
An project lead by @aromicsbiotech, focused on developing a novel Berberine derivative to treat Mesothelioma, an asbestos-related cancer.🔬💊
Barcelona, España
Joined January 2019
- Tweets5.4K
- Following927
- Followers617
- Likes34.8K
Pinned Tweet
La exposición al #amianto es el principal factor de riesgo para el #mesothelioma. El objetivo principal @bermesproject es el desarrollo una nueva molécula con un mecanismo de acción disruptivo y que ha mostrado actividad farmacológica en este tumor difícil de tratar. #asbestos
Bermes Project retweeted
This was the equivalent of extending human lifespan by 9 years
science.org/content/article/… @ScienceMagazine
Bone as a modulator of aging.
Dynamic force on the tibia is sensed and extends healthspan and lifespan, with reduced inflammation, and bone marrow immunosenescence in the experimental model. When Piezo1 (pressure sensor) receptor was deleted, resulting in lack of sensing, the opposite occurred @Cell_Metabolism
sciencedirect.com/science/ar…
Bermes Project retweeted
Malcolm Gladwell literally gave a 1-hour masterclass on why some people succeed and others don't and it will completely change how you think about talent, effort, and success.
Bermes Project retweeted
🧬🔬 What if correcting batch effects in single-cell RNA sequencing actually removes the biological differences we are trying to discover?
A new study in Science Advances introduces CellDiffusion, an unsupervised graph neural diffusion framework designed to integrate scRNA-seq datasets while preserving genuine cellular heterogeneity.
The central innovation: Better integration does not mean mixing all cells together. It means aligning biologically equivalent cells while protecting distinct cell states and developmental trajectories.
🔬 1. The hidden problem in single-cell integration
Combining scRNA-seq datasets across laboratories, patients, and experimental conditions introduces technical batch effects.
Conventional approaches can produce two major errors:
• Underalignment: Equivalent cell types remain separated by batch.
• Misalignment: Biologically distinct populations are incorrectly merged through overcorrection.
These problems become particularly severe when disease and control samples contain different cell populations.
⚡ 2. How CellDiffusion works
CellDiffusion integrates three computational strategies:
① Optimized mutual nearest neighbors (OMNN) establish candidate connections between cells across batches.
② Graph neural diffusion propagates transcriptional features between connected cells.
③ Trainable attention mechanisms strengthen biologically meaningful associations while suppressing misleading connections.
A graph-reconstruction objective preserves the original within-batch cellular structure.
The result is a batch-corrected embedding that maintains biological identity.
🧬 3. Preserving dynamic human endometrial cell states
Researchers evaluated CellDiffusion using human endometrial scRNA-seq datasets spanning proliferative and secretory menstrual phases.
These tissues undergo extensive physiological remodeling, creating major differences in cellular composition.
Conventional Harmony integration incorrectly merged distinct stromal populations.
In contrast, CellDiffusion preserved:
eStromal cycling → eStromal → Early decidual stromal → Mid/late decidual stromal states
It also retained biologically meaningful developmental relationships.
Across the evaluated metrics, CellDiffusion outperformed Harmony and CellANOVA in this dataset.
🧠 4. Detecting subtle immune-cell heterogeneity
In human bone marrow mononuclear cells, CellDiffusion distinguished closely related immune populations that competing methods frequently merged.
Examples included:
• IGKC⁺ versus IGKC⁻ B-cell populations
• Plasma-cell and plasmablast subpopulations
• Distinct monocyte and T-cell states
These findings demonstrate the importance of preserving subtle transcriptional differences during integration.
🔍 5. Improved annotation when reference atlases are incomplete
Many annotation algorithms assume that every query cell type exists in the reference dataset.
CellDiffusion addresses this limitation by identifying cells without reliable reference counterparts.
In benchmark experiments, annotation accuracy declined by only 1–14% when transitioning from complete to incomplete references.
Competing approaches showed reductions of 16–62%.
This could be particularly valuable when analyzing rare disease-associated populations or previously uncharacterized cellular states.
