Machine Learning Screening Identifies AML Drug Candidates

Researchers at Rice University have developed a machine learning-driven screening pipeline capable of identifying potent drug candidates for acute myeloid leukemia by analyzing over four million compounds. The findings, published in the journal Leukemia, outline a novel computational approach that reverses traditional drug discovery by targeting specific biological functions rather than predefined molecular structures. Led by Professor Natasha Kirienko, the research team programmed the software to prioritize compounds that induce apoptosis, enhance autophagy, and inhibit glutathione reductase. This reverse-engineered screening process efficiently filtered the massive compound library, yielding fewer than one hundred matches. The researchers selected the top twenty candidates for laboratory validation. The study reveals that glutathione reductase inhibition exploits a critical metabolic vulnerability in leukemia cells. AML cells exhibit exceptionally high energy demands, relying heavily on mitochondrial function which inevitably generates reactive oxygen species. While healthy cells possess redundant neutralization pathways to manage this oxidative stress, AML cells lack sufficient backup mechanisms. Inhibiting glutathione reductase disrupts the primary antioxidant defense, causing lethal reactive oxygen species accumulation specifically within cancer cells. Experimental validation confirmed the compounds selective toxicity. Testing against AML cell lines and primary cells derived from patient samples demonstrated significant cancer cell mortality. When administered alongside established leukemia therapies, the new compounds exhibited strong synergistic effects, eliminating between 95 and 97 percent of malignant cells while sparing only 5 percent of healthy hematopoietic cells. The project leveraged multidisciplinary collaboration, incorporating expertise from medicinal chemists at the University of Houston and clinical researchers at the University of Texas MD Anderson Cancer Center. Access to the MD Anderson Leukemia Sample Bank was instrumental in verifying patient-derived efficacy. Kirienko emphasized that the reverse screening methodology establishes a scalable framework for drug discovery, with immediate potential for adaptation to other malignancies beyond acute myeloid leukemia. This computational pipeline marks a significant advancement in precision oncology, demonstrating how machine learning can streamline target identification and compound selection while minimizing off-target toxicity.