BioByte 106: B cell depletion therapies branching out, glycoRNAs for targeted therapeutic delivery, breaking from tradition in scientific publishing and optimizing data for AI-enabled drug discovery

Mikaela Kimpton

Pablo Lubroth

Thomas Vanderstichele

, and 2 others

Mar 07, 2025

Welcome to Decoding Bio’s BioByte: each week our writing collective highlight notable news—from the latest scientific papers to the latest funding rounds—and everything in between. All in one place.

What we read

Blogs

Science transcends form: publishing beyond papers [Jessica Polka, Astera Institute, February 2025]

There are many types of documents and data that don’t fit the mold of traditional papers or preprints: “grant applications, results from projects that are discontinued, null results, individual results that don’t conform to a traditional paper format, code in GitHub, data in generalist repositories, written descriptions of the researchers’ motivations, methods, and interpretations can make these outputs more useful for humans and machines alike”.

The team at Astera have identified several important features of tools that can host non-standard scientific content. Each platform can do some but not necessarily all: hosts non-paper content, customizable display, issues DOIs, offers permanent archiving, enables public commenting, is discoverable, is indexed and has hosted repositories. For more information for each of these, please refer to the article.

The team also identified seven platforms that allow for non-traditional publication of scientific content and summarized their features in the table below:

CTRL+ALT+DEPLETE [Bauer LeSavage, Dimension Research, February 2025]

Bauer LeSavage wrote a comprehensive report on reprogramming the immune system with B cell depletion therapy, which is not a new therapeutic approach but one that has potential to be wielded for autoimmune diseases among other indications.

B cell depletion therapies (BCDTs) have already transformed oncology, particularly in B cell malignancies like Non-Hodgkin Lymphoma, where drugs like rituximab have delivered remarkable remission rates. The question now turns to whether the same approach can be harnessed to tackle autoimmune disease.

Given their success in cancer, BCDTs are being repurposed to selectively eliminate the pathogenic B cells driving autoimmune diseases, offering a potential reset for the immune system. Roche’s obinutuzumab (a CD20-targeting antibody) has shown promise for lupus nephritis, and bispecific T cell engagers (TCEs) are emerging as precision tools for B cell depletion in autoimmunity. However, unlike in cancer, where complete B cell removal may be justified, autoimmune therapies must balance efficacy with safety. Researchers are now exploring ways to fine-tune depletion—targeting only the most problematic B cell subtypes while preserving immune function. Bauer goes into immunomodulation and specific methods for reprogramming the immune system in this detailed report—a fun dive for anyone curious about creative therapeutic strategies for the immune system.

Four ways to power-up AI for drug discovery [Anthony King, Nature, February 2025]

As many who are familiar with it know, drug discovery is an extraordinarily difficult process due in large part to the high upfront temporal and monetary costs. While recent developments in AI have been praised as a potential pathway to making this process less pricey and more efficient by visually depicting 3D atomic structures and identifying targets, there are several areas that must be addressed to realize this endeavor:

Standardize reporting and methods

While the scrappiness and flexibility of scientists in collecting data using materials on hand is both admirable and extremely valuable, training AI models necessitates a consistency of how results and methods are reported and experimentation is conducted. Batch effects and inconsistencies in naming conventions can lead to difficulties in making direct comparisons, especially for an ML model. This highlights the import of large, publicly available databases such as ChEMBL and the Human Cell Atlas, which emphasize experimental rigor and standardization (though even these databases are not entirely immune to variation).

Recognize the value of negative results

Many may know this bias as the ‘file drawer problem’ or the ‘publication bias’—wherein only the successful findings are published and the so-called failed studies never see the light of day—but these negative findings are in reality just as vital to the scientific community at large, and especially for informing generative AI models. Should they be omitted, the models are often shown to propose much more idealistic solutions to biologics problems which are known, from unpublished negative results, to not be grounded in reality. This second point serves to further underline the value of open-access databases, particularly those focused on toxicology and avoid-ome characterizations.

Share industry data and expertise

The aforementioned issue with the suppressing of negative results is exacerbated by the proclivity of pharmaceutical companies to reinforce this practice, as sharing of negative results can drastically impact share prices as well as confer a competitive advantage to their opponents. To encourage big pharma to share results, certain projects have been conceived such as EU-funded Mellody, which allowed ten companies to collaboratively train predictive software while protecting sensitive data IP. Still, anonymization and lack of experimental standardization between companies breeds challenges broached in the first point, leaving another approach for pharma behemoths in advancing the AI endeavor: monetary contributions to public projects like UK Biobank.

Do more with what you have

Despite all of these challenges with existing data, many argue the importance of pushing forward in their utilization, regardless of their insufficiencies. As an immediate solution is not imminent, this is perhaps a nod back to the scrappiness of scientists in their laudably dogged pursuit of ultimate discovery. In adopting this strategy, it becomes even more requisite that AI models be trained on greater quantities of data, enabling them to have greater success in making generalizations. Insilico Medicine is one example featured as a firm finding prosperity in this manner, producing 22 preclinical candidates since the integration of their gen-AI model, Chemistry42, which scrapes existing databases to find potential target-blocking compounds, in 2019. Though some available data sources are easier to parse than others, even those deemed less-than-ideal must be used to train AI, if for no other reason than for all the animals sacrificed to obtain it, as one scientist interviewed argues. In these instances and more, smaller datasets of higher quality can be created to test these models trained on larger, hairier data sources. As they say “where there’s a will…,” and, one thing is for certain: science (and scientists) will always find a way.

