The arrival of AMP-Diffusion marks a significant advancement from previous methodologies that primarily relied on sifting through vast datasets to isolate promising antibiotic candidates. Prior breakthroughs at Penn had already demonstrated that AI could effectively sort through massive amounts of biological data and identify antibiotic prospects. However, the current study takes a revolutionary leap forward by demonstrating that AI can also concoct antibiotic candidates from scratch. With the increasing urgency of developing new antibiotics, especially in the wake of alarming rates of antibiotic resistance, the promise of AMP-Diffusion could not be more timely.

Pranam Chatterjee, Assistant Professor in Bioengineering and Computer and Information Science at Penn, along with César de la Fuente, Presidential Associate Professor in Bioengineering and Chemical and Biomolecular Engineering, spearheaded this innovative project. Chatterjee emphasizes the ability to leverage AI not merely as a tool for analysis but as a creator capable of designing new antibiotic molecules. The collaborative efforts of both labs are foundational, blending their unique expertise to push the boundaries of what’s achievable in antibiotic discovery.

The methodology of the AMP-Diffusion model mirrors techniques used in popular AI platforms like DALL·E and Stable Diffusion, which have gained prominence for their ability to generate images based on textual descriptions. Instead of “denoising” pixels as in these more visual AI applications, AMP-Diffusion undergoes a similar process for sequences of amino acids—gradually refining random noise into biologically relevant sequences. In this intricate process, the model begins with a chaotic array of possibilities and hones in on effective peptide structures.

While traditional generative models typically rely on predicting the next element in a sequence, AMP-Diffusion takes advantage of pre-existing protein language models, specifically ESM-2 developed by Meta. This foundational model had been trained on a staggering number of natural protein sequences, providing AMP-Diffusion with a comprehensive internal framework of how proteins are structured. By starting with this robust “mental map,” AMP-Diffusion can expedite the generation of candidate AMPs while ensuring that these candidates adhere to the biological realities governing effective peptides.

In total, AMP-Diffusion produced approximately 50,000 candidate sequences, an incredible volume far surpassing what conventional testing methods could evaluate. Recognizing the impracticality of testing every candidate, the researchers employed an AI tool previously developed by de la Fuente’s lab, known as APEX 1.1, to filter candidates based on various parameters. The screening process not only sought sequences with strong antimicrobial properties but also filtered out redundancies by eliminating peptides too similar to existing AMPs. This level of filtration ensures a diverse array of candidate types, thus broadening the scope of potential discoveries.

From the pool of candidates, the teams synthesized 46 of the most promising AMPs for comprehensive testing. The subsequent evaluations in human cells and animal models yielded remarkable results: two of these AMP candidates demonstrated efficacy comparable to that of FDA-approved antibiotics such as levofloxacin and polymyxin B. Astonishingly, these AI-generated molecules managed to treat skin infections in mice without causing any adverse effects, validating the effectiveness of machine learning in drug discovery.

The implications of these findings extend beyond antibiotic treatment; they represent a paradigm shift in how researchers can expedite the timeline of antibiotic discovery, which frequently spans many years. Chatterjee outlines this potential transformation, expressing hope that future iterations of AMP-Diffusion could allow for the crafting of drug candidates with even more specific therapeutic goals in mind. This could mean producing antibiotics tailored for particularly stubborn bacterial strains or even for different types of infections.

Looking ahead, the researchers plan to refine the capabilities of AMP-Diffusion, enhancing its ability to target specific properties in future designs to elevate the effectiveness of generated antibiotics. Each refinement brings scientists one step closer to realizing their ambition of reducing the antibiotic discovery timeline from years to mere days. Such efficiency could usher in a new era of drug development, one where generating effective antibiotics becomes a streamlined and rapidly attainable goal.

This research is not merely a demonstration of technology; it represents a broader vision of battling antibiotic resistance through innovation. As the urgency of developing new antibacterial treatments increases, AMP-Diffusion positions itself as a beacon of hope for medical science, providing the tools necessary to forge new paths in the fight against drug-resistant bacteria.

