Computational prediction of anti-cancer drug metabolizing enzymes in the human microbiome

Drugs can be modified by human gut bacteria, leading to variability in efficacy and side-effects across people. Yet most of the ~19,000 FDA-approved drugs have not been tested for bacterial metabolism, and for those that have been screened, the responsible microbial enzymes are rarely known. Anti-cancer drugs epitomize this knowledge gap, with huge patient-to-patient variability and multiple documented links to specific bacterial strains and genes that alter drugs post-administration. This hinders our ability to design, prescribe, and dose cancer chemotherapies accurately and safely. A major roadblock is the immense diversity of microorganisms within a person’s gastrointestinal tract (the gut microbiota), including dynamic variability in enzyme presence/absence across strains of the same species, making it necessary to track causal genes not just taxa. Furthermore, state- of-the-art experimental screening approaches have insufficient scale to accommodate the rapidly growing list of drugs subject to gut bacterial metabolism. To remove these obstacles, we propose to develop a computational technology platform based on chemical and protein similarity that matches microbial enzymes with the drugs they are likely to modify. Supporting feasibility, our multi-PI team developed a prototype of this platform, called Similarity algorithms that Identify MicrobioMe Enzymatic Reactions (SIMMER). In the proposed project, we now aim to overcome three key limitations preventing the SIMMER prototype from being broadly applicable: the paucity of validated reactions for training and evaluation (Aim 1), variable performance across enzyme classes (Aim 2), and inability to query starting from a protein sequence rather than a chemical reaction (Aim 3). We will tackle these challenges by using large language models to incorporate protein structural similarity alongside sequence homology, linking traditionally siloed reaction-centric and sequence-based databases, and generating large-scale functional data to iteratively evaluate and improve SIMMER’s algorithms. The resulting tool will enable users to predict drugs that a given protein could modify and to prioritize gut microbial enzymes capable of performing known drug transformations. We have opted to focus on anti-cancer drugs as an initial proof-of-concept, given the rigorous prior literature implicating the microbiome in cancer therapy and the broad potential for translational impact. SIMMER 2.0 will speed up the discovery of chemotherapy-metabolizing enzymes, enabling focused work on specific drug classes and types of cancer. In addition, SIMMER predictions themselves will be useful for drug design and as inputs to personalized dosing algorithms. This cancer-focused project will be a key milestone towards a comprehensive map of all FDA-approved drugs and their microbial interactions. More broadly, the proposed methods will be easily extendable to other chemicals besides drugs, including diet- and host-derived small molecules. Project Number: 1R01CA313882-01A1 | Fiscal Year: 2026 | NIH Institute/Center: National Cancer Institute (NCI) | Principal Investigator: KATHERINE POLLARD (+1 co-PI) | Institution: J. DAVID GLADSTONE INSTITUTES, SAN FRANCISCO, CA | Award Amount: $747,790 | Activity Code: R01 | Study Section: Biodata Management and Analysis Study Section[BDMA] View on NIH RePORTER: https://reporter.nih.gov/project-details/11363855