🌱 6. Better clustering and trajectory reconstruction
The diffusion framework captures two complementary properties:
Local equilibrium: Closely related cells become more tightly grouped.
Global equilibrium: Relationships between biologically connected populations are preserved.
Across Tabula Muris, cortical, and PBMC datasets, CellDiffusion improved unsupervised clustering performance.
In planarian regeneration datasets, its trajectory inference also better recovered established differentiation relationships than benchmarked alternatives.
💡 7. Implications for disease research
CellDiffusion could be particularly useful for:
• Cancer microenvironment analysis
• Aging-associated cellular remodeling
• Adipose tissue plasticity
• Immune-cell differentiation
• Regenerative medicine
• Cross-cohort single-cell atlas integration
For example, when comparing young and aged tissues, conventional integration may inadvertently eliminate age-specific cellular states.
Preserving those differences may be essential for identifying disease mechanisms and therapeutic targets.
⚠️ Important limitations
CellDiffusion assumes that batches share some common cell populations and that technical variation does not overwhelm biological differences.
It is computationally more demanding than Harmony, and scaling to millions of cells remains challenging.
Furthermore, improved computational integration does not independently establish the biological function of newly resolved populations.
The broader implication
Single-cell integration should not maximize batch mixing at the expense of biological fidelity.
The objective is to distinguish technical artifacts from meaningful differences in cell identity, state, and lineage.
The next generation of single-cell analysis may shift from aggressive batch correction toward biologically informed reconstruction of cellular relationships.
📄 Liu Y-C, Huang Z, Lu SL, et al.
CellDiffusion: Enhancing scRNA-seq data integration with accurate inference of cell-cell association.
Science Advances. 2026;12.
DOI: doi.org/10.1126/sciadv.aed41…
💻 Open-source implementation:
github.com/CZCBLab/CellDiffu…
#SingleCellRNAseq #CellDiffusion #Bioinformatics #GraphNeuralNetworks #Transcriptomics #BatchCorrection #CellAtlas #CancerResearch #Aging #AdiposePlasticity #ComputationalBiology
Bermes Project retweeted
Integrating RNA-seq, metabolomics & proteomics, @GeraldineMaier mapped how exercise rewires metabolism across tissues:
• Glycolysis
• Fat oxidation
• Ketone use
• BCAA metabolism
@WuTsaiAlliance
cell.com/cell/abstract/S0092…
Bermes Project retweeted
🧬🔥 What if preventing skin inflammation requires more than blocking TNF-induced cell death—and instead depends on maintaining Wnt/β-catenin signaling?
A new study in Science Advances identifies an unexpected function of OTULIN, a linear ubiquitin-specific deubiquitinase, in protecting epithelial cells from inflammatory injury.
The central mechanism:
OTULIN → β-catenin stabilization → Wnt/TCF3/4 signaling → Keratinocyte survival → Skin inflammation ↓
🔬 1. OTULIN connects ubiquitination to epithelial survival
OTULIN selectively removes M1-linked (linear) ubiquitin chains and regulates LUBAC-dependent inflammatory signaling.
Loss-of-function mutations cause OTULIN-related autoinflammatory syndrome (ORAS), characterized by severe systemic inflammation and skin involvement.
In keratinocyte-specific OTULIN-knockout mice, investigators observed:
• Severe dermatitis
• Increased keratinocyte apoptosis and necroptosis
• Epidermal barrier dysfunction
• Progression toward squamous tumors
Unexpectedly, single-cell RNA-seq revealed profound suppression of canonical Wnt signaling.
⚡ 2. OTULIN directly regulates β-catenin stability
Mechanistic experiments demonstrated that β-catenin undergoes linear ubiquitination in keratinocytes.
Without OTULIN:
M1-linked ubiquitination dysregulation
↓
K48-linked β-catenin ubiquitination ↑
↓
Proteasomal degradation ↑
↓
β-catenin stability ↓
↓
Canonical Wnt signaling ↓
Coimmunoprecipitation and recombinant OTULIN experiments supported its interaction with β-catenin and regulation of linear ubiquitination.
