Companion Article — Technology & Diagnostics

How AI Cell-Reading Technology Could Revolutionize Prostate Cancer Diagnosis

A new AI model trained on over half a million prostate cancer cells can identify cell types with 94% accuracy — and points toward a future where every biopsy tells a complete molecular story, not a single architectural grade

IPCSG Research Staff · Companion to: Song et al., bioRxiv 2026 · Industry Analysis Included

BLUF — Bottom Line Up Front

The UCSF prostate cancer atlas described in our companion article does more than advance our scientific understanding of how prostate cancer progresses — it introduces a specific AI tool called PCformer that achieves 94% accuracy in identifying individual cancer and normal cell types from single-cell genetic data. This technology, combined with the commercial infrastructure already being built by genomics giants like Illumina, points toward a fundamental transformation of pathology: from a system built on what cells look like under a microscope, to one built on what cells are doing at the molecular level. For prostate cancer patients, this means diagnostic reports that could one day reveal not just a Gleason grade but a complete cell-by-cell molecular map of their tumor — including dangerous transitions invisible to any microscope. This transformation is not science fiction. The instruments, the algorithms, and the commercial momentum are already in motion.

This is a companion article. It should be read alongside our primary article: "Mapping Prostate Cancer One Cell at a Time: A New Molecular Atlas Reveals Why Tumors Escape Treatment" (IPCSG Newsletter, March/April 2026), which covers the biological findings of the Song et al. atlas study in detail.

What Gleason Is — And What It Cannot See

In 1966, a pathologist named Donald Gleason at the Minneapolis VA Hospital developed a grading system for prostate cancer that remains in use today, sixty years later. His insight was elegant and practical: the more a tumor's glandular architecture had diverged from normal prostate tissue — the more chaotic and disorganized the cellular patterns looked — the more aggressive the cancer tended to be. A pathologist could look at a biopsy under a microscope, assess the architectural pattern on a scale from 1 to 5, add the scores from the two most prevalent patterns, and arrive at a Gleason score from 2 to 10 that predicted clinical outcomes with meaningful reliability.

For its era, this was a remarkable clinical tool. It was reproducible enough to guide treatment decisions, correlate with survival, and become the most important single piece of information in a prostate cancer diagnosis. Six decades of outcome data have been collected against it. Every prostate cancer guideline in the world references it. Nearly every prostate cancer patient reading this newsletter has had their disease characterized by it.

But Gleason's method has a fundamental and increasingly consequential limitation: it grades what cells look like, not what they are doing. And in 2026, we now know that what cells are doing — which genes they are actively expressing, which molecular programs they have activated, which dangerous transitions they have already begun — determines far more about how a tumor will behave than its architectural pattern alone.

1966

Year Gleason grading was introduced — still the standard today

94%

PCformer's accuracy classifying prostate cell types from gene expression data

560K

Individual cells PCformer was trained on from the UCSF atlas

Cell target for Illumina's Billion Cell Atlas program over three years

The Specific Things Gleason Misses

The UCSF atlas study examined 560,492 cells from 128 patients across all stages of prostate cancer, reading the genetic activity of each cell individually. What it found directly exposes the diagnostic blind spots of Gleason grading — not as a criticism of a flawed system, but as a precise scientific accounting of what sixty-year-old light microscopy simply cannot see.

Blind Spot 1: Luminal Infidelity

The atlas identified eight distinct gene modules — coordinated programs of gene activity — that define the normal luminal cell identity of prostate tissue. These modules contain the genes that encode PSA, AR, TMPRSS2, NKX3-1, and other proteins considered hallmarks of prostate identity. In healthy tissue, most luminal cells express six, seven, or all eight of these modules simultaneously.

In cancer, the atlas found something striking: different cells within the same tumor were losing these modules in different combinations. One cell might retain PSA and AR expression but lose NKX3-1. Another might lose PSA entirely while retaining AR. A third might express a basal-cell module alongside luminal markers, occupying a hybrid identity that has no clean name in traditional pathology. The atlas called this pattern "luminal infidelity" — and found that it varied more within Gleason grade groups than between them in some comparisons. In other words, two tumors with identical Gleason scores could have radically different molecular identities that Gleason cannot distinguish.

Blind Spot 2: The Invisible Transformation

Perhaps the most clinically consequential finding is that the study detected intermediate levels of neuroendocrine (NE) gene module activation in cells from tumors not clinically classified as neuroendocrine prostate cancer. Fifty-four localized prostate cancer samples had at least 25% of cells with intermediate levels of the novel HES6 neuroendocrine module — despite appearing as ordinary adenocarcinoma under the microscope.