The study not only underscores the synergy between biology and artificial intelligence but also serves as a springboard for future investigations. By tapping into generative AI’s potential, researchers can explore uncharted territories in drug discovery and rekindle the fight against some of humanity’s most pressing health challenges. Ultimately, the integration of AI in the process illuminates a bright future, one where antibiotics can be designed, tested, and deployed rapidly, thereby offering a significant countermeasure to the perilous rise of antibiotic-resistant infections globally.

Subject of Research: Animals
Article Title: Generative latent diffusion language modeling yields anti-infective synthetic peptides
News Publication Date: 2-Sep-2025
Web References: DOI link
References:
Image Credits: Credit: Sylvia Zhang

Keywords

Artificial Intelligence, Antibiotic Resistance, Antimicrobial Peptides, Drug Discovery, Generative AI, Bioengineering, Peptide Design, Innovation in Medicine, Computational Biology, Synthetic Biology

Oligosaccharides are fascinating biomolecules made up of three to ten linked sugar units known as monosaccharides. Unlike large structural polysaccharides like cellulose or energy-storing starches, oligosaccharides frequently appear on cell surfaces where they act as crucial signaling and recognition agents. Their structural diversity—including variations in sugar type, bonding sites, and spatial configuration of glycosidic linkages—makes them incredibly versatile but also notoriously difficult to manufacture with exactitude. The sheer number of possible five-unit oligosaccharides exceeds 100 million, complicating chemical synthesis and isolation from natural sources.

The heart of the challenge lies in controlling the stereochemistry of the linkages—that is, the “handedness” or spatial orientation of the bonds between sugars. Traditional chemical reactions often yield mixtures of right- and left-handed connections because the newly forming bond can occur in more than one spatial arrangement. This ambiguity contrasts starkly with proteins and nucleic acids, where bonds form in a uniform manner without producing stereoisomeric mixtures. An oligosaccharide’s complexity increases exponentially as it grows, potentially producing more than 2,000 stereochemical permutations for a 10-sugar chain, rendering nondirected synthesis inefficient and impractical.

Isolation from natural sources offers no viable solution either, as biological degradation of oligosaccharides results in a complex amalgam of similar molecules that defy straightforward purification. While enzymatic synthesis can provide specificity, enzyme development is both costly and time-consuming, especially for early-phase research requiring rapid access to diverse oligosaccharides. Consequently, the development of a broadly applicable, chemically robust, and automatable synthetic method has been a pressing goal in carbohydrate chemistry for over a century.

Addressing this long-standing obstacle, the UCSB and MPIKG team devised a strategy based on bimolecular nucleophilic substitution (SN2) chemistry. Unlike less selective processes, SN2 reactions proceed via a single concerted step where an incoming sugar molecule displaces a leaving group simultaneously, resulting in inversion of stereochemistry at the reaction center. This feature inherently enables precise control over the orientation of the glycosidic bond, ensuring the exclusive formation of either right- or left-handed linkages depending on reaction design. Achieving this stereoselectivity in the complex milieu of oligosaccharide synthesis, however, is nontrivial and requires delicate regulation of reaction kinetics.

The researchers enhanced the SN2 reaction’s fidelity by introducing a directing molecule tethered to the departing group. This molecular “hand” guides the incoming sugar to attack in the correct spatial orientation just before the leaving group departs. The approach works effectively across multiple sugar types and linkage positions, and can be performed under mild, near-neutral conditions compatible with solid-phase synthesis. Anchoring oligosaccharides onto a polymer support allows for iterative addition of sugar units, with the support-bound molecule retained while unreacted reagents and undesired byproducts are washed away at each step—streamlining the purification process and conserving reagents.

Solid-phase synthesis has revolutionized peptide and oligonucleotide production, earning its original developer a Nobel Prize in 1984. Peter Seeberger, a co-author and trailblazer in carbohydrate solid-phase synthesis since 2001, has spent over two decades adapting and refining this technique for sugars. The current breakthrough in stereocontrolled glycosylation now makes fully automated assembly of oligosaccharides not only feasible but accessible to scientists without specialized synthetic chemistry expertise. This democratization of carbohydrate synthesis is expected to accelerate research dramatically.

