Real-World Validation of Big Omics-Powered AI Drug Sensitivity Predictor for Acute Myeloid Leukemia Treatment

1. Overview

(1) scientific rationale and importance

Despite the increasing use of targeted therapies, acute myeloid leukemia (AML) remains a highly fatal disease with poor long-term survival. One major barrier is the inability to predict individual patient responses to treatment, especially in cases without known gene mutations. To address this, we are testing our under-development multi-omics–powered AI model, which integrates transcriptomic and proteomic data to predict drug sensitivities with high resolution.

This project aims to validate the AI-predicted therapeutic candidates using patient-derived xenograft (PDX) models, offering a critical real-world test of the platform’s translational value. Launching and completing this project as an open science initiative will significantly enhance the impact and accessibility of our AI predictor for both researchers and clinical healthcare.

(2) confirmation of funding/facilities/ethics approvals

We have secured access to relevant RNA-seq, proteomics, and drug sensitivity datasets, assembled a team of AI engineers and bioinformatics experts, and obtained initial funding to establish the AI predictor within three months. We have also established formal collaborations with leading institutions worldwide—Dana-Farber Cancer Institute (USA), Monash University (Australia), Southeast University (China), and the University of Urbino (Italy)—who will provide PDX resources with ethical approvals, experimental facilities, and data analysis support. Researchers from One Patient One Cure will lead the project by designing experiments, analyzing data, and managing results.

(3) estimated timeline

Months 0–3: RNA-seq and proteomic analyses of PDX samples to generate AI input data

Months 3–6: Drug treatment and response monitoring in PDX models

Months 6–9: Data analysis, AI model refinement, and validation of updated model

Months 9–12: Final data analysis and reporting

(4) data sharing commitment.

All anonymized multi-omic data, AI-predicted drug lists, and treatment response data will be freely available via One Patient One Cure’s public repository and ResearchHub.

2. Introduction

Theoretical Background

As a nonprofit research organization, One Patient One Cure is pioneering a new era of cancer precision treatment by integrating artificial intelligence (AI) with multi-omics data—including transcriptomics and proteomics—to more accurately predict how individual patients will respond to specific drugs. Unlike conventional treatments that rely on one-size-fits-all chemotherapy or limited genomic mutations, our approach focuses on the functional state of the tumor—its gene expression, protein activity, and signaling dynamics—to drive therapy selection.

While genomic variant–based precision medicine has helped a subset of patients, most cancers do not harbor actionable mutations¹. Even mutation-guided therapies often yield short-lived or partial responses. Our platform goes beyond DNA, using AI to analyze high-dimensional molecular profiles and predict optimal treatments tailored to each patient’s unique biology.

Real-world validation and continuous refinement are essential to the success of any AI-driven model. As we begin building our predictor using secured datasets, we have also formed global collaborations to ensure seamless transition to real-world testing. This includes conducting validation experiments using academic partners' PDX resources and platforms.

Acute myeloid leukemia (AML) is the most common acute leukemia, with an estimated ~20,800 new cases and ~11,220 related deaths in the U.S. in 2024². Despite advances in treatment—including targeted therapies and improved chemotherapies—a cure remains elusive for most patients, largely due to the disease’s intrinsic heterogeneity^3,4. For example, combining the BCL-2 inhibitor venetoclax with hypomethylating agents has improved outcomes by better targeting leukemia stem cells^5,6. However, existing biomarkers are drug-centered—matching drugs to patients—without a patient-centered method to identify the best drugs based on real-time biology. Moreover, AML heterogeneity evolves with disease progression and treatment. A patient-centered AI predictor using RNA and protein data could dynamically guide therapy, improving both understanding and treatment outcomes.

We already have the necessary expertise, resources, and research funding through collaborators. However, relying solely on their institutional funding would delay data release for years due to publication timelines. Therefore, we seek dedicated funding for One Patient One Cure to secure intellectual property and release results publicly—fulfilling our mission and that of ResearchHub.

Research Questions

1. Can AI-predicted drugs outperform current first-line chemotherapies in individual AML cases?

2. Can the AI predictor detect dynamic changes in AML heterogeneity and adapt drug predictions accordingly?

3. Will real-world validation help improve the predictive accuracy of the AI model?

Hypotheses

1. AI-predicted compounds will provide superior therapeutic effects and a better therapeutic index than current chemotherapies.

2. The AI predictor can track dynamic changes in AML biology and suggest effective treatment accordingly.

3. Real-world validation feedback will improve future prediction accuracy.

Impact Statement

We anticipate validating all three hypotheses. However, even in the event of unexpected results, this study will provide key insights into the strengths and limitations of AI models. The findings will facilitate wider academic use of our predictor, inform clinical decision-making tools, and advance patient-centered precision medicine for those without actionable mutations.