This matters enormously. Neuroendocrine transformation — cancer cells shifting from prostate identity to a nerve-cell-like identity — is one of the most dangerous events in prostate cancer progression, conferring resistance to hormone therapy and poor prognosis. Current pathology can only detect it after it has become histologically apparent: when enough cells have completed the transformation to produce visible neuroendocrine features on a standard slide. The atlas shows that the molecular transformation begins far earlier, in cells that look completely normal to any pathologist.

"The molecular plasticity begins to occur before occult histological transformation — these intermediate states are detectable by gene expression analysis in cells that appear as ordinary adenocarcinoma under any conventional microscope."

— Song et al., bioRxiv, March 2026 (paraphrased)

Blind Spot 3: Heterogeneity Within a Single Gland

The atlas used the Xenium 5K spatial transcriptomics platform to map individual cells within intact tissue sections, overlaying their molecular profiles on standard H&E histological images. Within a single prostatic gland — a structure a pathologist would assign a single Gleason pattern — the team found cells in basal-like, club-like, and luminal-like states co-existing simultaneously. Gleason assigns one grade to one gland. The molecular reality is a neighborhood of cells in different states, some stable and some already transitioning.

Blind Spot 4: The Immune Microenvironment

Gleason grades epithelial cancer cells. It provides no information whatsoever about the immune and stromal environment surrounding those cells — which the atlas showed is itself highly prognostic. The atlas identified three distinct immune signaling axes in tumor cells that correlated with completely different immune microenvironments: one associated with an immune-supportive environment with proliferating T cells, one with an immunosuppressive environment dominated by regulatory T cells and reduced CD8+ cells, and one with a myeloid-dominant pattern. These distinctions have direct relevance to immunotherapy decisions — yet none of them is visible to Gleason grading.

What PCformer Changes

The UCSF atlas introduced PCformer — the Prostate Cancer Single-Cell Annotation Transformer — as a practical tool built to make the atlas's molecular classification system deployable and reproducible by any laboratory. Understanding what PCformer actually does, and what it represents architecturally, is essential for appreciating why it points toward a transformation of pathology rather than simply an incremental improvement.

PCformer was trained on the same transformer architecture that underlies modern large language models — the technology behind AI text tools. But instead of learning the statistical patterns of human language, PCformer learned the statistical patterns of prostate cell gene expression: which genes tend to be active together in a luminal cell, which combinations mark a basal cell, which patterns characterize a club cell, a Schwann cell, an exhausted T cell, a cancer-associated fibroblast.

The training data was the 560,492 cells in the atlas. Each cell was represented as a ranked list of up to 512 gene tokens — the genes most actively expressed in that cell, ordered from most to least active — fed into the transformer's attention mechanism, which learned contextual relationships between gene expression patterns exactly as a language model learns contextual relationships between words.

The result achieved 94% overall accuracy across 12 prostate cell type categories when tested on held-out data. When validated on a completely independent dataset from a different institution — the Wong et al. cribriform prostate cancer dataset — it achieved precision and recall above 96% for most major cell lineages. These are not research-quality approximations. They are performance figures that would be acceptable for clinical diagnostic tools.

Diagnostic Dimension Gleason Grading Today PCformer-Based Molecular Pathology
What it measures Glandular architecture — what cells look like Gene expression — what each cell is actively doing
Resolution Pattern averaged across thousands of cells simultaneously Individual cell — every cell characterized separately
NE transformation detection Only after histological transformation is complete and visible Intermediate molecular states detectable before histological change
Intra-tumoral heterogeneity Not captured — single grade per core or gland Fully characterized — distribution of all cell states per tumor
Immune microenvironment Not assessed Fully characterized — immune cell types, states, and signaling axes
Spatial information Architecture only — no molecular location data Cell-by-cell molecular map overlaid on tissue coordinates Reproducibility Significant interobserver variability — same slide, different grades from different pathologists
Algorithmic — same data produces identical result every time Treatment guidance Risk stratification only — does not identify therapeutic targets Identifies active molecular pathways, resistance mechanisms, and targetable programs
Ancestry sensitivity Not assessed — same grading criteria applied regardless of ancestry Atlas identified elevated Th17 inflammatory T-cells in African American patients — a biologically meaningful difference

The Decipher Precedent: This Transition Has Happened Before

The idea that molecular gene expression analysis could replace or augment architectural pathology in prostate cancer is not hypothetical — it has already happened once, at smaller scale. The Decipher Genomic Classifier is a 22-gene RNA expression test run on prostate biopsy or prostatectomy tissue that provides a risk score independent of Gleason grade. It went from research concept to inclusion in the National Comprehensive Cancer Network (NCCN) guidelines in roughly a decade and is now reimbursed by Medicare and many commercial insurers.