The implications for biomedical science are profound. Oligosaccharides play critical roles in immune signaling, host-pathogen interactions, developmental biology, and disease biomarkers. Until now, obtaining pure, structurally defined oligosaccharides required months of manual labor and substantial financial investment. The new automated platform can produce these molecules rapidly and with precise configuration, facilitating explorations in vaccine development, autoimmune disease diagnostics, and antimicrobial agents targeting hospital-acquired infections.

Despite the significant progress, some challenges remain. The synthesis of certain complex bonds, such as the elusive “beta mannosidic” linkage, has so far resisted full stereocontrol. The UCSB and MPIKG teams are actively pursuing solutions to these remaining obstacles, expanding the methodology’s scope to include rare and bacterial sugars that hold promise for novel therapeutic applications. Continued advances in carbohydrate synthesis techniques will underpin future breakthroughs in drug development and disease prevention.

This discovery underscores the critical importance of fundamental methodological advances in chemistry for driving applied biomedical innovation. By unlocking the ability to reliably build complex carbohydrates with tailored properties, this work opens new frontiers in glycobiology research and pharmaceutical design. The vision of a “one-size-fits-all” synthetic method for carbohydrates appears closer than ever, thanks to the harmonious integration of reaction engineering, mechanistic insight, and automation championed by these pioneering scientists.

In summary, this research heralds a new era where the structural complexity of carbohydrates is no longer a barrier but an opportunity. With robust stereochemical control and automated synthesis, oligosaccharides can now be produced on demand, fueling discoveries that were previously out of reach. The ripple effect of this achievement will likely permeate biomedical research, leading to more precise diagnostic tools, next-generation vaccines, and innovative therapeutic agents targeting some of today’s most challenging medical problems.

Subject of Research: Automated stereoselective synthesis of oligosaccharides
Article Title: New Method Enables Precise Automated Assembly of Oligosaccharides for Biomedical Innovations
Web References: https://www.nature.com/articles/s44160-025-00846-z
Image Credits: Matt Perko

Keywords

Applied sciences and engineering, Oligomerization, Chemical engineering, Biochemical engineering, Stereochemistry

In a study published in the prestigious journal Nature Physics, MIT researchers detail how the fluid nestled between cells significantly influences a tissue’s capacity for deformation and relaxation. Traditionally, scientists have ascribed tissue compliance primarily to intracellular elements—such as the cytoskeleton and organelles—overlooking the contributions arising from extracellular fluids. This new research challenges that dogma by demonstrating that the ease with which intercellular fluid moves profoundly governs how tissues bear mechanical stress. Specifically, when the spaces between cells allow for free fluid flow, tissues exhibit greater compliance and faster relaxation rates following deformation. Conversely, tightly packed cells that limit this fluid movement yield stiffer, more resistant tissues.

This insight carries ramifications across biology and medicine, emphasizing the importance of considering the extracellular fluid environment when analyzing tissue behavior. For instance, adaptations of muscular tissue during exercise or recovery may hinge not merely on the cellular architectures but on how effectively intercellular fluid redistributes under strain. Similarly, pathologies such as cancer, diabetes, and neuromuscular disorders involve tissue deformation and remodeling where this interstitial fluid dynamics might play previously unappreciated roles. Adjusting this fluid flow could influence disease progression or tissue resilience, highlighting a new target for therapeutic intervention.

The researchers also propose groundbreaking applications in the realm of biomedical engineering, particularly in designing artificial tissues and organs. By optimizing the permeability and flow dynamics of interstitial fluids within engineered tissues, scientists could enhance their functional properties, making artificial implants more durable and physiologically accurate. Additionally, controlled facilitation of fluid movement could become a strategy for targeted drug or nutrient delivery, exploiting the natural pathways between cells to penetrate dense tissue matrices, a concept that may revolutionize treatments for tumors or degenerative conditions.

Ming Guo, associate professor of mechanical engineering at MIT and a co-author of the study, stresses the historical neglect of intercellular fluid flow in biomechanical research. “While it’s recognized that a substantial amount of fluid exists in tissue interstices, its role in tissue deformation has been largely ignored,” Guo remarks. “Our observations now reveal that this fluid movement dominates tissue mechanical behavior during deformation, urging scientists to integrate this understanding into both disease study and tissue engineering efforts.”

