for multi-modal data.

4. A data portal that does not enable data access, sharing, analysis, querying, and visualization with other performers as well as the public.

C. PROGRAM STRUCTURE AND INTEGRATION

The ARPA-H ADAPT program will develop and demonstrate an adaptive cancer treatment platform that integrates research and clinical practice to enable rapid improvements in cancer therapy. All TA1, TA2, and TA3 performers are expected to work together to achieve program goals. The ADAPT Program’s TA1, TA2, and TA3 performers collaborate to create a system that enables advanced analysis methods to build biomarkers (TA1) that match the right patient to the right therapy at the right time by providing individualized cancer care in a new evolutionary trial (TA2). TA2 performers will collect a broad diversity of longitudinal tumor measurements and clinical data from the evolutionary clinical trial. TA1.1 performers will apply multi-model fusion methods to this data, which will enable resistant trait modeling by TA1.2 and resistant cancer biomarker development by TA1.3. In subsequent iterations, TA1.2 will provide resistance mechanism data and TA1.3 will provide therapy recommendations to TA2 performers based on biomarkers that predict drug response. TA2 performers will evaluate the predictive accuracy of the biomarkers throughout the course of the program. As such, TA2 will improve cancer treatment by tightly integrating advanced tumor analysis methods from TA1 to identify resistant traits as these emerge so clinicians can adapt treatments to target such traits. This adaptive approach uses the predictive biomarkers developed in TA1 for selecting the optimal standards of care therapies that best align with the biology of the patient’s tumor in the initiation phases of clinical trials. Biomarkers are subsequently used for identifying new treatments to overcome emerging resistance.

In addition to data and process sharing, performers across TAs will be responsible for validation and verification efforts. TA3 performers will integrate the data and methods produced by TA1 and TA2 into the portal created in TA3 and ensure the TA1 algorithms run properly so that these algorithms can automatically process new TA2 data as it is delivered to the TA3 platform. TA3 will create a treatment and analysis platform that will enable near-real-time availability of curated and linked data, protocol distribution, and standardization of what constitutes a new biomarker. This platform and set of tools will allow clinicians and researchers to design, analyze, and share innovations from TA1 and TA2 including biomarkers, tumor genomic profiles, health records, models, treatment strategies, and trial protocols. TA3 performers will send the output from the algorithms to TA2 clinical teams so they can interpret the output and check for consistency. Performers from each TA are expected to work closely together, and proposals should address plans to incorporate data from other TAs and collaborate with those performers.

Cumulatively, the ADAPT program includes an infrastructure that enables continuous and iterative learning from patient care data collected and analyzed from the evolutionary trial. This constant and interconnected learning process between TA1-TA3 will foster rapid research innovation whereby clinical practice provides evidence for therapy recommendations and this evidence, in turn, will rapidly inform and shape clinical practice.

ARPA-H anticipates funding multiple performers for TA1-TA3. Proposers may submit a single proposal addressing any one TA (TA1, TA2 or TA3) or sub-TA of TA1 (TA1, TA1.1, TA1.2, TA1.3) or performers may submit a single proposal addressing a combination of multiple TAs. Four (4) to six (6) awards are anticipated for TA1. Four (4) performers are expected to be selected for TA2; TA2 proposals must address all TA2 sub-TAs (TA2.1, TA2.2, and TA2.3) and one of the following cancer types: breast (two performers), lung (one performer), or colon (one performer). Two (2) performers will be selected for TA3, with a planned down-selection to a single performer taking place at program month 12.

Collaboration between multiple types of organizations, academic institutions and companies is highly encouraged.

D. PROGRAM METRICS

The overall program goals for ADAPT are to enroll and treat over 500 and up to 1,000 cancer patients with at least 95% compliance of multi-modal data collection during treatment, to identify a multitude of new drug resistance mechanisms, and to incorporate at least 3 new non-standard-of-care drug therapies based on predictive biomarkers that are derived from multi-modal data and computational analysis. New treatment indications should target resistant cancer traits based on tumor measurements and treat patients predicted to respond as assessed by the biomarker. The program will build, test, and implement multiple predictive biomarkers, and improve patient progression-free survival (PFS) by >50% in at least one clinical trial patient subgroup. Further, an open-source data portal will be populated with harmonized, audited and deidentified data that will be made available to the research and clinical communities. Metrics for each phase of each of the 3 TAs are outlined in the tables below.

Table 1. TA1 Metrics for Each Phase

* Validated biomarkers must have a p-value of <0.05 or R2>0.8 Table 2. TA2 Metrics for Each Phase

Table 3. TA3 Metrics for Each Phase

E. SCHEDULE/MILESTONES

The ARPA-H ADAPT program is a 6-year effort composed of 2 Phases. Phase I is divided into 2 Stages:

an initial six-month Stage 1 to establish…