PCformer-based molecular pathology is orders of magnitude more complex than Decipher — 13,745 genes analyzed at single-cell resolution across all cell types versus 22 genes in bulk tissue from cancer cells only. But the infrastructure trajectory is instructive:

2010s — Decipher's Path 22-gene RNA test developed from bulk tissue research. Regulatory submission, clinical validation studies, NCCN guideline inclusion by approximately 2017. Medicare coverage followed.
2022 — Single-Cell Infrastructure Matures 10x Genomics Xenium platform commercialized — the same instrument used in the Song et al. atlas. Single-cell sequencing costs drop below $1,000 per sample at scale.
February 2025 — Illumina + Broad Clinical Labs Illumina announces collaboration with Broad Clinical Labs to scale single-cell sequencing and build a 5 billion cell atlas over three years, explicitly targeting clinical translation.
January 2026 — Illumina Billion Cell Atlas Illumina launches the Billion Cell Atlas with AstraZeneca, Merck, and Eli Lilly as founding partners — explicitly aimed at drug discovery AI and disease biology mapping at clinical scale.
March 2026 — PCformer Published Song et al. atlas introduces PCformer as an open-access prostate-specific transformer model achieving 94% cell type classification accuracy — a public research resource deployable by any laboratory.
2028–2032 — Projected Bridge Period Expected condensed gene signatures derived from atlas modules (analogous to Decipher's condensation of transcriptomic data) enter clinical validation studies. First FFPE-compatible single-cell diagnostic assays likely submitted for FDA review.
2030s — Projected Clinical Integration Molecular cell-state profiling likely enters NCCN guidelines as adjunct to Gleason for high-risk and metastatic disease. Full replacement of Gleason for localized disease may take longer given cost and infrastructure requirements.

The Industry Landscape: Who Is Building This Future

The commercial infrastructure for molecular cell-state pathology is being assembled right now, and the players involved suggest this transition will be driven as much by market forces as by academic research. Several major firms have made commitments specifically relevant to what the UCSF atlas represents.

Key Industry Players and Their Prostate Cancer / Single-Cell Stakes

Illumina (ILMN) The dominant force in DNA and RNA sequencing globally. In January 2026, Illumina launched the Billion Cell Atlas — explicitly targeting 5 billion cells over three years with AstraZeneca, Merck, and Eli Lilly as founding partners. In February 2025, Illumina partnered with Broad Clinical Labs (the clinical arm of the Broad Institute of MIT and Harvard, where PCformer's co-investigator Dr. Joshua Campbell is based) specifically to scale single-cell sequencing toward clinical deployment. Illumina's PIPseq V single-cell RNA-seq platform and NovaSeq X sequencer are the commercial infrastructure on which atlas-scale studies run. The connection to the Campbell lab at Boston University / Broad is direct and structural.
10x Genomics Manufacturer of the Xenium 5K Prime spatial transcriptomics platform used directly in the Song et al. atlas study. 10x Genomics sells both the scRNA-seq (Chromium) and spatial (Xenium) platforms that produced the atlas data. Clinical adoption of Xenium for diagnostic spatial transcriptomics would be a direct commercial application of the atlas methodology. The company has been actively promoting Xenium for FFPE tissue — the standard clinical specimen format.
Foundation Medicine (Roche) The leading clinical genomic profiling company for oncology. FoundationOne CDx is already standard of care for metastatic prostate cancer genomic profiling. The natural next commercial step — as single-cell and spatial data matures — is expanding from mutation-level profiling to cell-state-level profiling. Foundation Medicine's existing relationships with oncologists and reimbursement infrastructure represent the most likely commercial pathway for single-cell molecular pathology to reach patients.
Tempus AI Clinical AI and genomics company with existing prostate cancer RNA sequencing products. Tempus has the clinical data infrastructure, oncologist relationships, and AI expertise to integrate a PCformer-like cell-state classification model into its existing diagnostic portfolio.
NeoGenomics Specialized cancer diagnostics company offering comprehensive genomic testing. Already provides whole transcriptome sequencing and spatial biology through its proprietary MultiOmyx platform. A natural commercial partner for translating atlas-derived signatures into standardized diagnostic products.
Veracyte (Decipher) Owner of the Decipher prostate cancer genomic classifier — the most direct commercial precedent for PCformer-based diagnostics. Veracyte has already navigated the regulatory and reimbursement pathway for RNA-based prostate cancer testing once. Expanding Decipher from 22 bulk-tissue genes to a PCformer-derived single-cell signature is a commercially logical next step that builds on existing infrastructure.
Pharma Partners (AZ, Merck, Lilly) The three founding partners of Illumina's Billion Cell Atlas are among the largest prostate cancer drug developers in the world. Their investment in single-cell atlas infrastructure is explicitly aimed at drug target validation and patient stratification — the direct commercial application of what PCformer enables. Drug trials that stratify patients by cell-state module profile rather than by Gleason grade alone are a predictable near-term output of this investment.

