Weekly reads 3/2/25

Week 2, let’s try to create a streak. This week's selection spans a diverse range of topics, from advances in causal modeling of gene regulation to unexpected immune system players in cancer defense and cutting-edge computational tools for single-cell analysis. Currently sticking to the same format as it aligns smoothly with my note-taking.

Preprints/articles that i managed to read this week

Mapping Genetic Effects from Regulators to Traits: A Breakthrough in Causal Modeling

Ota et al. (2025). Causal modeling of gene effects from regulators to programs to traits: integration of genetic associations and Perturb-seq. bioRxiv. https://doi.org/10.1101/2025.01.22.634424

The paper in one sentence

By integrating genetic association data with Perturb-seq experiments, this study constructs causal models that trace how genetic variations impact biological programs and ultimately influence human traits.

Summary

Genetic association studies have identified thousands of gene-trait relationships, but understanding the biological mechanisms behind these associations remains a challenge. This study bridges the gap by combining loss-of-function burden tests with Perturb-seq data to map how genes regulate biological programs, which in turn influence traits. As a proof-of-concept, the researchers constructed a causal model explaining how gene regulatory networks control three blood traits—mean corpuscular hemoglobin (MCH), red cell distribution width (RDW), and immature reticulocyte fraction (IRF). The approach not only identifies key regulators but also predicts gene function and interactions, providing a framework for future genetic studies.

Personal highlights

• Causal gene mapping: The study builds a causal graph linking genetic variants to traits through their effects on regulatory programs.

• Integration of genetic and functional data: By combining GWAS and Perturb-seq, researchers can infer gene effects more accurately than with GWAS alone.

• Proof-of-concept with blood traits: The model successfully explains the regulation of blood-related traits, with insights into cell cycle and hemoglobin synthesis pathways.

Why should we care?

Many diseases, from cancer to cardiovascular disorders, have genetic underpinnings that remain poorly understood. This study introduces a powerful approach to map genetic variations to their biological effects, helping researchers pinpoint actionable targets for treatment. By improving our ability to interpret genetic associations, this work accelerates the path toward precision medicine, where treatments can be tailored based on a person’s unique genetic makeup.

Eosinophils as Unexpected Defenders Against Colorectal Cancer Metastasis

Handler et al. (2024). Eosinophils restrict CRC metastasis by inhibiting pro-tumorigenic SPP1+ macrophage differentiation. bioRxiv. https://doi.org/10.1101/2024.12.13.628333

The paper in one sentence:

Eosinophils, immune cells traditionally linked to allergies, play a surprising role in preventing colorectal cancer (CRC) metastasis by blocking the formation of pro-tumorigenic SPP1+ macrophages.

Summary:

While eosinophils are best known for their role in allergic reactions, this study uncovers their unexpected function in shaping the tumor microenvironment (TME). Using transcriptomic analysis of CRC patient samples and mouse models, the researchers found that eosinophils actively suppress the differentiation of SPP1+ tumor-associated macrophages (TAMs), a subset that promotes immune evasion and cancer metastasis. Depleting eosinophils in mice led to an increase in metastatic spread, highlighting their protective role.

Personal highlights:

1. Eosinophils suppress metastasis: Despite declining numbers in late-stage CRC, eosinophils counteract pro-tumor macrophages, reducing the spread of cancer cells.

2. SPP1+ macrophages drive metastasis: These macrophages support immune evasion and tumor progression. Eosinophils inhibit their conversion

3. Loss of eosinophils worsens outcomes: In mice, depleting eosinophils led to increased tumor spread, confirming their critical role in preventing metastasis.

Why Should a Broader Audience Care?

This study challenges the conventional view of eosinophils as mere allergy-related cells and highlights their crucial role in cancer defense. With CRC metastasis being a major cause of mortality, enhancing eosinophil activity—or mimicking their effects—could provide a new strategy for treatment.