🧪 3. β-catenin stabilization prevents inflammatory skin disease
Genetic stabilization of β-catenin in OTULIN-deficient mice produced remarkable protection:
• Complete prevention of tail dermatitis
• Delayed or prevented back-skin inflammation
• Reduced circulating IL-6
• Improved epidermal barrier integrity
• Protection against exophytic tumor formation
Importantly, activating β-catenin after inflammatory lesions had developed also attenuated disease progression.
Wnt restoration therefore showed both preventive and therapeutic effects in this genetic mouse model.
🧠 4. TCF3/4 emerge as downstream survival regulators
OTULIN deficiency reduced expression of TCF3 and TCF4, transcription factors involved in epithelial stem-cell maintenance.
Restoring TCF3 in OTULIN-deficient human keratinocytes significantly reduced TNF-induced cytotoxicity, although protection was incomplete.
WNT3A pretreatment similarly improved cellular survival.
These findings suggest that canonical Wnt signaling determines the threshold for TNF-induced keratinocyte death.
🔗 5. Potential implications beyond skin inflammation
OTULIN deficiency also reduced Tcf4 expression in mouse intestinal epithelial cells.
Analysis of human colorectal cancer single-cell datasets revealed increased OTULIN, TCF3, and TCF4 expression in tumor epithelial populations.
However, these cancer observations are associative and do not establish OTULIN as a therapeutic cancer target.
💡 The broader implication
OTULIN protects epithelial homeostasis through two complementary mechanisms:
① LUBAC-dependent regulation of inflammatory cell-death signaling
② β-catenin stabilization and Wnt-dependent survival transcription
This reveals an important connection between ubiquitin signaling, epithelial stem-cell biology, and inflammatory disease.
⚠️ Therapeutic translation remains preclinical. Because sustained Wnt activation can promote tumorigenesis in other contexts, systemic β-catenin activation would require careful safety evaluation.
The next generation of anti-inflammatory therapies may need to preserve epithelial survival programs—not simply suppress inflammatory cytokines.
📄 Lecomte K, Hertens P, Wylleman V, et al.
Canonical Wnt induction by OTULIN prevents keratinocyte death and skin inflammation.
Science Advances. 2026;12(41).
DOI: doi.org/10.1126/sciadv.adz01…
#OTULIN #WntSignaling #BetaCatenin #Ubiquitination #Inflammation #CellDeath #SkinBiology #Autoinflammation #DrugDiscovery
Bermes Project retweeted
🧬🔥 What if pancreatic cancer could be treated BEFORE it becomes cancer?
A remarkable 2026 Science study provides preclinical proof-of-concept for:
PHARMACOLOGICAL CANCER INTERCEPTION
Instead of waiting for pancreatic ductal adenocarcinoma (PDAC) to emerge, Than et al. targeted its dominant oncogenic driver—KRAS—while disease was still confined to premalignant PanIN lesions.
The result:
KRAS inhibition → PanIN regression → delayed PDAC → dramatically prolonged survival. science.aec7929
🧬 Why PDAC may be uniquely suited to interception
Most PDAC develops through:
normal pancreas
→ ADM
→ PanIN
→ high-grade PanIN
→ invasive PDAC
And remarkably:
>92% of PanIN lesions already carry KRAS mutations.
KRAS activation therefore occurs long before invasive cancer develops. science.aec7929
The investigators asked a provocative question:
Why wait until KRAS-mutant cells have accumulated additional mutations, plasticity and resistance mechanisms?
ATTACK KRAS EARLY.
🐭 First: can KRAS inhibition eliminate precancerous lesions?
Using autochthonous KPC mice, the investigators treated tumor-free animals harboring PanINs with:
RMC-7977
→ multiselective RAS(ON) inhibitor
or
RMC-9945
→ KRAS^G12D-selective RAS(ON) inhibitor.