What a PCformer-Era Pathology Report Would Look Like

For patients, the most important question is what this actually means for the experience of diagnosis and treatment. A useful way to think about it is to compare what a pathology report looks like today with what it could look like in a decade.

Today's report tells you: the biopsy from core 3 of 12 shows Gleason 4+3=7 (Grade Group 3) prostate adenocarcinoma involving 60% of the core length. Perineural invasion present. No extraprostatic extension identified.

This is genuinely important information. But it tells you nothing about whether the tumor cells have already begun expressing intermediate neuroendocrine gene programs. It tells you nothing about whether RB1 and TP53 are being silenced, which would predict lineage plasticity risk. It tells you nothing about whether the immune microenvironment is suppressive or active, which affects immunotherapy decisions. It gives you one grade for a tumor that may contain cells in five or six different molecular states simultaneously.

A future PCformer-era report might read instead: 68% of tumor cells are luminal-like with an average of 5.2 of 8 luminal modules expressed (moderate infidelity score, associated with Grade Group 3 disease). 12% of tumor cells are basal-like (TP63+) with elevated EMT and inflammatory signatures. 8% show intermediate neuroendocrine module activation (HES6/SCG intermediate — precursor state). 12% are in active cell cycle (Ki67+). Immune microenvironment: Axis 2 dominant (immunosuppressive — elevated Treg and CD1C+ cells, reduced CD8+ T cells). No high-confidence NE-transformed cells detected. Spatial distribution: basal-like cells concentrated in Gleason 3 glandular areas; intermediate NE cells distributed throughout tumor margin.

This report answers different questions — and in many cases, more clinically important ones. It identifies the 8% of cells that may already be on the path to treatment resistance. It characterizes the immune environment that will determine immunotherapy eligibility. It gives the oncologist not a single number but a cell-state landscape that can be compared to future biopsies to track molecular evolution in real time.

What This Means for Active Surveillance Patients

One of the most direct near-term applications of molecular cell-state profiling is in active surveillance — the management strategy for low-risk prostate cancer that avoids immediate treatment. Currently, surveillance decisions are made primarily on the basis of PSA kinetics, repeat biopsy Gleason grades, and MRI findings. The atlas finding that many cells in seemingly low-grade tumors already show intermediate molecular changes — lineage infidelity, intermediate NE module activation — suggests that molecular profiling could identify the subset of surveillance patients whose tumors are already in molecular transition, even before any change in Gleason grade or PSA. This could allow earlier, more targeted intervention for exactly the patients who need it, while sparing the majority of stable patients from unnecessary treatment.

What This Means for Advanced Disease Patients

For men with metastatic or castration-resistant prostate cancer — the population IPCSG serves most directly — molecular cell-state profiling addresses a specific and urgent unmet need: knowing when neuroendocrine transformation is beginning, before it becomes clinically apparent. The atlas found that intermediate NE module expression was detectable in lymph node and other metastatic samples not clinically classified as NEPC. A biopsy of a progressing lesion analyzed with PCformer-based profiling could detect this transition early enough to adjust treatment — adding platinum-based chemotherapy, reconsidering PSMA-targeted therapy sequencing, or enrolling in trials targeting the NE transformation mechanism — rather than discovering it only after clinical deterioration makes the transition undeniable.

The Honest Obstacles

Enthusiasm for this technology must be tempered by honesty about what stands between the current research state and widespread clinical deployment. These are real obstacles, not rhetorical hedges.