Engineered Fat Cells as a Novel Cancer Therapy: Outcompeting Tumors for Nutrients

Nguyen et al. (2025). Implantation of engineered adipocytes suppresses tumor progression in cancer models. Nature Biotechnology. https://doi.org/10.1038/s41587-024-02551-2

The paper in one sentence:

By engineering adipocytes to consume more glucose and fatty acids, researchers have developed a novel therapy that starves tumors of essential nutrients, significantly slowing cancer progression.

Summary:

Tumors are highly efficient at hijacking the body’s metabolic resources to fuel their growth. This study introduces a new therapeutic approach—Adipose Manipulation Transplantation (AMT)—which uses engineered fat cells to outcompete cancer cells for glucose and fatty acids. The researchers modified human adipocytes to upregulate UCP1, a gene involved in energy metabolism, leading to increased glucose and fat uptake. When implanted near tumors in preclinical models, these engineered adipocytes suppressed tumor growth, reduced hypoxia, and inhibited angiogenesis. Further experiments demonstrated that AMT is effective across multiple cancer types, including breast, pancreatic, and prostate cancers, offering a promising new strategy to complement existing therapies.

Personal highlights:

1. Engineered fat cells compete with tumors: Adipocytes modified to express UCP1 consume excess glucose and fatty acids, depriving cancer cells of essential nutrients.

2. Significant tumor suppression: In mouse models of breast, pancreatic, and prostate cancer, implantation of engineered adipocytes led to over 50% reduction in tumor volume.

3. Reduces tumor hypoxia and angiogenesis: Engineered adipocytes not only slow tumor growth but also decrease oxygen deprivation and blood vessel formation, making tumors less aggressive.

4. Ease of clinical translation: Since adipose tissue is easily accessible via liposuction and can be re-implanted with minimal complications, AMT offers a practical and patient-friendly approach.

5. Customizable metabolic targeting: By modifying adipocytes to target specific metabolic pathways, this strategy could be adapted for different tumor types.

Why should we care?

Cancer treatments often focus on directly killing tumor cells, but this study introduces an entirely different approach—starving cancer by modifying the metabolic environment. The ability to repurpose fat cells as a therapeutic tool could revolutionize cancer treatment, offering a minimally invasive, highly adaptable strategy that complements existing therapies.

The role of adipocytes in cancer therapy: a practical advantage

One of the biggest challenges in developing innovative cancer treatments is ensuring their feasibility for real-world use. The AMT approach has a major advantage: adipose (fat) tissue is one of the easiest and most abundant cell types to access in the human body. Liposuction is a routine, well-established procedure that allows for the collection of a patient’s own fat cells with minimal risk. Unlike other forms of cell-based therapy, which require complex stem cell extractions or bone marrow harvesting, adipocytes can be easily obtained and engineered for therapeutic use.

Once modified, these engineered fat cells can be reintroduced into the body through straightforward transplantation techniques. Fat grafting, commonly used in cosmetic and reconstructive surgery, already provides a clinical precedent for safe and effective adipocyte implantation. Because adipose tissue integrates well into the body with low rejection risk, AMT could be seamlessly incorporated into existing medical workflows, making it more accessible for widespread use.

A potentially game-changing therapy

By leveraging a patient’s own fat cells, AMT provides a highly personalized and low-risk treatment option that could revolutionize how we approach cancer therapy. With further research and clinical trials, this method could become a mainstream strategy to slow tumor growth in a minimally invasive manner, offering new hope for patients with aggressive and treatment-resistant cancers.

scCausalVI: A Causality-Aware Generative Model for Disentangling Single-Cell Perturbation Responses

An et al. (2025). scCausalVI disentangles single-cell perturbation responses with causality-aware generative model. bioRxiv. https://doi.org/10.1101/2025.02.02.636136

The paper in one sentence:

scCausalVI is a deep learning model that integrates structural causal modeling with variational inference to disentangle intrinsic cellular states from treatment-induced effects in single-cell RNA sequencing (scRNA-seq) data.

