Biotechnology · us
Tracing Drug Action with RNA Fingerprints: Broader Effects Emerge for Two HDAC6 Inhibitors
Changes in cellular gene activity can offer clues to how drugs act. A new method compares these changes with known experiments to help researchers identify possible targets and reveal how “selectivity” changes with dose and cellular context.
After a drug enters a cell, the gene activity it changes is often easier to measure than the molecules through which it causes those changes. This gap matters for drug development: does the intended target truly drive the response, or do other effects also contribute? A research team from the New York Genome Center, New York University, and Columbia University has proposed an “RNA fingerprinting” method to trace possible causes from cellular responses. According to an October 9 announcement from the research institution, the findings have been published in *Cell*.
The method’s reference data come from single-cell RNA sequencing performed after researchers deliberately reduced the expression of specific genes, applied drugs, or introduced immune signals. Each known intervention may leave a distinctive pattern of gene expression changes. The model first uses multi-condition latent factor analysis to reduce noise and separate out preexisting differences between cells, then uses Bayesian regression to identify one or more fingerprints that best explain the response in a new sample.
Matching does not guarantee a single answer. If several interventions produce similar transcriptional responses, the method retains a group of possible explanations, called a Bayesian credible set; it can also leave a sample unassigned when evidence is insufficient. An early preprint from 2025 tested the method using data from known genetic interventions. In one validation across replicate experiments, approximately 59% of cells received assignments, and 88% of their highest-ranked credible sets contained the correct intervention. These figures reflect performance under specific test conditions and cannot be treated directly as the accuracy for identifying all drug targets.
Drug tests demonstrated the utility of this matching approach. The research announcement stated that cells treated with a BCR-ABL inhibitor matched the BCR fingerprint among thousands of genetic interventions; KDM1A and CDK9 inhibitors also yielded clear matches. The team further examined two compounds described as selective HDAC6 inhibitors. According to the early preprint, ACY-738 and ricolinostat were each tested at eight doses in K562 cells, with AR-42 and tacedinaline used as controls. Neither compound matched the fingerprint of HDAC6 gene knockdown at any tested dose; low-dose responses remained unassigned, while higher doses showed broader signatures of HDAC inhibition.
This result provides a lead that warrants further investigation into claims of selectivity, but the limits must remain clear: similar RNA responses do not amount to direct measurements of a drug’s binding to each protein. Measurements of protein markers in the preprint broadly supported more extensive effects. However, at one tested concentration, ACY-738 produced marker changes more consistent with HDAC6 selectivity, so the authors did not fully rule out selectivity under that condition. These cellular experiments also cannot be used to directly infer efficacy or toxicity in humans.
The method is available through a public R package. The repository includes installation instructions, usage examples, and precomputed genome-wide Perturb-seq and immune stimulation reference fingerprints, supporting comparisons for individual cells or groups of cells. Researchers can use these resources to analyze their own single-cell data and connect drug responses with existing intervention experiments. The repository currently still links to the 2025 preprint; that version is explicitly labeled on PubMed Central as not yet peer reviewed and should be distinguished from the current journal publication.
The practical value of RNA fingerprinting lies in giving the next round of validation clearer direction: which genes or signals deserve priority testing, which doses may introduce additional effects, and which explanations remain difficult to distinguish. Its capabilities are also limited by the reference data; different cell types may respond differently to the same intervention, and effects not covered by the database may go unrecognized. Moving from transcriptional clues to a reliable mechanism of drug action still requires follow-up validation through direct biochemical and functional experiments.