A melanoma cell can dodge a targeted drug in dozens of different ways, each driven by a different genetic mutation. But what if many of those escape routes funnel into the same handful of molecular dead ends? That is the central finding from a team at The Rockefeller University, whose new single-cell screening platform, PerturbFate, tracked more than 300 genetic mutations simultaneously in drug-treated cancer cells and found that strikingly different mutations often push cells toward shared resistant states. The work, published in Nature and publicly available as of June 2026, suggests that targeting those convergence points could be a smarter strategy than chasing each mutation one by one.
What PerturbFate actually does
Most genetic screens measure one thing at a time: either which genes are active or how the cell’s DNA is packaged. PerturbFate captures three streams of information from every individual cell in a single experiment. First, it reads chromatin accessibility, revealing whether stretches of DNA are wound tightly shut or open for business. Second, it measures pre-existing RNA, a snapshot of the genes the cell was already using. Third, it tags and sequences newly made RNA using a chemical label called 5-ethynyl uridine, showing which genes the cell is turning on or off right now.
That third layer is what gives the method its name. By distinguishing old transcripts from fresh ones, PerturbFate can infer not just where a cell sits in its molecular landscape but the direction it is heading, essentially capturing cell fate in motion. Each cell also carries a CRISPR-delivered barcode that identifies which specific mutation it harbors, so researchers can link every molecular profile back to a known genetic change.
The team applied PerturbFate to A375 melanoma cells carrying the BRAF(V600E) mutation, a driver found in roughly half of all melanomas. These cells were treated with vemurafenib, a frontline targeted therapy for BRAF-mutant melanoma that initially shrinks tumors but frequently fails as resistance develops. Raw data and experimental metadata are deposited in the Gene Expression Omnibus (accession GSE291147), allowing independent researchers to reanalyze the results.
Why convergence matters for treatment
The headline biological result is that many genetically distinct mutations funneled cells into a small number of overlapping resistant states. Senior author Junyue Cao framed the significance in a Rockefeller University statement: if dozens of mutations converge on the same downstream program, the number of molecular bottlenecks that need to be targeted may be far smaller than the number of mutations driving resistance.
For patients, the implication is practical. Rather than developing a separate drug for every resistance mutation, clinicians might block the shared state that many mutations depend on. That logic mirrors how combination therapies already work in oncology, but PerturbFate offers a systematic way to identify which states are truly shared and which are outliers, at a resolution that previous single-modality screens could not achieve.
Ambitions beyond melanoma
Cao’s team has stated plans to extend PerturbFate to aging and Alzheimer’s disease, where genome-wide association studies have identified scores of risk variants whose functional effects on individual brain cells remain largely unknown. In principle, the same framework could reveal whether Alzheimer’s risk genes converge on shared pathological states in neurons, microglia, or astrocytes, much as melanoma mutations converge on resistant phenotypes.
That application, however, remains aspirational. No public data, preprint, or timeline describes how PerturbFate will be adapted to neuronal models or postmortem human brain tissue. The technical hurdles are substantial: neurons are fragile, brain cell populations are regionally specialized, and modeling complex human risk variants with CRISPR knockouts may oversimplify the biology. Until pilot data emerge, the Alzheimer’s extension is best understood as a stated research direction, not a demonstrated capability.
What the tool has not yet proven
The published proof of concept rests on a single cancer cell line grown in culture. Whether the convergence patterns hold in more complex settings, such as patient-derived tumor samples, xenograft models, or tumors shaped by immune and stromal interactions, is an open question the current data cannot answer. Tumor heterogeneity in real patients could scramble the tidy convergence seen in a controlled dish.
Scalability is another unknown. The screen exceeded 300 perturbations while collecting multi-omic data, a significant technical achievement. But it is not yet clear how readily other laboratories can reproduce that throughput, or whether pushing toward thousands of perturbations will compromise data quality or require prohibitive sequencing costs.
There is also a gap between identifying a convergent resistant state and turning that knowledge into a drug. A shared transcriptional or chromatin program does not automatically point to a protein that is both druggable and safe to inhibit. Some resistance programs may depend on essential cellular machinery that cannot be blocked without harming healthy tissue.
Computational tools working in a related space have shown promise. A peer-reviewed method called PerturbNet demonstrated the ability to predict single-cell responses to unseen chemical and genetic perturbations using machine-learning models trained on single-modality data. Whether PerturbFate’s multi-omic approach yields more accurate or clinically actionable predictions has not been tested head to head; direct comparisons would require standardized benchmarks and shared perturbation libraries.
Separately, a company called PerturbAI recently announced what it describes as the world’s largest in vivo CRISPR atlas. That effort operates in the broader field of large-scale perturbation biology but has no documented connection to the PerturbFate team or dataset, and its claims have not yet undergone peer review.
How the evidence layers from peer review to press release
The strongest foundation for PerturbFate’s claims comes from two sources: the peer-reviewed Nature paper, which details the method’s design, scale, and biological findings; and the public GEO repository, which archives raw data and metadata in a government-backed database open to independent reanalysis. Together, they establish that coupling pooled CRISPR perturbations with combinatorial single-cell multi-omics and metabolic labeling at a scale exceeding 300 mutations is technically feasible, and that convergence onto shared resistant states occurs in at least one controlled melanoma system.
Everything beyond that, including applications to other cancers, neurodegenerative disease, and clinical decision-making, remains a hypothesis awaiting data. The Rockefeller press release outlines the team’s ambitions, but institutional communications carry an inherent incentive to present findings favorably. Independent replication, cross-platform comparisons, and eventual clinical studies will determine whether multi-omic perturbation screens like PerturbFate change how complex diseases are understood and treated, or whether the convergence story proves more complicated once it leaves the culture dish.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