After only:
10 DAILY DOSES
both dramatically reduced pancreatic preneoplastic burden. science.aec7929
The 3D CODA reconstructions in Figure 1, page 2 are particularly striking: untreated pancreata contain extensive ADM/PanIN lesions, whereas RAS(ON)-treated pancreata show profound disappearance of preneoplasia. science.aec7929
💥 How do PanINs disappear?
RAS inhibition caused:
pERK ↓
↓
MAPK signaling suppressed
and:
cleaved caspase-3 ↑
↓
APOPTOSIS.
Importantly, amylase staining did not support substantial restoration of PanIN cells toward an acinar phenotype.
So the major mechanism was:
KRAS OFF
→ PanIN survival program collapses
→ NEOPLASTIC CELL DEATH
rather than simply redifferentiation. science.aec7929
This suggests that even very early pancreatic neoplasia is already:
KRAS-ADDICTED.
🧪 What happens if treatment is stopped?
The investigators treated 7-week-old tumor-free KPC mice for:
28 DAYS
and then discontinued therapy.
RMC-7977 shifted median tumor onset:
94 → 136 days
and median survival:
135 → 159 days.
But KRAS^G12D-selective RMC-9945 was even more impressive:
tumor onset:
83 → 162 days
overall survival:
120 → 200 days.
science.aec7929
The Kaplan-Meier curves in Figure 2, page 3 show a striking rightward shift after just four weeks of premalignant treatment. science.aec7929
And importantly:
INTERCEPTION DID NOT CREATE MORE AGGRESSIVE TUMORS.
Tumors that eventually emerged were similar in size and subsequent growth rate to tumors in untreated animals. science.aec7929
🤯 Then came the decisive experiment: LONG-TERM INTERCEPTION
Because new PanINs continuously arise in KPC mice, the authors tested a metronomic schedule:
RMC-7977
1 week ON
1 week OFF
repeated long-term
starting while mice were still tumor-free. science.aec7929
The result was extraordinary.
MEDIAN TUMOR-FREE SURVIVAL
Control:
105 days
Interception:
329 days
→ approximately 3× longer
And:
MEDIAN PDAC-SPECIFIC OVERALL SURVIVAL
Control:
138 days
Interception:
376 DAYS
P < 0.0001. science.aec7929
The survival curves in Figure 3, page 5 are the centerpiece of the paper: the intercepted animals remain tumor-free for months while virtually all control animals rapidly develop lethal PDAC. science.aec7929
🔥 And early treatment was BETTER than treating established cancer
The authors then used the same metronomic RAS(ON) regimen but initiated therapy:
AFTER PDAC DIAGNOSIS.
It still improved survival.
But:
PREMALIGNANT INTERCEPTION PROVIDED GREATER SURVIVAL BENEFIT.
science.aec7929
That is perhaps the most important conceptual finding.
The same oncogene.
The same pharmacological strategy.
But:
TIMING CHANGES THE OUTCOME.
Why?
A PanIN is biologically simpler than an advanced cancer.
It has had less time to acquire:
genomic instability
intratumoral heterogeneity
alternative signaling
stromal adaptation
therapeutic resistance.
The authors propose that early lesions may simply have fewer evolutionary options when KRAS signaling disappears. science.aec7929
🎯 Remarkably, complete PanIN eradication was NOT necessary
Long-term control pancreata contained:
7.66% preneoplastic tissue
versus only:
0.64%
after RMC-7977 interception. science.aec7929
But some lesions remained.
Yet PDAC onset was dramatically delayed.
Therefore:
CANCER INTERCEPTION ≠ NECESSARILY ERADICATING EVERY PREMALIGNANT CELL.
Reducing the pool of lesions capable of evolutionary progression may be enough to dramatically shift cancer latency.
This is a powerful concept.
🧬 What about resistance?
A legitimate concern is:
If we expose premalignant cells to KRAS inhibition for months, will we simply pre-select highly resistant PDAC?
Surprisingly:
NO CLEAR EVIDENCE OF THAT OCCURRED.
Escape tumors remained responsive to RAS inhibition.