Summary:

scRNA-seq enables researchers to analyze how individual cells respond to perturbations such as drugs, genetic modifications, or environmental changes. However, current computational methods struggle to separate true treatment effects from inherent cellular heterogeneity. scCausalVI addresses this by incorporating structural causal modeling (SCM) into a deep learning framework, explicitly modeling causal relationships between intrinsic cellular states and treatment responses. The model achieves this by learning two distinct latent spaces: one for baseline cellular heterogeneity and another for treatment-specific effects. By leveraging this causality-aware approach, scCausalVI enables in silico predictions of how cells would behave under different conditions, making it a powerful tool for drug response modeling, precision medicine, and biological discovery.

Personal highlights:

1. Causal disentanglement in single-cell data: scCausalVI uniquely separates inherent cellular variation from treatment-induced changes using deep structural causal modeling, unlike conventional machine learning models that rely on correlation-based methods.

2. Cross-condition in silico prediction: The model can predict how a given cell’s gene expression would change under hypothetical treatment conditions, facilitating virtual perturbation experiments.

3. Attention-based cell-specific scaling: scCausalVI employs Squeeze-and-Excitation Networks (SENet) to adaptively scale treatment effects for each cell, capturing the heterogeneity in response across different cell types.

4. Batch effect correction with causal inference: Unlike traditional batch correction techniques, scCausalVI explicitly models batch effects separately from treatment responses, preserving biological signals while eliminating technical noise.

Why should we care?

As scRNA-seq technology becomes increasingly central to biomedical research, the ability to accurately model and predict cellular responses to drugs, diseases, and environmental stimuli is critical. scCausalVI provides a breakthrough by applying causal inference principles to single-cell data, moving beyond correlation-based approaches that often misattribute effects. By enabling computationally driven in silico experiments, scCausalVI reduces the need for expensive and time-consuming lab work.

Figure 1 from An et al.: overview of scCausalVI

bpAI-TAC: Advancing Sequence-to-Function Models by Increasing Resolution in Chromatin Accessibility Predictions

Chandra et al. (2025). Refining the cis-regulatory grammar learned by sequence-to-activity models by increasing model resolution. bioRxiv. https://doi.org/10.1101/2025.01.24.634804

The Paper in one sentence:

bpAI-TAC, a multi-task neural network, enhances chromatin accessibility prediction by modeling ATAC-seq data at base-pair resolution, significantly improving our understanding of transcription factor (TF) binding and gene regulation.

Summary:

Regulatory DNA sequences control when and how genes are expressed, shaping cell identity and function. ATAC-seq technology captures chromatin accessibility across the genome, revealing active regulatory regions. Traditional sequence-to-function (S2F) models, such as AI-TAC, predict chromatin accessibility based on DNA sequences but operate at lower resolution, missing finer regulatory details. The newly developed bpAI-TAC improves upon this by integrating base-pair resolution ATAC-seq profiles, allowing for more precise identification of TF binding sites. By using a multi-task learning approach, bpAI-TAC models chromatin accessibility across 90 immune cell types, outperforming previous models in predicting differential gene regulation. This advance refines our understanding of cis-regulatory grammar, improving the ability to interpret non-coding genetic variations and their impact on disease.

Personal highlights:

1. Base-pair resolution for chromatin accessibility: Unlike previous models that predict accessibility over broader genomic regions, bpAI-TAC leverages ATAC-seq cut-site profiles at base-pair precision, enhancing its ability to detect transcription factor (TF) footprints.

2. Multi-task learning improves predictions: Instead of training separate models for different cell types, bpAI-TAC learns regulatory patterns across 90 immune cell types simultaneously, leading to better generalization and efficiency.

3. Improved identification of regulatory motifs: The model refines motif predictions by accounting for TF binding footprints, capturing sequence-dependent variability in regulatory grammar more accurately than AI-TAC.