Their investigative journey originates from a previous 2020 study focusing on tumor mechanics, where the researchers observed that squeezing tumors enhanced fluid flow from their interior to periphery via intercellular channels. This phenomenon appeared to facilitate tumor invasion by transporting fluids and potentially signaling molecules, fostering more aggressive cancer spread. Building on these findings, the team sought to determine whether similar fluid dynamics play a role in noncancerous tissues, such as those found in healthy organs.

The experimental approach combined biological and mechanical ingenuity. The researchers cultured microtissue clusters composed of tens of thousands of pancreatic tissue-derived cells, each cluster measuring under a quarter-millimeter in diameter. To assess mechanical responses, they engineered a novel testing apparatus combining a high-precision microbalance capable of detecting infinitesimal changes in weight with a step motor designed for nanometer-scale compression control. This setup allowed them to compress tissue clusters meticulously while monitoring their relaxation behavior over time.

As the tissues were gently pressed, videos captured the transformation from a spherical to a flattened “pancake” shape, accompanied by dynamic shifts in weight indicating fluid displacement. The investigative hypothesis proposed that if intercellular fluid movement governs the relaxation response, larger tissue clusters would necessitate longer times for fluid to percolate through and for the tissue to rebound. Conversely, if the mechanical response depended purely on tissue architecture, relaxation times would display size-independence.

Experiments yielded consistent and striking results: relaxation times scaled with tissue size, confirming that fluid flow between cells significantly dictates mechanical behavior. Larger clusters exhibited slower relaxation due to longer pathways for fluid drainage. This pattern held across multiple tissue types, underscoring a universal principle in tissue biomechanics. Fan Liu, the study’s lead author and MIT postdoctoral researcher, emphasizes the broader impact, noting that these findings necessitate a paradigm shift in understanding tissue mechanics and could influence future biomedical device design and therapeutic strategies.

Looking ahead, the research team envisions extending this work to the brain, a complex organ where fluid dynamics critically influence function and pathology. They speculate that facilitating interstitial fluid flow may aid in the clearance of metabolic waste products implicated in neurodegenerative diseases like Alzheimer’s. Thus, modulating these microfluidic environments might become a powerful tool for maintaining or restoring brain health.

Guo also highlights the concept of therapeutic mechanical stimulation: “As our study shows, applying targeted pressure to tissues drives fluid flow. We might one day develop non-invasive techniques to ‘massage’ tissues, enhancing nutrient delivery and waste removal within cellular environments.” Such innovations could transform rehabilitation protocols, sports medicine, and chronic disease management.

This research not only advances fundamental biomechanical knowledge but also exemplifies the interdisciplinary synergy of mechanical engineering, cell biology, and materials science. Supported by the MIT Department of Mechanical Engineering, the study underscores the importance of precise measurement tools and conceptual frameworks to unravel complex biological phenomena. As the team continues their investigations, the prospects for integrating intercellular fluid dynamics into clinical practice and bioengineering promise to catalyze transformative health innovations.

In summary, this seminal study unveils the pivotal influence of intercellular fluid flow on tissue poroelasticity, altering the scientific narrative around tissue mechanics. By marrying meticulous experimentation with insightful analysis, MIT engineers have illuminated a previously hidden determinant of tissue behavior under mechanical stress. The implications ripple through biomedical research, clinical strategies, and engineered tissue development, heralding a new era in understanding how the fluid-filled interstices between cells shape life’s fundamental processes.

Subject of Research: Mechanics of tissue deformation and the role of intercellular fluid flow
Article Title: “Intracellular flow dominates the poroelasticity of multicellular tissues”
References: Nature Physics
Image Credits: Courtesy of Ming Guo, Fan Liu, et al
Keywords: Biomechanics, Biophysics, Mechanics, Cell biology, Cellular physiology, Cell behavior, Cell responses, Health and medicine, Tissue engineering, Biomedical engineering, Mechanical engineering