Whole-genome sequencing found no reproducible resistance-associated genetic pattern; KRAS/BRAF copy-number gains were not selectively enriched, and no consistent Kras or Cypa mutations explained escape. science.aec7929
The authors therefore speculate that escape may involve:
CELLULAR PLASTICITY
or
MICROENVIRONMENTAL ADAPTATION
rather than fixed genetic resistance.
🧠 The pre-tumor microenvironment changes too
PanINs already possess a stromal ecosystem containing:
fibroblasts
macrophages
T cells.
RAS inhibition caused a modest early reduction in:
F4/80⁺ macrophages
and reduced immune infiltration around residual PanINs in 3D.
However, depleting CD4⁺ and CD8⁺ T cells did not prevent acute PanIN regression. science.aec7929
Thus the immediate effect appears predominantly:
KRAS-dependent epithelial killing
rather than T-cell-mediated immune clearance.
But the rapid collapse of the PanIN-associated stromal niche suggests another interesting principle:
KRAS SIGNALING MAY MAINTAIN THE PRE-TME BEFORE CANCER EVEN EXISTS. science.aec7929
👥 Who could actually receive this strategy?
Here lies the enormous translational challenge.
PanINs are:
MICROSCOPIC
and
COMMON IN ADULTS.
We currently cannot reliably identify which individual PanIN will progress.
Therefore population-wide KRAS chemoprevention is unrealistic.
Instead, the authors envision:
HIGH-RISK PDAC POPULATIONS
such as individuals with:
BRCA1 mutations
BRCA2 mutations
PALB2 mutations
hereditary pancreatitis
other strong inherited/familial risk factors. science.aec7929
In these populations, treatment eligibility might be based on:
PREDICTED CANCER RISK
rather than direct visualization of every microscopic PanIN.
💊 And the drugs are moving closer to humans
The experimental compounds themselves are preclinical, but related RAS(ON) inhibitors are already being developed for RAS-mutant cancers.
The authors highlight:
DARAXONRASIB
→ close chemical relative of RMC-7977
and:
ZOLDONRASIB
→ close relative of RMC-9945. science.aec7929
Interestingly, because human PanINs can harbor multiple KRAS alleles:
MULTISELECTIVE RAS(ON) INHIBITION
may ultimately be more useful for interception than a single allele-specific inhibitor.
⚠️ But the bar for PREVENTION is fundamentally different from cancer treatment
This cannot be overstated.
Patients with advanced PDAC may accept substantial drug toxicity.
A healthy BRCA2 carrier who might never develop PDAC will not.
So interception requires:
extremely favorable safety
long-term tolerability
minimal cumulative toxicity
accurate risk stratification
optimal intermittent dosing
biomarkers of target engagement.
The authors explicitly emphasize that safety becomes critical because these drugs would be given to people without a cancer diagnosis. science.aec7929
💡 The deeper conceptual shift
Traditional oncology:
CANCER
↓
diagnosis
↓
treatment.
Cancer interception:
GENETIC RISK
↓
premalignant clone
↓
oncogenic dependency
↓
TARGETED THERAPY
↓
precursor burden ↓
↓
malignant evolution delayed
↓
CANCER PREVENTED / POSTPONED
And PDAC may be particularly attractive because:
KRAS IS ALREADY PRESENT AT THE BEGINNING.
Rather than waiting until:
KRAS
TP53
genomic instability
stromal remodeling
immune suppression
metastasis
have assembled into one of the most treatment-resistant cancers in medicine, perhaps we should attack:
KRAS ALONE—WHILE THE LESION IS STILL VULNERABLE.
This paper provides compelling preclinical evidence that:
WHEN WE TARGET AN ONCOGENE MAY BE AS IMPORTANT AS WHICH ONCOGENE WE TARGET.
📄 Than MT, Dequiedt L, Sor R, et al.
Cancer interception with KRAS inhibitors in preclinical models of pancreatic ductal adenocarcinoma.
