Machine learning identifies leukemia drug candidates and vulnerable target

A machine-learning screening method developed by researchers at Rice University identified fewer than 100 small molecules predicted to combine three functions associated with activity against acute myeloid leukemia (AML), including inhibition of the enzyme glutathione reductase. The researchers selected 20 candidates for laboratory testing and found that they selectively affected AML cells. In combination with existing AML drugs, the compounds killed as many as 95% to 97% of cancer cells while affecting about 5% of healthy blood cells, according to the experimental study.

The work reverses the usual direction of computer-based drug discovery. In a conventional in silico screen, researchers specify a molecular target and ask a model to find compounds capable of acting against it. Natasha Kirienko, a Rice University professor of biosciences and corresponding author, said the team instead began with functions observed in successful drugs from earlier wet-laboratory screens against acute myeloblastic leukemia cells. The researchers analyzed which functional properties were shared by those successful compounds, then trained machine-learning software to search among more than four million molecules for compounds likely to exhibit all three. This approach narrowed the field to just under 100 candidates, from which the team chose 20 for testing. The pipeline was designed to uncover both potential drug compounds and biological targets rather than requiring a target to be known in advance.

Bernadetta Meika, a graduate student and co-author, said the three functions were increased cell death through apoptosis, increased recycling of cellular components through autophagy, and inhibition of glutathione reductase. The compounds with the strongest predicted combinations were evaluated in AML cell lines and in cancer cells donated by patients with AML. Across the experimental models, the researchers observed selective activity against AML cells. This result concerns cells in laboratory studies and does not establish efficacy, safety or clinical benefit in people. The authors also combined the new compounds with existing AML drugs and observed a synergistic effect. Those combinations eliminated up to 95% to 97% of cancer cells in the tested models, compared with about 5% of healthy blood cells under the reported conditions.

Investigating glutathione reductase revealed why inhibiting the enzyme may be particularly effective in AML cells. Mitochondria generate chemical energy but also produce reactive oxygen species, or ROS, which can damage cellular components and must be neutralized. Glutathione reductase is an important part of that neutralization system. Healthy cells have secondary mechanisms that can compensate when the enzyme is inhibited. AML cells, however, have high energy requirements and depend heavily on mitochondrial activity, resulting in a level of ROS production that the backup mechanisms cannot adequately handle. The combined effect of reduced antioxidant capacity and sustained oxidative stress damages and kills the leukemia cells. The study therefore identifies inhibition of glutathione reductase as a specific vulnerability of AML cells in the tested models, while also showing that the machine-learning framework can reveal targets through predicted compound functions.

The project brought together computational biologists, chemists, biologists and clinicians, with collaborators from the University of Houston and The University of Texas MD Anderson Cancer Center. MD Anderson also provided access to patient-derived samples through its Leukemia Sample Bank. Kirienko said the resulting pipeline could be adapted to search for new compounds and targets in other cancers, but that broader application would require additional testing. The experimental study was published in the journal Leukemia as “Cheminformatic identification of small molecules targeting acute myeloid leukemia.” Its findings support further investigation of the screened molecules and glutathione reductase as an AML target, rather than a claim that the candidates are ready for therapeutic use.

Subject of Research: Machine-learning identification of small-molecule candidates and a metabolic vulnerability in acute myeloid leukemia cells

Article Title: New method identifies leukemia drug candidates and target

Article References: New method identifies leukemia drug candidates and target. Original publication (Bibliographic details unavailable.)

Image Credits: AI Generated

DOI: Not provided

Keywords: acute myeloid leukemia, machine learning, drug discovery, cheminformatics, glutathione reductase, autophagy, apoptosis, reactive oxygen species, cancer cell lines, small molecules, method, identifies