3. Methods

Participants/Sample

We will test 10 AML PDX cases from our collaborators’ established libraries, each with paired RNA-seq data⁷. Case selection will be random to ensure unbiased sampling, with the availability of sufficient cryopreserved vials as a basic requirement.

Materials and Procedures

To address Research Question 1, we will perform RNA-seq and proteomic analyses for the 10 PDX cases, input the data into our AI predictor, and identify a drug list for each case. The top three clinically applicable drugs will then be tested using a rapid ex vivo drug screening platform⁸. Standard AML therapies—cytarabine, venetoclax, and azacitidine—will serve as benchmarks. Sensitivity will be measured using IC50 and AUC values with CellTiter-Glo assays, and CD34+ normal bone marrow cells will be used to assess the therapeutic index.

For Research Question 2, we will isolate therapy-persistent leukemia cells from three PDX cases post-chemotherapy using dead cell removal kits, and perform RNA-seq on residual disease. The AI predictor will then generate updated drug lists, and the top three drugs per case will be tested ex vivo. Here, pretreated PDX cells will serve as controls to better evaluate the AI’s capacity to respond to evolving disease biology.

To answer Research Question 3, we will refine the AI model based on the outcomes of the first tests. In cases with suboptimal responses, the improved AI predictor will generate new drug predictions, and the revised top three drugs will be tested to assess performance gains.

Planned Analyses

We have developed a complete pipeline to preprocess RNA-seq and proteomic data before AI input. Drug sensitivity will be evaluated using IC50 and AUC. Therapeutic index will be assessed by comparing drug responses between AML cells and either CD34+ normal cells or pretreated PDX controls. t-tests or one-way ANOVA with post-hoc comparisons will be used, depending on the specific comparisons.

Ethics and Data Management

All ethical approvals are secured by our academic collaborators. All sensitive data will be anonymized, and all results will be made publicly available via One Patient One Cure and ResearchHub.

4. Pilot Data (Optional)

We have established validated platforms for rapid ex vivo drug testing using primary patient samples or PDXs, as well as for analyzing RNA-seq and proteomic data to correlate with drug sensitivity. The related data was published in Nature. 2024 Jun;630(8015):198-205. doi: 10.1038/s41586-024-07410-3. PMID: 38720074

5. Budget

Cell culture media and supplies $5,000

Drugs and reagents $5,000

RNA-seq $5,000

Proteomics $10,000

Data sharing and hosting $5,000

Project manager compensation $20,000

Total: $50,000

6. References

1 Malani, D. et al. Implementing a Functional Precision Medicine Tumor Board for Acute Myeloid Leukemia. Cancer Discov 12, 388-401 (2022). https://doi.org:10.1158/2159-8290.CD-21-0410

2 Key Statistics for Acute Myeloid Leukemia (AML), <https://www.cancer.org/cancer/types/acute-myeloid-leukemia/about/key-statistics.html> (2024).

3 Kantarjian, H. M., Kadia, T. M., DiNardo, C. D., Welch, M. A. & Ravandi, F. Acute myeloid leukemia: Treatment and research outlook for 2021 and the MD Anderson approach. Cancer 127, 1186-1207 (2021). https://doi.org:10.1002/cncr.33477

4 Kantarjian, H. et al. Acute myeloid leukemia: current progress and future directions. Blood Cancer J 11, 41 (2021). https://doi.org:10.1038/s41408-021-00425-3

5 Pollyea, D. A. et al. Venetoclax with azacitidine disrupts energy metabolism and targets leukemia stem cells in patients with acute myeloid leukemia. Nat Med 24, 1859-1866 (2018). https://doi.org:10.1038/s41591-018-0233-1

6 Waclawiczek, A. et al. Combinatorial BCL2 Family Expression in Acute Myeloid Leukemia Stem Cells Predicts Clinical Response to Azacitidine/Venetoclax. Cancer Discov 13, 1408-1427 (2023). https://doi.org:10.1158/2159-8290.CD-22-0939

7 Townsend, E. C. et al. The Public Repository of Xenografts Enables Discovery and Randomized Phase II-like Trials in Mice. Cancer Cell 29, 574-586 (2016). https://doi.org:10.1016/j.ccell.2016.03.008

8 Luo, Q. et al. Targetable leukaemia dependency on noncanonical PI3Kgamma signalling. Nature 630, 198-205 (2024). https://doi.org:10.1038/s41586-024-07410-3

Funding

Funding

Real-World Validation of Big Omics-Powered AI Drug Sensitivity Predictor for Acute Myeloid Leukemia Treatment

Funding Progress

Nonprofit Recipient

Funders

Topics

DOI