Cost: Single-cell RNA sequencing currently costs several hundred to several thousand dollars per sample depending on scale and platform. Standard pathology is a fraction of this. The economics must shift further — and they are shifting, but not yet to the levels required for population-scale screening. Spatial transcriptomics remains even more expensive. The Illumina Billion Cell Atlas program and the broader commercial infrastructure being built suggests costs will continue to decline, but the timeline for reaching clinical affordability is measured in years, not months.

Fresh tissue requirement: Live single-cell RNA sequencing requires fresh tissue processed within hours of collection — which is incompatible with the standard formalin fixation that happens to essentially all clinical biopsy specimens. Spatial transcriptomics using the Xenium platform can work with FFPE tissue, as demonstrated in the atlas itself, but with reduced resolution. Expanding clinical compatibility with standard specimen processing remains a critical technical challenge. Some promising approaches using FFPE-compatible single-nucleus RNA sequencing are under active development.

Regulatory pathway: A new diagnostic modality requires FDA clearance and CMS reimbursement before it becomes standard of care. That process typically takes five to ten years from initial clinical evidence submission, with no guarantee of outcome. Gleason grading benefits from six decades of outcome data. PCformer-based profiling will require prospective clinical validation demonstrating that its outputs improve patient outcomes — not just that they are scientifically interesting — before regulators and payers will adopt it.

Interobserver variability solved, interpretive complexity introduced: One of Gleason's recognized limitations is interobserver variability — different pathologists assign different grades to the same slide. PCformer eliminates this: the same data produces the same result algorithmically. But it introduces a new challenge: the output is vastly more complex than a single number from 6 to 10. Training oncologists and pathologists to interpret a multi-dimensional cell-state profile, and embedding that interpretation into clinical decision support systems, is a non-trivial workflow redesign challenge.

The Near-Term Bridge: Condensed Signatures

The most realistic near-term path from the atlas to clinical practice is not deploying full single-cell sequencing on every biopsy — it is extracting condensed signatures from the atlas's findings that can be run with existing clinical tools.

The atlas itself demonstrated this approach. From the 166 gene modules and 13,745 genes in the full PCformer system, the team derived an 18-gene neuroendocrine signature that outperformed all previously published NE signatures in distinguishing NEPC from non-NEPC samples. This 18-gene signature can be run on bulk RNA from standard FFPE tissue — a workflow already established by tests like Decipher. It requires no single-cell infrastructure, no special tissue handling, and is compatible with existing clinical laboratory workflows.

This condensation strategy — extracting the highest-information genes from a comprehensive single-cell atlas and deploying them as a clinically practical test — is exactly how Decipher was built from complex transcriptomic research data. It is the bridge between today's research tools and tomorrow's clinical practice. The atlas has now provided the comprehensive map from which multiple such condensed signatures can be derived — for NE transformation risk, for luminal infidelity score, for immune microenvironment classification, for metastatic site-specific gene program activation.

For IPCSG Members: What to Watch For

The transformation of pathology described in this article will unfold over years, not months. But there are specific developments worth watching that will signal how quickly it is progressing:

The 18-gene NE signature from the atlas entering clinical validation studies is the most imminent development to watch. If a commercial laboratory (most likely Veracyte, Tempus, or Foundation Medicine) adopts and validates this signature as a clinical add-on to standard biopsy processing, it could be available within two to three years. Ask your oncologist about it specifically.

FFPE-compatible spatial transcriptomics entering routine use at NCI-designated cancer centers is the next milestone. UCSF, UCSD, MD Anderson, Memorial Sloan Kettering, and other major centers are building this infrastructure now. Patients treated at these centers may have access to spatial molecular profiling within their standard of care or on clinical trials in the next three to five years.

The Illumina Billion Cell Atlas and its pharmaceutical partners producing disease-specific cell-state maps across cancer types — including prostate cancer at metastatic stages — will create the training data needed for PCformer-like models to be extended to other cancers and other disease contexts. Watch for announcements from this consortium about prostate-cancer-specific datasets and companion diagnostic programs.

Clinical trials that stratify patients by cell-state molecular profile — rather than by Gleason alone or PSA kinetics alone — will be the first point at which this science intersects directly with patient access to advanced therapies. IPCSG will monitor trial registrations for this kind of molecular stratification criterion and report when such trials open.