Papers

RNA-binding proteins and glycoRNAs form domains on the cell surface for cell-penetrating peptide entry [Perr et al., Cell, February 2025]

Recent discoveries have shown that glycoRNAs – RNAs modified with glycan sugars – are present on the surface of cells. However, the precise roles for many of these RNAs remains unknown. In a recent study, Ryan Flynn’s lab explored whether these glycoRNAs are organised on the cell surface by cell-surface RNA-binding proteins (csRBPs)

The researchers first identified csRBPs from publicly available datasets and validated their presence on the membrane using a combination of experimental techniques. Imaging studies revealed that these csRBPs form distinct clusters on the cell surface. Further analysis using protein proximity labeling demonstrated that these clusters are enriched with glycoRNAs. A particularly intriguing discovery was that these glycoRNA-csRBP clusters exhibited highly regular patterns in terms of both size and spacing. This structured organization suggests a level of regulation in how glycoRNAs and their associated proteins interact on the cell surface. The study also revealed that cell-penetrating peptides rely on glycoRNA-csRBP clusters for cellular entry. This suggests that glycoRNAs may play a crucial role in facilitating peptide uptake, potentially due to the cationic charge of RNA.

There is increasing interest in harnessing cell-surface glycoRNAs for targeted therapeutic delivery. The identification of glycoRNA-csRBP clusters not only enhances our understanding of cellular biology but also opens up exciting possibilities for designing novel cell-specific therapeutics. This discovery could pave the way for more precise drug delivery systems, leveraging glycoRNAs as key mediators of cellular entry mechanisms. To achieve this it will be necessary to focus future studies on cell-type-specific glycoRNA-csRBP clusters and how these could be specifically targeted.

The interplay between age at menopause and synaptic integrity on Alzheimer’s disease risk in women [Alexander et al., Science Advances, March 2025]

Women constitute two-thirds of Alzheimer’s dementia diagnoses, suggesting a significant link between sex and disease incidence. However, this link is one that has not as of yet been thoroughly explored. Prior research has suggested that certain hormones—particularly estradiol—seem to exhibit a protective effect against cognitive decline. Since these hormones change dramatically during menopause, the authors sought to uncover whether the age of menopause onset—and thus the dip in estradiol and increase in gonadotropin hormones like follicle stimulating hormone (FSH)—had any effect on the correlation between synaptic health and cognitive decline, particularly as it pertains to the onset of Alzheimer’s disease.

To examine this effect, Alexander et al. analyzed postmortem brain tissue samples from over 200 women as part of the Rush Memory and Aging Project (MAP). Analysis focused largely on three synaptic biomarkers: complexin-I, complexin-II, and SNARE protein-protein interactions, while also factoring in the presence and quantity of tau protein tangles, an established indicator of AD. Findings indicate that with earlier onset menopause, the correlation between lower levels of the synaptic biomarkers and cognitive decline was strengthened. The proposed explanation for this increase was an uptick in the concentration of tau tangles in the tissue samples, as this correlation was not observed with β amyloid. This reinforces the observation that elevated rates of cognitive decline in women may be caused by a unique susceptibility to tau burdens. This effect was attenuated in women who underwent hormone therapies, indicating that the hormones involved in menopause play a key role in mediating this relationship.

While many directions for future study were suggested, these findings offer key insights into the widespread and debilitating indication that is Alzheimer’s disease. They also underline the critical need for further research that focuses on women’s health, not just the sake of female-identifying individuals, but for the benefit of all.

Notable deals

Lilly and Magnet Biomedicine have entered a multi-target collaboration, featuring an upfront investment of $40M, equity, and near-term milestones to develop molecular glues for oncology. This partnership not only advances their joint efforts but also provides Magnet with essential funding to accelerate its internal discovery efforts. Magnet’s proprietary TrueGlue platform harnesses advanced screening technologies, unique chemical libraries, and a strategic approach to target and presenter protein selection to create its molecular glue therapeutics. Andrew Pannu shared a visual (below) on the landscape of recent molecule glue deals, note that the values quoted are total biobucks, which are not equally obtainable in all cases.

Novo Nordisk has partnered with Gensaic in a deal worth up to $354 million, reinforcing its strategy to secure novel targets and delivery technologies for metabolic diseases. This collaboration originated through Novo’s BioInnovation Hub, where Gensaic received lab space and mentorship, allowing Novo to assess its technology early. Gensaic’s FORGE platform leverages protein evolution and machine learning to engineer delivery vehicles. While financial details remain undisclosed, this partnership reflects a broader trend of pharma giants embedding within startups to mitigate risk and gain a deeper understanding of emerging technologies before making significant investment commitments.
TRIMTECH has raised $31M (£25M) in seed funding to advance its targeted protein degradation pipeline for neurodegenerative diseases. This funding will support the continued development of novel therapeutics designed to address a critical unmet need for patients with limited treatment options.
AbbVie enters the obesity arena with Gubra, a Danish biotech, for $350M upfront and $1.87B in milestones for a long-acting amylin analog.Notably, the deal is focused solely on amylin (excluding any incretin assets) which may indicate that AbbVie is targeting a monotherapy approach, or that further deal-making is anticipated. Given that amylin analogs have demonstrated a better tolerability profile than incretins to date, they could be particularly well-suited for patients aiming for modest weight loss goals.