Science. 2026;391. science.aec7929
DOI: 10.1126/science.aec7929 science.aec7929
#PancreaticCancer #PDAC #KRAS #CancerInterception #CancerPrevention #PanIN #RAS #PrecisionOncology #OncogeneAddiction #CancerBiology #Chemoprevention
Immune Checkpoint VISTA Doubles as a Growth Engine Inside Mesothelioma Cells bioengineer.org/immune-check…
20
Bermes Project retweeted
Sam Altman (CEO of OpenAI):
"You no longer need to write prompts."
In just 38 minutes, he explains how to use ChatGPT at a level that most people can't even imagine.
It's a talk he gave to Stanford students. A friend sent me the recording last night.
After watching it, I realized I was only taking advantage of about 15% of what this tool can really do.
Watch it in full and then read the guide I leave below on how to create a system that prompts itself.
Bermes Project retweeted
How do cells die? What used to be thought were 2 ways, now it's >20
nature.com/articles/d41586-0…
New mesothelioma drug turns cancer’s own defenses against it sciencedaily.com/releases/20…
34
Scientists find an immune “false alarm” that may drive rapid aging sciencedaily.com/releases/20…
16
Bermes Project retweeted
Strongest Nature paper for us who love #cancer #immunology
Tumors use nerves to tell the brain to suppress immune response to all the cancer to grow.
nature.com/articles/s41586-0…
Asbestos, 9/11 and the World Trade Center asbestos.com/world-trade-cen…
23
Bermes Project retweeted
Sarcopenia, the age-related loss of muscle mass and strength, carries an increased risk of morbidity and mortality. It can, in part, be prevented by lifestyle factors, especially exercise. But there's no approved treatment. Here's a new optimistic review that this is about to change @NatRevDrugDisc
nature.com/articles/s41573-0…
Bermes Project retweeted
The first drug in development (rentosertib, anti-fibrotic, designed with AI) to show slowing of proteomic aging clocks, 6 of 6, in a small trial @gladyshev_lab
@NatureBiotech @biogerontology @InSilicoMeds
nature.com/articles/s41587-0…
Bermes Project retweeted
for the next 18-24 mo., the biggest bottleneck in AI drug discovery is experimental validation and verification
however it is increasingly clear that in 36 mo., the biggest bottleneck in AI drug discovery will be clinical trial scarcity
the challenge with US clinical trial infrastructure goes far beyond simply regulatory reform
in fact, outside of Australia, US FDA is the fastest in the world in getting INDs reviewed (30 day response time vs 60 in UK and Western Europe, and 60 working days in China)
Yet, median time to site activation after trial regulatory approval for oncology trials is:
US (academic oncology site) - 171 days (~6 mo)
Australia Phase 1 - 63 days (~1.5 mo)
China - as quick as 4 weeks
It gets worse. Based on July 2026 analysis, US sites not only take median 124 days to get sites activated, it takes *another* 93 days to get FPI (first patient in).
Australia sites can usually get from site activation to FPI in 2 weeks.
To top it off, it takes *309 days* on average for US trials to get 75% of sites in a trial activated
what's going on?
the challenge is multifactorial - in the US, drawn out budget negotiations at every clinical site, significant overlapping local ethics and protocol reviews (esp at high profile academic research centers), and extensive site contract negotiations lead to months / quarters / years of delays.
There is no magic bullet solution - no fell swoop of regulatory or policy change will address this - wholesale revamp is needed.
simultaneously, US and western biotechs (and frontier labs) can't just wait around for systemic change to unfold. They need to play the game on the ground in order to establish a human data edge
a new clinical execution paradigm + AI-native translational layer is sorely needed
Bermes Project retweeted
Metformin: A Promising Candidate for Slowing the Progression of Cardiovascular Aging
aginganddisease.org/EN/10.14…
Bermes Project retweeted
Cancer cells survive by rewiring metabolism.
This review shows how #FattyAcidMetabolism drives #BloodCancer progression and therapy resistance, revealing targets for precision therapies. @ShandongUni1901
#OpenAccess #GenesAndDiseases: doi.org/10.1016/j.gendis.202…




