Verified Sources & Formal Citations

PRIMARY ATLAS PAPER (PREPRINT):
Song H, Xu J, Velazquez-Arcelay K, et al. A single-cell and spatial atlas of prostate cancer reveals the combinatorial nature of gene modules underlying lineage plasticity and metastasis. bioRxiv. 2026 Mar 27. doi: 10.64898/2026.03.25.711335
ILLUMINA BILLION CELL ATLAS LAUNCH:
Illumina, Inc. Press Release. Illumina introduces Billion Cell Atlas to accelerate AI and drug discovery. January 13, 2026.
URL: https://www.illumina.com/company/news-center/press-releases/
ILLUMINA + BROAD CLINICAL LABS COLLABORATION:
Illumina, Inc. Press Release. Illumina and Broad Clinical Labs usher in new era of drug discovery with collaboration to rapidly scale single-cell solutions. February 21, 2025.
URL: https://investor.illumina.com/news/press-release-details/2025/
ILLUMINA MULTIOMICS PORTFOLIO (AGBT 2025):
Illumina, Inc. Press Release. Illumina transforms multiomic research with new technologies. February 24, 2025.
URL: https://investor.illumina.com/news/
ILLUMINA PIPseq V SINGLE-CELL RNA-SEQ:
Illumina. Single-cell RNA sequencing with PIPseq chemistry. 2026.
URL: https://www.illumina.com/techniques/sequencing/rna-sequencing/
SINGLE-CELL SEQUENCING IN CLINICAL DIAGNOSTICS — REVIEW:
Elshiekh A et al. Single-cell sequencing in molecular diagnostics: Transformative yet untapped potential. Frontiers in Genetics. 2025;16:1621081. doi: 10.3389/fgene.2025.1621081
URL: https://pmc.ncbi.nlm.nih.gov/articles/PMC12671893/
AI FOR GLEASON GRADING — EXPLAINABLE AI (Nature Communications, 2025):
Pathologist-like explainable AI for interpretable Gleason grading in prostate cancer. Nature Communications. 2025;16:8959. doi: 10.1038/s41467-025-64712-4
URL: https://www.nature.com/articles/s41467-025-64712-4
ADVANCES IN AI IN PROSTATE CANCER PATHOLOGY — UCSF (2026):
Advances in artificial intelligence in prostate cancer pathology. Seminars in Diagnostic Pathology. 2026 Mar;43(2):150995. doi: 10.1016/j.semdp.2026.150995
PubMed: https://pubmed.ncbi.nlm.nih.gov/41807227/
PANDA CHALLENGE — AI GLEASON GRADING (Nature Medicine, 2022):
Artificial intelligence for diagnosis and Gleason grading of prostate cancer: the PANDA challenge. Nature Medicine. 2022. doi: 10.1038/s41591-021-01620-2
URL: https://www.nature.com/articles/s41591-021-01620-2
PROSTATE CANCER CELL ATLAS (PCCAT) — RELATED SINGLE-CELL ATLAS:
Deciphering single-cell heterogeneity and cellular ecosystem dynamics during prostate cancer progression. bioRxiv. December 2024. doi: 10.1101/2024.12.18.629070
URL: https://www.biorxiv.org/content/10.1101/2024.12.18.629070v1
GARVAN INSTITUTE EARLY-STAGE ATLAS (Cancer Research, 2026):
Apostolov E et al. Single-Cell and Spatial Transcriptomic Profiling Reveals Epithelial Functional States and Fibroblast Phenotypes in Hormone Therapy-Naïve Localized Prostate Cancer. Cancer Research. 2026:OF1–OF18. doi: 10.1158/0008-5472.CAN-25-1202
LINEAGE PLASTICITY JAK/STAT (Science, 2022):
Chan JM et al. Lineage plasticity in prostate cancer depends on JAK/STAT inflammatory signaling. Science. 2022;377:1180–1191. doi: 10.1126/science.abn0478
FRANKLIN HUANG LAB — UCSF (Official Profile):
Franklin W. Huang, MD, PhD. UCSF Helen Diller Family Comprehensive Cancer Center.
URL: https://cancer.ucsf.edu/people/huang.franklin

All URLs verified as of March 30, 2026. The primary atlas paper (Source 1) is a preprint and has not yet undergone formal peer review. Commercial and industry analyses represent the editorial assessment of IPCSG research staff based on publicly available information and do not constitute investment advice. IPCSG has no financial relationships with any companies mentioned in this article.

Search This Blog

Informed Prostate Cancer Patient Support

Pathology Beyond Gleason? What are the cells doing?