4. Bias-correction for ATAC-seq profiles: To distinguish true TF signals from sequencing bias, bpAI-TAC integrates a bias model trained on protein-free DNA, allowing it to filter out spurious ATAC-seq artifacts.

5. Better interpretation of variant effects: By improving the resolution of chromatin accessibility modeling, bpAI-TAC enables more precise predictions of how genetic mutations affect gene regulation, aiding functional genomics and disease research.

Why should we care?

Deciphering how non-coding DNA variations impact gene expression is critical for understanding diseases such as cancer, immune disorders, and neurodegenerative conditions. bpAI-TAC’s ability to detect regulatory sequences with improved accuracy makes it a powerful tool for precision medicine, drug discovery, and synthetic biology.

How TLR7 Shapes Sex Differences in Aging and Alzheimer’s-Related Demyelination

Lopez-Lee et al. (2024). Tlr7 drives sex differences in age- and Alzheimer’s disease–related demyelination. Science, 386, eadk7844. https://doi.org/10.1126/science.adk7844

The paper in one sentence:

This study identifies Toll-like receptor 7 (TLR7) as a central regulator of sex-specific immune responses in the aging brain, linking it to differences in microglial activation and demyelination in Alzheimer’s disease (AD).

Summary:

Demyelination, or the loss of myelin—the protective sheath around nerve fibers—is a hallmark of neurodegenerative diseases such as Alzheimer’s disease (AD). While sex differences in AD risk are well established, little has been known about whether these differences extend to demyelination. Using a mouse model of demyelination and single-nucleus RNA sequencing, the researchers found that aged female mice exhibited more severe myelin loss and motor impairments, while male mice with tau pathology experienced heightened microglial activation and demyelination. The X-linked immune gene TLR7 emerged as a key driver of these sex-specific effects. In males, TLR7 amplified interferon (IFN) signaling in microglia, exacerbating myelin loss. When TLR7 was deleted, these sex differences were diminished, and demyelination was reduced. Importantly, pharmacological inhibition of TLR7 protected male mice from tau-induced motor deficits and myelin loss, highlighting its role as a potential therapeutic target.

Personal highlights:

1. Females experience more severe demyelination with age: Aged female mice showed greater myelin loss and motor deficits than their male counterparts, an effect linked to interactions between sex chromosomes and gonadal hormones.

2. Males exhibit stronger microglial activation in Alzheimer’s models: In mice with tau pathology, male microglia had a heightened interferon (IFN) response, leading to increased demyelination.

3. TLR7 is a key regulator of sex-specific demyelination: This X-linked immune gene drives male-biased IFN signaling, promoting microglial activation and myelin loss in Alzheimer’s models.

4. Deleting TLR7 protects against demyelination: Removing TLR7 reduced the male-specific IFN response and prevented excessive myelin loss in both aging and tauopathy models.

5. Sex-specific therapeutic potential: Inhibiting TLR7 pharmacologically successfully mitigated tau-induced motor impairment and demyelination in male mice, suggesting a possible intervention for Alzheimer’s disease.

Why should we care?

This study reshapes our understanding of sex differences in neurodegenerative diseases by showing that immune signaling pathways differ between males and females, influencing disease progression in distinct ways. The discovery that TLR7 drives male-specific microglial activation and myelin loss underscores the importance of considering biological sex in Alzheimer’s research and treatment development.

Other papers that peeked my interest and were added to the purgatory of my “to read” pile

• Quantifying cell divisions along evolutionary lineages in cancer: https://www.nature.com/articles/s41588-025-02078-5

• Learning single-cell spatial context through integrated spatial multiomics with CORAL: https://www.biorxiv.org/content/10.1101/2025.02.01.636038v1

• Nucleotide dependency analysis of DNA language models reveals genomic functional elements: https://www.biorxiv.org/content/10.1101/2024.07.27.605418v1

Thanks for reading.

Cheers,

Seb.